[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:figure:zh":3,"public-menus:all":45,"post:what-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split:zh":531,"related:post:what-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split:zh:1":1592},{"statusCode":4,"data":5,"message":44},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":14,"robotsIndex":15,"socialLinks":14,"reservedSlugs":14,"seoPolicy":16},"figure","zh","en","https:\u002F\u002Ffigure.rocks","info@stajic.de","Figure Rocks","\u002Ffavicon-32x32.png","",null,true,{"branding":17,"relatedContent":18,"crossDomainLinks":19},{"logoUrl":12},{"enabled":15},[20,23,26,29,32,35,38,41],{"url":21,"label":22,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":24,"label":25,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002Fstajic.de","stajic.de",{"url":27,"label":28,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002Fbazify.com","bazify.com",{"url":30,"label":31,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002Fbazify.de","bazify.de",{"url":33,"label":34,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002F2mesta.com","2mesta.com",{"url":36,"label":37,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002F2mesta.de","2mesta.de",{"url":39,"label":40,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002Fbazify.at","bazify.at",{"url":42,"label":43,"isActive":15,"showInFooter":15,"includeInSameAs":15},"https:\u002F\u002Fweb-hoch3.de","web-hoch3.de","Portal settings resolved",[46],{"id":47,"name":48,"location":49,"isActive":15,"isDefault":15,"items":50},2,"Main Menu","sidebar",[51,65,189,282,358,424],{"id":52,"title":53,"url":61,"target":62,"icon":63,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":64},"1",{"de":54,"en":55,"es":56,"fr":57,"it":55,"ru":58,"sr":59,"zh":60},"Startseite","Home","Inicio","Accueil","Главная","Početna","首页","\u002F","_self","i-lucide-home",[],{"id":66,"title":67,"url":76,"target":62,"icon":77,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":78},"3",{"de":68,"en":69,"es":70,"fr":71,"it":72,"ru":73,"sr":74,"zh":75},"Spiele","Games","Juegos","Jeux","Giochi","Игры","Igre","游戏","\u002Fgames","i-lucide-gamepad-2",[79,92,107,121,135,148,162,176],{"id":80,"title":81,"url":76,"target":62,"icon":90,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":91},"3-1",{"de":82,"en":83,"es":84,"fr":85,"it":86,"ru":87,"sr":88,"zh":89},"Spiele-Hub","Games Hub","Centro de juegos","Hub de jeux","Hub Giochi","Игровой центр","Centar za igre","游戏中心","i-lucide-layout-grid",[],{"id":93,"title":94,"url":103,"target":62,"icon":104,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":106},"1-1",{"de":95,"en":96,"es":97,"fr":98,"it":99,"ru":100,"sr":101,"zh":102},"Neu & Angesagt","New & Trending","Novedades & Tendencias","Nouveautés & Tendances","Novità & Tendenze","Новинки & тренды","Novo & Popularno","新品 & 热门","\u002Fnews\u002Fgaming-news","i-lucide-sparkles","custom",[],{"id":108,"title":109,"url":118,"target":62,"icon":119,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":120},"1-2",{"de":110,"en":111,"es":112,"fr":113,"it":114,"ru":115,"sr":116,"zh":117},"Angebote","Deals","Ofertas","Offres","Offerte","Акции","Ponude","优惠","\u002Fdeals","i-lucide-badge-percent",[],{"id":122,"title":123,"url":132,"target":62,"icon":133,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":134},"1-3",{"de":124,"en":125,"es":126,"fr":127,"it":128,"ru":129,"sr":130,"zh":131},"Release-Kalender","Release Calendar","Calendario de lanzamientos","Calendrier des sorties","Calendario delle uscite","Календарь релизов","Kalendar izdanja","发布日历","\u002Freleases","i-lucide-calendar-days",[],{"id":136,"title":137,"url":145,"target":62,"icon":146,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":147},"3-2",{"de":138,"en":139,"es":140,"fr":141,"it":142,"sr":143,"zh":144},"amiibo-kompatible Spiele","amiibo-Compatible Games","Juegos compatibles con Amiibo","Jeux compatibles Amiibo","Giochi compatibili con amiibo","Amiibo kompatibilne igre","Amiibo 兼容游戏","\u002Freference\u002Fcompatibility","i-lucide-check-circle-2",[],{"id":149,"title":150,"url":159,"target":62,"icon":160,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":161},"3-3",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Wie amiibo funktionieren","How amiibo work","Cómo funcionan los Amiibo","Comment fonctionnent les Amiibo","Come funzionano gli Amiibo","Как работают Amiibo","Kako Amiibo rade","Amiibo 如何运作","\u002Fgames\u002Fhow-amiibo-unlocks-work","i-lucide-puzzle",[],{"id":163,"title":164,"url":173,"target":62,"icon":174,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":175},"3-4",{"de":165,"en":166,"es":167,"fr":168,"it":169,"ru":170,"sr":171,"zh":172},"Beste Amiibo pro Spiel","Best amiibo per Game","Mejores Amiibo por juego","Meilleurs Amiibo par jeu","Migliori Amiibo per gioco","Лучшие Amiibo по играм","Najbolji Amiibo po igri","各游戏最佳 Amiibo","\u002Fgames\u002Fbest-amiibo-per-game","i-lucide-award",[],{"id":177,"title":178,"url":186,"target":62,"icon":187,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":188},"3-5",{"de":179,"en":180,"es":181,"fr":182,"it":183,"ru":184,"sr":185},"Freischaltbares & Belohnungen","Unlockables & Rewards","Desbloqueables & Recompensas","Déblocables & Récompenses","Sbloccabili & Ricompense","Разблокировки & награды","Otključavanja & nagrade","\u002Fgames\u002Funlocks-and-benefits","i-lucide-gift",[],{"id":190,"title":191,"url":199,"target":62,"icon":200,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":201},"4",{"de":192,"en":193,"es":194,"it":195,"ru":196,"sr":197,"zh":198},"Sammeln","Collecting","Coleccionismo","Collezionismo","Коллекционирование","Kolekcionarstvo","收藏","\u002Fcollecting","i-lucide-gem",[202,213,227,240,254,268],{"id":203,"title":204,"url":211,"target":62,"icon":90,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":212},"4-1",{"de":205,"en":206,"fr":207,"it":208,"ru":209,"zh":210},"Sammel-Hub","Collecting Hub","Hub de collecte","Hub di raccolta","Центр сбора","收藏中心","\u002Famiibo\u002Fcollecting",[],{"id":214,"title":215,"url":224,"target":62,"icon":225,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":226},"4-2",{"de":216,"en":217,"es":218,"fr":219,"it":220,"ru":221,"sr":222,"zh":223},"Kaufberatung","Buying Guide","Guía de compra","Guide d'achat","Guida all'acquisto","Гид покупателя","Vodič za kupovinu","购买指南","\u002Famiibo-library\u002Famiibo-buying-smart","i-lucide-shopping-bag",[],{"id":228,"title":229,"url":237,"target":62,"icon":238,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":239},"4-3",{"de":230,"en":230,"es":231,"fr":232,"it":233,"ru":234,"sr":235,"zh":236},"Fake vs Real","Falso vs. Real","Faux vs Vrai","Falso vs Vero","Фейк vs Оригинал","Lažno vs Pravo","真假对比","\u002Famiibo-library\u002Famiibo-buying-smart\u002Favoid-fake-listings","i-lucide-scan",[],{"id":241,"title":242,"url":251,"target":62,"icon":252,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":253},"4-4",{"de":243,"en":244,"es":245,"fr":246,"it":247,"ru":248,"sr":249,"zh":250},"Aufbewahrung & Präsentation","Storage & Display","Almacenamiento & exhibición","Rangement & Présentation","Contenitori & Esposizione","Хранение & витрины","Odlaganje & izlaganje","存储 & 展示","\u002Famiibo-library\u002Famiibo-care-and-storage","i-lucide-box",[],{"id":255,"title":256,"url":265,"target":62,"icon":266,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":267},"4-5",{"de":257,"en":258,"es":259,"fr":260,"it":261,"ru":262,"sr":263,"zh":264},"Preisübersicht","Price Guide","Guía de precios","Guide des prix","Guida ai prezzi","Гид по ценам","Cenovnik","价格指南","\u002Fcollections\u002Fprice-guide","i-lucide-bar-chart-3",[],{"id":269,"title":270,"url":279,"target":62,"icon":280,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":281},"4-6",{"de":271,"en":272,"es":273,"fr":274,"it":275,"ru":276,"sr":277,"zh":278},"Raritäten","Rare Finds","Hallazgos únicos","Trouvailles rares","Pezzi rari","Редкие находки","Retki nalazi","稀有发现","\u002Fcollections\u002Frare-and-notable","i-lucide-trophy",[],{"id":283,"title":284,"url":292,"target":62,"icon":293,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":294},"5",{"de":285,"en":286,"es":287,"fr":286,"it":288,"ru":289,"sr":290,"zh":291},"Technik","Tech","Tecnología","Tecnologia","Технологии","Tehnologija","科技","\u002Ftech","i-lucide-cpu",[295,306,318,331,344],{"id":296,"title":297,"url":292,"target":62,"icon":90,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":305},"5-1",{"de":298,"en":298,"es":299,"fr":300,"it":301,"ru":302,"sr":303,"zh":304},"Tech Hub","Centro tecnológico","Hub Tech","Hub Tecnologico","Технохаб","Tehnološki centar","科技中心",[],{"id":307,"title":308,"url":315,"target":62,"icon":316,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":317},"5-2",{"de":309,"en":310,"fr":311,"it":312,"ru":313,"zh":314},"Audio- & Mikrofonqualität","Audio & Mic Quality","Qualité audio & micro","Qualità audio & mic","Качество звука & микрофона","音频 & 麦克风质量","\u002Fgear\u002Faudio","i-lucide-mic",[],{"id":319,"title":320,"url":328,"target":62,"icon":329,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":330},"5-3",{"de":321,"en":322,"es":323,"fr":324,"it":325,"sr":326,"zh":327},"Controller & Zubehör","Controllers & Accessories","Mandos & accesorios","Manettes & accessoires","Controller & Accessori","Kontroleri & dodatna oprema","控制器 & 配件","\u002Fgear\u002Fcontrols","i-lucide-joystick",[],{"id":332,"title":333,"url":341,"target":62,"icon":342,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":343},"5-4",{"de":334,"en":335,"es":336,"fr":337,"it":338,"ru":339,"sr":340},"Displays & Aufnahme","Displays & Capture","Pantallas & Captura","Écrans & Capture","Display & Acquisizione","Дисплеи & Захват","Ekrani & snimanje","\u002Fgear\u002Fdisplays","i-lucide-monitor",[],{"id":345,"title":346,"url":355,"target":62,"icon":356,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":357},"5-5",{"de":347,"en":348,"es":349,"fr":350,"it":351,"ru":352,"sr":353,"zh":354},"Netzwerkstabilität","Network Stability","Estabilidad de la red","Stabilité du réseau","Stabilità della rete","Стабильность сети","Stabilnost mreže","网络稳定性","\u002Fplaybooks\u002Fnetwork-stability","i-lucide-wifi",[],{"id":359,"title":360,"url":368,"target":62,"icon":369,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":370},"6",{"de":361,"en":361,"es":362,"fr":363,"it":364,"ru":365,"sr":366,"zh":367},"Shop","Tienda","Boutique","Negozio","Магазин","Prodavnica","商店","\u002Fshop","i-lucide-shopping-cart",[371,383,391,399,412],{"id":372,"title":373,"url":368,"target":62,"icon":381,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":382},"6-1",{"de":374,"en":375,"es":376,"fr":377,"ru":378,"sr":379,"zh":380},"Shop-Hub","Shop Hub","Centro de compras","Espace Boutique","Центр магазина","Centar za kupovinu","购物中心","i-lucide-store",[],{"id":384,"title":385,"url":388,"target":62,"icon":389,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":390},"6-2",{"de":386,"en":386,"es":387,"fr":387,"it":387,"sr":387,"zh":387},"amiibo","Amiibo","\u002Famiibo-shop","i-lucide-scan-line",[],{"id":392,"title":393,"url":396,"target":62,"icon":397,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":398},"6-3",{"de":394,"en":394,"es":394,"fr":394,"it":394,"ru":394,"sr":394,"zh":395},"LEGO","乐高","\u002Flego-shop","i-lucide-blocks",[],{"id":400,"title":401,"url":409,"target":62,"icon":410,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":411},"6-4",{"de":402,"en":403,"es":404,"fr":405,"it":406,"sr":407,"zh":408},"Figuren & Sammlerstücke","Figures & Collectibles","Figuras & Coleccionables","Figurines & Objets de collection","Figure & Collezionismo","Figure & kolekcionarstvo","手办 & 收藏品","\u002Fshop\u002Ffigures","i-lucide-package",[],{"id":413,"title":414,"url":421,"target":62,"icon":422,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":423},"6-5",{"de":415,"en":415,"fr":416,"it":417,"ru":418,"sr":419,"zh":420},"Gaming Gear","Équipement gaming","Accessori gaming","Игровое снаряжение","Gejming oprema","游戏装备","\u002Fgaming-gear-shop","i-lucide-headphones",[],{"id":425,"title":426,"url":427,"target":62,"icon":389,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":428},"2",{"de":386,"en":386,"es":386,"fr":386,"it":386,"ru":386,"sr":386,"zh":386},"\u002Famiibo",[429,440,447,461,475,489,503,517],{"id":430,"title":431,"url":427,"target":62,"icon":90,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":439},"2-1",{"de":432,"en":433,"es":434,"fr":435,"it":435,"ru":436,"sr":437,"zh":438},"amiibo-Hub","amiibo Hub","Centro amiibo","Hub amiibo","Хаб amiibo","amiibo centar","amiibo 中心",[],{"id":441,"title":442,"url":444,"target":62,"icon":445,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":446},"1772724791061",{"de":443,"en":443},"Franchise","\u002Famiibo\u002Ffranchise","i-lucide-star",[],{"id":448,"title":449,"url":458,"target":62,"icon":459,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":460},"2-3",{"de":450,"en":451,"es":452,"fr":453,"it":454,"ru":455,"sr":456,"zh":457},"Region & Verpackung (EU\u002FUS\u002FJP)","Region & Packaging (EU\u002FUS\u002FJP)","Región y embalaje (UE\u002FEE. UU.\u002FJP)","Région & Emballage (UE\u002FUS\u002FJP)","Regione & Confezione (EU\u002FUS\u002FJP)","Регион & упаковка (EU\u002FUS\u002FJP)","Region & pakovanje (EU\u002FUS\u002FJP)","地区 & 包装 (EU\u002FUS\u002FJP)","\u002Famiibo-library\u002Famiibo-editions-and-regions","i-lucide-globe",[],{"id":462,"title":463,"url":472,"target":62,"icon":473,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":474},"2-4",{"de":464,"en":465,"es":466,"fr":467,"it":468,"ru":469,"sr":470,"zh":471},"Schnellidentifikations-Checkliste","Fast Identification Checklist","Lista de identificación rápida","Liste d'identification rapide","Checklist identificazione rapida","Чек-лист быстрой идентификации","Kontrolna lista za brzu identifikaciju","快速识别清单","\u002Famiibo\u002Fidentify-checklist","i-lucide-list-checks",[],{"id":476,"title":477,"url":486,"target":62,"icon":487,"isActive":15,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":488},"2-6",{"de":478,"en":479,"es":480,"fr":481,"it":482,"ru":483,"sr":484,"zh":485},"Tipps für Sealed-Sammlungen","Sealed Collection Tips","Consejos de colección sellada","Conseils collection scellée","Consigli collezione sigillata","Советы по коллекции Sealed","Saveti za zapečaćene kolekcije","密封收藏贴士","\u002Famiibo\u002Fsealed-tips","i-lucide-shield-check",[],{"id":490,"title":491,"url":500,"target":62,"icon":501,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":502},"2-5",{"de":492,"en":493,"es":494,"fr":495,"it":496,"ru":497,"sr":498,"zh":499},"Zustand & Bewertung","Condition & Grading","Estado & Graduación","État & Évaluation","Condizioni & Valutazione","Состояние & Оценка","Stanje & ocenjivanje","品相与分级","\u002Famiibo-library\u002Famiibo-condition-and-grading","i-lucide-badge-check",[],{"id":504,"title":505,"url":514,"target":62,"icon":515,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":516},"2-2",{"de":506,"en":507,"es":508,"fr":509,"it":510,"ru":511,"sr":512,"zh":513},"Ausgaben & Nachdrucke","Editions & Reprints","Ediciones & Reimpresiones","Éditions & Réimpressions","Edizioni & ristampe","Издания & переиздания","Izdanja & reizdanja","版本 & 重印","\u002Famiibo\u002Fcollecting\u002F","i-lucide-layers",[],{"id":518,"title":519,"url":528,"target":62,"icon":529,"isActive":15,"type":105,"productId":14,"categoryId":14,"shopCategoryId":14,"articleId":14,"pageId":14,"portfolioId":14,"children":530},"2-7",{"de":520,"en":521,"es":522,"fr":523,"it":524,"ru":525,"sr":526,"zh":527},"Nachdruck-Timeline","Reprint Timeline","Cronología de reimpresión","Chronologie des réimpressions","Cronologia ristampe","Хронология переизданий","Hronologija reprinta","重印时间轴","\u002Freference\u002Ftimelines\u002Famiibo-reprint-timeline","i-lucide-timer",[],{"statusCode":4,"data":532,"message":1591},{"id":533,"title":534,"slug":535,"content":536,"contentJson":537,"excerpt":14,"featuredImage":989,"featuredImageAlt":990,"featuredImageCaption":14,"featuredImageTitle":14,"featuredImageCopyright":14,"featuredImageAuthor":14,"featuredImageSourceUrl":14,"featuredImageLicense":14,"featuredImageIsAiGenerated":606,"status":991,"publishedAt":992,"createdAt":993,"updatedAt":994,"seoLocalePaths":995,"categories":1004,"author":1024,"translations":1027},"456","一台类似Spark的游戏PC究竟会带来什么改变？统一内存、AI计算与经典的GPU分工","what-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u003Cp>DGX Spark 是为 AI 开发而构建的，并非为了取代 GeForce 游戏台式机。这正是它作为一个思想实验有价值的原因。它将 CPU、GPU、Tensor 和 RT 能力围绕一个大型一致内存池打包在一起，使整个系统看起来不像一台拥有两个主要内存孤岛的经典 PC，而更像一台异构计算机器。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--info my-6 rounded-xl border p-5 border-blue-300 bg-blue-50 dark:border-blue-900 dark:bg-blue-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">直接回答\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">\u003Cstrong>一台类 Spark 的游戏 PC 改变数据拓扑的程度，会超过它改变 GPU 概念的程度。\u003C\u002Fstrong> 潜在收益是一个大型共享池，CPU、图形和本地 AI 工作负载可以更一致地访问它。直接代价是共享容量不会自动提供独立 GPU 级别的带宽，而更紧密的集成通常会降低模块化可升级性。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">这是一个架构思想实验\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">DGX Spark 的规格是事实。后文提出的游戏架构\u003Cstrong>不是\u003C\u002Fstrong>已发布的 NVIDIA 产品，本文也不是在预测“DGX Spark 游戏版”。Spark 证明了一个商用 Blackwell 系统可以结合 Arm CPU 核心、GPU\u002FTensor\u002FRT 资源和大型一致内存池。“类 Spark”意味着借用这些设计原则进行游戏思想实验，而不是预测特定 SKU 或路线图。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"目录\">\u003Cstrong class=\"editorjs-toc__title\">目录\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-5\" class=\"editorjs-toc__link\">Spark 实际上是什么\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-7\" class=\"editorjs-toc__link\">经典 PC 分离造成重复和移动\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-10\" class=\"editorjs-toc__link\">统一设计改变了问题\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-13\" class=\"editorjs-toc__link\">容量和带宽讲述了截然不同的故事\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-17\" class=\"editorjs-toc__link\">压缩和预测可以使低带宽统一池更具可行性\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">为什么神经渲染让统一内存更有趣\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-22\" class=\"editorjs-toc__link\">游戏版本需要什么Spark未优化的功能？\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-24\" class=\"editorjs-toc__link\">最有趣的版本可能是混合型\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-27\" class=\"editorjs-toc__link\">买家获得什么——以及放弃什么\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-29\" class=\"editorjs-toc__link\">什么会证伪这个想法？\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">主要来源\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Ch2 id=\"section-5\">Spark 实际上是什么\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">DGX Spark 规格\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">官方数值\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">为何对本思想实验重要\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">架构\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Grace Blackwell \u002F GB10\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">CPU 和 Blackwell GPU 被设计为一个集成计算平台\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">CPU\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">20 核 Arm\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">打破了传统 x86 游戏台式机的假设\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">系统内存\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">64 GB 或 128 GB 一致统一 LPDDR5X\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">大型公共池，而非独立的 CPU RAM + 专用 VRAM\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">内存带宽\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">273 GB\u002Fs\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">对集成系统而言很强，但远低于顶级独立显卡内存带宽\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">AI 计算\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">最高 1 PFLOP FP4\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">展示了设计优先级可以多大程度转向矩阵\u002FTensor 风格工作负载\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">RT 硬件\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">第四代 RT 核心\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">SoC 仍包含专用实时图形加速\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-7\">经典 PC 分离造成重复和移动\u003C\u002Fh2>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">传统独立显卡游戏数据路径\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">CPU 侧系统内存\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">游戏逻辑、流式缓冲区和许多资源从系统 RAM 开始。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">PCIe\u002F数据移动\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">资源根据需要被复制或映射到 GPU。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">独立 VRAM\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">GPU 将其工作集保存在单独的高带宽池中。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">渲染\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">着色器、RT 和计算消耗这些驻留资源。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">驱逐\u002F流式传输\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">当工作集变化时，数据再次移动。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Cp>这种分离本身并不坏。它允许 GPU 拥有极快的专用内存，并让用户独立升级组件。但它也创造了边界：容量可能被困在一侧，数据可能被重复，应用程序需要显式的移动和驻留策略。\u003C\u002Fp>\n\u003Ch2 id=\"section-10\">统一设计改变了问题\u003C\u002Fh2>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">类 Spark 数据模型\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">一个一致池\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">CPU 和 GPU 看到一个大得多的公共可寻址内存资源。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">专用引擎\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">图形、RT 和矩阵\u002FTensor 工作负载从同一系统级池中获取。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">更少的强制重复\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">某些工作负载可以避免维护单独的 CPU 和 GPU 副本。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">新的争用问题\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">所有计算引擎现在可以竞争同一内存带宽。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">软件变得更加重要\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">调度和局部性决定统一内存是优雅还是拥堵。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Csection class=\"editorjs-comparison my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">独立显卡游戏 PC 与类 Spark 原则\u003C\u002Fh3>\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left dark:border-gray-700 dark:bg-gray-900\">\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">独立显卡游戏 PC\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">类 Spark 集成系统\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">内存容量\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">系统 RAM + 独立 VRAM 池\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">大型一致共享池\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">峰值图形带宽\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">使用 GDDR\u002FHBM 风格专用内存可以达到极高\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">必须设计得足够高以应对图形 + CPU + AI 争用\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">可升级性\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">GPU、RAM 和有时 CPU 可独立更换\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">更可能是平台级更换\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">本地 AI 适配性\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">AI 竞争 GPU VRAM 或独立 NPU\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">大型共享池可以使模型更容易容纳\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">数据移动\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">显式传输和重复可能很重要\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">可能减少复制，但局部性仍然重要\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-13\">容量和带宽讲述了截然不同的故事\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">参考架构\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">内存容量\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">内存带宽\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">设计教训\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">DGX Spark\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">128 GB 一致统一 LPDDR5X\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">273 GB\u002Fs\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">巨大的共享容量，但不是独立高端图形带宽\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">RTX PRO 6000 Blackwell 工作站\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">96 GB GDDR7\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">1,792 GB\u002Fs\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">专业独立 GPU 展示了高端图形\u002FAI 工作负载可以要求多少带宽\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">GeForce RTX 5090\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">32 GB GDDR7，512 位\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">高端独立设计\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">游戏旗舰强调带宽和专用本地内存，而非巨大的统一容量\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Cp>重点不是一种架构普遍更好。Spark 展示了一致容量；RTX PRO 6000 展示了极端的本地带宽。试图将类 Spark 的一致性与旗舰级图形结合的游戏架构，需要调和这些优先级，而不是原样复制某一产品。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--warning my-6 rounded-xl border p-5 border-amber-300 bg-amber-50 dark:border-amber-900 dark:bg-amber-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">带宽墙\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">Spark 的 \u003Cstrong>273 GB\u002Fs\u003C\u002Fstrong> 是一个有用的提醒：内存容量和内存带宽是不同的规格。游戏衍生品不能简单地复制 128 GB LPDDR5X 并假设它能取代高端独立 GPU。它需要更激进的带宽、缓存、压缩或工作负载调度——可能四者都需要。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-17\">压缩和预测可以使低带宽统一池更具可行性\u003C\u002Fh2>\n\u003Cp>未来的集成设计不必通过提高引脚速度来解决所有问题。压缩可以减少移动的字节，神经表示可以减少存储的细节，预测性预取可以隐藏一些访问延迟，缓存层次结构可以使热图形数据靠近 GPU。这些技术不会消除带宽墙，但它们可以减少软件撞墙的频率。\u003C\u002Fp>\n\u003Ch2 id=\"section-19\">为什么神经渲染让统一内存更有趣\u003C\u002Fh2>\n\u003Cp>经典图形学主要关注GPU能否将纹理、几何体和缓冲区保持得足够近以渲染场景。神经图形学可以增加模型权重、潜在表示、历史缓冲区和中间张量。一个大的连贯内存池使这些新数据类别更容易在CPU侧游戏系统和GPU侧推理之间共享，尤其是当本地游戏助手或NPC模型也在运行时。\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Ffigure.rocks\u002Fzh\u002Fblog\u002Fdirectx-is-becoming-an-ml-platform-what-linear-algebra-and-neural-shaders-mean-for-future-games\" class=\"flex flex-col sm:flex-row gap-4 rounded-xl border border-gray-200 dark:border-gray-700 p-4 transition hover:border-primary-500\">\u003Cdiv class=\"min-w-0 flex-1\">\u003Cstrong class=\"block text-lg text-gray-900 dark:text-gray-100\">DirectX正在成为ML平台：线性代数和神经着色器对未来游戏意味着什么\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">这一转变的软件方面：机器学习数学正在进入图形编程模型本身。\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">阅读DirectX ML分析 →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Ch2 id=\"section-22\">游戏版本需要什么Spark未优化的功能？\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">游戏需求\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">为什么类似Spark的集成有帮助\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">仍需解决的问题\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">高刷新率图形\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">紧密的CPU\u002FGPU协调\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">更高的持续图形带宽和低延迟\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">大型神经模型\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">共享容量减少适配\u002F重复问题\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">推理不能不可预测地占用帧时间\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">资产流式传输\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">单一池简化驻留关系\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">存储到内存的路径和解压缩必须保持快速\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">竞技延迟\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">减少传输可能有帮助\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">统一调度不能增加争用或长队列\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">长产品寿命\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">集成平台可以节能\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">维修\u002F升级路径可能比模块化PC更差\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-24\">最有趣的版本可能是混合型\u003C\u002Fh2>\n\u003Caside class=\"editorjs-callout editorjs-callout--success my-6 rounded-xl border p-5 border-emerald-300 bg-emerald-50 dark:border-emerald-900 dark:bg-emerald-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">Figure Rocks假设\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">未来的高端游戏系统可能不会在\u003Cstrong>完全独立\u003C\u002Fstrong>和\u003Cstrong>完全统一\u003C\u002Fstrong>之间选择。混合型可以结合大型连贯内存池与高带宽本地缓存或图形内存层，允许AI和游戏状态工作负载共享容量，同时延迟敏感的渲染保持非常快的本地路径。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">可能的混合游戏内存层次结构\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">大型连贯池\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">存放游戏资产、AI模型、CPU状态和延迟不敏感的数据。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">预测性驻留管理器\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">在需求之前将可能需要的资源移动到更快的内存。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">高带宽图形层\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">保护当前渲染工作集。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">神经压缩\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">在需要之前保持选定的资产或表示更小。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">引擎调度器\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">根据帧截止时间平衡CPU、图形、RT和神经工作负载。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-27\">买家获得什么——以及放弃什么\u003C\u002Fh2>\n\u003Csection class=\"editorjs-comparison my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">买家权衡\u003C\u002Fh3>\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left dark:border-gray-700 dark:bg-gray-900\">\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">潜在收益\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">潜在成本\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">非常大的可用池\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">RAM、VRAM和模型内存之间的硬墙更少\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">共享争用和不太明显的性能规格\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">本地AI余量\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">更大的模型可以与游戏共存\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">AI工作负载可能与渲染竞争\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">集成\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">潜在的效率和紧凑性\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">热限制可能限制峰值图形性能\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">平台简单性\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">更少的兼容性边界\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">整个平台升级而不是仅GPU升级\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">优化\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">引擎可以协调更多系统\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">糟糕的软件支持可能浪费昂贵的硬件\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">预测性内存管理\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">预测准确时资产\u002F页面到达更少延迟\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">错误的预测可能浪费带宽、驻留和缓存\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">平台级优化\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">引擎、驱动和硬件可以更深入地协调数据放置和加速器\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">更大的生态系统锁定和更不透明的性能行为\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-29\">什么会证伪这个想法？\u003C\u002Fh2>\n\u003Cp>如果高端独立GPU继续在成本、带宽和延迟方面提供更好的表现，而本地游戏AI仍然是可选的，那么可能几乎没有理由接受统一架构的锁定和争用。只有当游戏越来越需要大型共享数据集和神经工作负载，而这些在传统的CPU-RAM + 独立VRAM分割中难以适配时，这个假设才会变得更强。\u003C\u002Fp>\n\u003Ch2 id=\"section-31\">主要来源\u003C\u002Fh2>\n\u003Ca href=\"https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fproducts\u002Fworkstations\u002Fdgx-spark\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">NVIDIA DGX Spark — 官方规格\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">官方GB10架构、统一内存、带宽、CPU、GPU和AI计算规格。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.nvidia.com\u002Fde-de\u002Fproducts\u002Fworkstations\u002Fprofessional-desktop-gpus\u002Frtx-pro-6000\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">NVIDIA — RTX PRO 6000 Blackwell工作站版\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">高端独立Blackwell GPU的官方96 GB GDDR7和1,792 GB\u002Fs带宽参考。\u003C\u002Fp>\u003C\u002Fa>\n\u003Csection class=\"editorjs-faq my-6 rounded-xl border border-gray-200 p-5 dark:border-gray-700\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">类似Spark的游戏架构\u003C\u002Fh3>\u003Cdiv id=\"faq1\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">DGX Spark是游戏PC吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">不是。NVIDIA将其定位为个人AI超级计算机。本文以其架构作为思想实验。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq2\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">128 GB统一内存会胜过32 GB高端GPU吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">不会自动。容量、带宽、延迟和软件优化是不同的约束。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq3\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">为什么统一内存对游戏AI有吸引力？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">它可以减少CPU、GPU和本地模型工作负载之间的硬容量边界和重复。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq4\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">最大的风险是什么？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">共享内存争用加上减少的模块化可升级性。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq5\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">混合型会更好吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">有可能。大型连贯池加上快速图形本地层可以在为AI和通用工作负载提供更多共享容量的同时保持带宽。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Csection class=\"editorjs-glossary my-6 rounded-xl border border-gray-200 dark:border-gray-700 p-5\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">架构术语\u003C\u002Fh3>\u003Cdl>\u003Cdiv id=\"coherent-memory\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">连贯统一内存\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">一种内存架构，多个处理器可以访问共享系统内存空间并具有一致性保证。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"memory-tier\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">内存层\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">层次结构中具有不同容量、带宽和延迟特性的层。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"heterogeneous-compute\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">异构计算\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">使用不同类型的处理器或加速器块来处理不同的工作负载类别。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"spark-like\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">类似Spark的游戏架构\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Figure Rocks的简写，指从DGX Spark等系统借鉴集成CPU\u002FGPU\u002FAI和连贯内存原则；并非已宣布的产品。\u003C\u002Fdd>\u003C\u002Fdiv>\u003C\u002Fdl>\u003C\u002Fsection>",{"time":538,"blocks":539,"version":988},1792340171384,[540,546,554,561,568,574,609,614,637,642,647,668,712,717,739,744,751,756,761,766,771,780,785,814,819,826,847,852,905,910,915,920,930,939,966],{"id":541,"data":542,"type":544,"tunes":545},"p-0111",{"text":543},"DGX Spark 是为 AI 开发而构建的，并非为了取代 GeForce 游戏台式机。这正是它作为一个思想实验有价值的原因。它将 CPU、GPU、Tensor 和 RT 能力围绕一个大型一致内存池打包在一起，使整个系统看起来不像一台拥有两个主要内存孤岛的经典 PC，而更像一台异构计算机器。","paragraph",{},{"id":547,"data":548,"type":552,"tunes":553},"c-0112",{"body":549,"title":550,"variant":551},"\u003Cstrong>一台类 Spark 的游戏 PC 改变数据拓扑的程度，会超过它改变 GPU 概念的程度。\u003C\u002Fstrong> 潜在收益是一个大型共享池，CPU、图形和本地 AI 工作负载可以更一致地访问它。直接代价是共享容量不会自动提供独立 GPU 级别的带宽，而更紧密的集成通常会降低模块化可升级性。","直接回答","info","callout",{},{"id":555,"data":556,"type":552,"tunes":560},"c-0113",{"body":557,"title":558,"variant":559},"DGX Spark 的规格是事实。后文提出的游戏架构\u003Cstrong>不是\u003C\u002Fstrong>已发布的 NVIDIA 产品，本文也不是在预测“DGX Spark 游戏版”。Spark 证明了一个商用 Blackwell 系统可以结合 Arm CPU 核心、GPU\u002FTensor\u002FRT 资源和大型一致内存池。“类 Spark”意味着借用这些设计原则进行游戏思想实验，而不是预测特定 SKU 或路线图。","这是一个架构思想实验","note",{},{"id":562,"data":563,"type":566,"tunes":567},"toc-0114",{"title":564,"maxLevel":565,"minLevel":47},"目录",3,"tableOfContents",{},{"id":569,"data":570,"type":572,"tunes":573},"h-0115",{"text":571,"level":47},"Spark 实际上是什么","header",{},{"id":575,"data":576,"type":607,"tunes":608},"t-0116",{"content":577,"stretched":606,"withHeadings":15},[578,582,586,590,594,598,602],[579,580,581],"DGX Spark 规格","官方数值","为何对本思想实验重要",[583,584,585],"架构","Grace Blackwell \u002F GB10","CPU 和 Blackwell GPU 被设计为一个集成计算平台",[587,588,589],"CPU","20 核 Arm","打破了传统 x86 游戏台式机的假设",[591,592,593],"系统内存","64 GB 或 128 GB 一致统一 LPDDR5X","大型公共池，而非独立的 CPU RAM + 专用 VRAM",[595,596,597],"内存带宽","273 GB\u002Fs","对集成系统而言很强，但远低于顶级独立显卡内存带宽",[599,600,601],"AI 计算","最高 1 PFLOP FP4","展示了设计优先级可以多大程度转向矩阵\u002FTensor 风格工作负载",[603,604,605],"RT 硬件","第四代 RT 核心","SoC 仍包含专用实时图形加速",false,"table",{},{"id":610,"data":611,"type":572,"tunes":613},"h-0117",{"text":612,"level":47},"经典 PC 分离造成重复和移动",{},{"id":615,"data":616,"type":635,"tunes":636},"flow-0118",{"steps":617,"title":633,"orientation":634},[618,621,624,627,630],{"label":619,"description":620},"CPU 侧系统内存","游戏逻辑、流式缓冲区和许多资源从系统 RAM 开始。",{"label":622,"description":623},"PCIe\u002F数据移动","资源根据需要被复制或映射到 GPU。",{"label":625,"description":626},"独立 VRAM","GPU 将其工作集保存在单独的高带宽池中。",{"label":628,"description":629},"渲染","着色器、RT 和计算消耗这些驻留资源。",{"label":631,"description":632},"驱逐\u002F流式传输","当工作集变化时，数据再次移动。","传统独立显卡游戏数据路径","auto","processFlow",{},{"id":638,"data":639,"type":544,"tunes":641},"p-0119",{"text":640},"这种分离本身并不坏。它允许 GPU 拥有极快的专用内存，并让用户独立升级组件。但它也创造了边界：容量可能被困在一侧，数据可能被重复，应用程序需要显式的移动和驻留策略。",{},{"id":643,"data":644,"type":572,"tunes":646},"h-0120",{"text":645,"level":47},"统一设计改变了问题",{},{"id":648,"data":649,"type":635,"tunes":667},"flow-0121",{"steps":650,"title":666,"orientation":634},[651,654,657,660,663],{"label":652,"description":653},"一个一致池","CPU 和 GPU 看到一个大得多的公共可寻址内存资源。",{"label":655,"description":656},"专用引擎","图形、RT 和矩阵\u002FTensor 工作负载从同一系统级池中获取。",{"label":658,"description":659},"更少的强制重复","某些工作负载可以避免维护单独的 CPU 和 GPU 副本。",{"label":661,"description":662},"新的争用问题","所有计算引擎现在可以竞争同一内存带宽。",{"label":664,"description":665},"软件变得更加重要","调度和局部性决定统一内存是优雅还是拥堵。","类 Spark 数据模型",{},{"id":669,"data":670,"type":710,"tunes":711},"cmp-0122",{"rows":671,"title":702,"layout":607,"columns":703},[672,678,684,690,696],{"id":673,"label":674,"values":675},"capacity","内存容量",{"disc":676,"unified":677},"系统 RAM + 独立 VRAM 池","大型一致共享池",{"id":679,"label":680,"values":681},"bandwidth","峰值图形带宽",{"disc":682,"unified":683},"使用 GDDR\u002FHBM 风格专用内存可以达到极高","必须设计得足够高以应对图形 + CPU + AI 争用",{"id":685,"label":686,"values":687},"upgrade","可升级性",{"disc":688,"unified":689},"GPU、RAM 和有时 CPU 可独立更换","更可能是平台级更换",{"id":691,"label":692,"values":693},"ai","本地 AI 适配性",{"disc":694,"unified":695},"AI 竞争 GPU VRAM 或独立 NPU","大型共享池可以使模型更容易容纳",{"id":697,"label":698,"values":699},"efficiency","数据移动",{"disc":700,"unified":701},"显式传输和重复可能很重要","可能减少复制，但局部性仍然重要","独立显卡游戏 PC 与类 Spark 原则",[704,707],{"id":705,"label":706},"disc","独立显卡游戏 PC",{"id":708,"label":709},"unified","类 Spark 集成系统","comparison",{},{"id":713,"data":714,"type":572,"tunes":716},"v3-numbers",{"text":715,"level":47},"容量和带宽讲述了截然不同的故事",{},{"id":718,"data":719,"type":607,"tunes":738},"v3-bandwidth-comparison",{"content":720,"stretched":606,"withHeadings":15},[721,724,728,733],[722,674,595,723],"参考架构","设计教训",[725,726,596,727],"DGX Spark","128 GB 一致统一 LPDDR5X","巨大的共享容量，但不是独立高端图形带宽",[729,730,731,732],"RTX PRO 6000 Blackwell 工作站","96 GB GDDR7","1,792 GB\u002Fs","专业独立 GPU 展示了高端图形\u002FAI 工作负载可以要求多少带宽",[734,735,736,737],"GeForce RTX 5090","32 GB GDDR7，512 位","高端独立设计","游戏旗舰强调带宽和专用本地内存，而非巨大的统一容量",{},{"id":740,"data":741,"type":544,"tunes":743},"v3-bandwidth-ratio",{"text":742},"重点不是一种架构普遍更好。Spark 展示了一致容量；RTX PRO 6000 展示了极端的本地带宽。试图将类 Spark 的一致性与旗舰级图形结合的游戏架构，需要调和这些优先级，而不是原样复制某一产品。",{},{"id":745,"data":746,"type":552,"tunes":750},"c-0123",{"body":747,"title":748,"variant":749},"Spark 的 \u003Cstrong>273 GB\u002Fs\u003C\u002Fstrong> 是一个有用的提醒：内存容量和内存带宽是不同的规格。游戏衍生品不能简单地复制 128 GB LPDDR5X 并假设它能取代高端独立 GPU。它需要更激进的带宽、缓存、压缩或工作负载调度——可能四者都需要。","带宽墙","warning",{},{"id":752,"data":753,"type":572,"tunes":755},"v3-compression-path",{"text":754,"level":47},"压缩和预测可以使低带宽统一池更具可行性",{},{"id":757,"data":758,"type":544,"tunes":760},"v3-compression-path-p",{"text":759},"未来的集成设计不必通过提高引脚速度来解决所有问题。压缩可以减少移动的字节，神经表示可以减少存储的细节，预测性预取可以隐藏一些访问延迟，缓存层次结构可以使热图形数据靠近 GPU。这些技术不会消除带宽墙，但它们可以减少软件撞墙的频率。",{},{"id":762,"data":763,"type":572,"tunes":765},"h-0124",{"text":764,"level":47},"为什么神经渲染让统一内存更有趣",{},{"id":767,"data":768,"type":544,"tunes":770},"p-0125",{"text":769},"经典图形学主要关注GPU能否将纹理、几何体和缓冲区保持得足够近以渲染场景。神经图形学可以增加模型权重、潜在表示、历史缓冲区和中间张量。一个大的连贯内存池使这些新数据类别更容易在CPU侧游戏系统和GPU侧推理之间共享，尤其是当本地游戏助手或NPC模型也在运行时。",{},{"id":772,"data":773,"type":778,"tunes":779},"ref-0126",{"url":774,"title":775,"excerpt":776,"ctaLabel":777},"https:\u002F\u002Ffigure.rocks\u002Fzh\u002Fblog\u002Fdirectx-is-becoming-an-ml-platform-what-linear-algebra-and-neural-shaders-mean-for-future-games","DirectX正在成为ML平台：线性代数和神经着色器对未来游戏意味着什么","这一转变的软件方面：机器学习数学正在进入图形编程模型本身。","阅读DirectX ML分析","referralArticle",{},{"id":781,"data":782,"type":572,"tunes":784},"h-0127",{"text":783,"level":47},"游戏版本需要什么Spark未优化的功能？",{},{"id":786,"data":787,"type":607,"tunes":813},"t-0128",{"content":788,"stretched":606,"withHeadings":15},[789,793,797,801,805,809],[790,791,792],"游戏需求","为什么类似Spark的集成有帮助","仍需解决的问题",[794,795,796],"高刷新率图形","紧密的CPU\u002FGPU协调","更高的持续图形带宽和低延迟",[798,799,800],"大型神经模型","共享容量减少适配\u002F重复问题","推理不能不可预测地占用帧时间",[802,803,804],"资产流式传输","单一池简化驻留关系","存储到内存的路径和解压缩必须保持快速",[806,807,808],"竞技延迟","减少传输可能有帮助","统一调度不能增加争用或长队列",[810,811,812],"长产品寿命","集成平台可以节能","维修\u002F升级路径可能比模块化PC更差",{},{"id":815,"data":816,"type":572,"tunes":818},"h-0129",{"text":817,"level":47},"最有趣的版本可能是混合型",{},{"id":820,"data":821,"type":552,"tunes":825},"c-0130",{"body":822,"title":823,"variant":824},"未来的高端游戏系统可能不会在\u003Cstrong>完全独立\u003C\u002Fstrong>和\u003Cstrong>完全统一\u003C\u002Fstrong>之间选择。混合型可以结合大型连贯内存池与高带宽本地缓存或图形内存层，允许AI和游戏状态工作负载共享容量，同时延迟敏感的渲染保持非常快的本地路径。","Figure Rocks假设","success",{},{"id":827,"data":828,"type":635,"tunes":846},"flow-0131",{"steps":829,"title":845,"orientation":634},[830,833,836,839,842],{"label":831,"description":832},"大型连贯池","存放游戏资产、AI模型、CPU状态和延迟不敏感的数据。",{"label":834,"description":835},"预测性驻留管理器","在需求之前将可能需要的资源移动到更快的内存。",{"label":837,"description":838},"高带宽图形层","保护当前渲染工作集。",{"label":840,"description":841},"神经压缩","在需要之前保持选定的资产或表示更小。",{"label":843,"description":844},"引擎调度器","根据帧截止时间平衡CPU、图形、RT和神经工作负载。","可能的混合游戏内存层次结构",{},{"id":848,"data":849,"type":572,"tunes":851},"h-0132",{"text":850,"level":47},"买家获得什么——以及放弃什么",{},{"id":853,"data":854,"type":710,"tunes":904},"cmp-0133",{"rows":855,"title":896,"layout":607,"columns":897},[856,862,867,873,878,884,890],{"id":857,"label":858,"values":859},"memory","非常大的可用池",{"cost":860,"gain":861},"共享争用和不太明显的性能规格","RAM、VRAM和模型内存之间的硬墙更少",{"id":691,"label":863,"values":864},"本地AI余量",{"cost":865,"gain":866},"AI工作负载可能与渲染竞争","更大的模型可以与游戏共存",{"id":868,"label":869,"values":870},"power","集成",{"cost":871,"gain":872},"热限制可能限制峰值图形性能","潜在的效率和紧凑性",{"id":685,"label":874,"values":875},"平台简单性",{"cost":876,"gain":877},"整个平台升级而不是仅GPU升级","更少的兼容性边界",{"id":879,"label":880,"values":881},"software","优化",{"cost":882,"gain":883},"糟糕的软件支持可能浪费昂贵的硬件","引擎可以协调更多系统",{"id":885,"label":886,"values":887},"predictive","预测性内存管理",{"cost":888,"gain":889},"错误的预测可能浪费带宽、驻留和缓存","预测准确时资产\u002F页面到达更少延迟",{"id":891,"label":892,"values":893},"platform","平台级优化",{"cost":894,"gain":895},"更大的生态系统锁定和更不透明的性能行为","引擎、驱动和硬件可以更深入地协调数据放置和加速器","买家权衡",[898,901],{"id":899,"label":900},"gain","潜在收益",{"id":902,"label":903},"cost","潜在成本",{},{"id":906,"data":907,"type":572,"tunes":909},"h-0134",{"text":908,"level":47},"什么会证伪这个想法？",{},{"id":911,"data":912,"type":544,"tunes":914},"p-0135",{"text":913},"如果高端独立GPU继续在成本、带宽和延迟方面提供更好的表现，而本地游戏AI仍然是可选的，那么可能几乎没有理由接受统一架构的锁定和争用。只有当游戏越来越需要大型共享数据集和神经工作负载，而这些在传统的CPU-RAM + 独立VRAM分割中难以适配时，这个假设才会变得更强。",{},{"id":916,"data":917,"type":572,"tunes":919},"h-0136",{"text":918,"level":47},"主要来源",{},{"id":921,"data":922,"type":928,"tunes":929},"src-0137",{"link":923,"meta":924},"https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fproducts\u002Fworkstations\u002Fdgx-spark\u002F",{"image":925,"title":926,"description":927},{"url":13},"NVIDIA DGX Spark — 官方规格","官方GB10架构、统一内存、带宽、CPU、GPU和AI计算规格。","linkTool",{},{"id":931,"data":932,"type":928,"tunes":938},"v3-src-pro6000",{"link":933,"meta":934},"https:\u002F\u002Fwww.nvidia.com\u002Fde-de\u002Fproducts\u002Fworkstations\u002Fprofessional-desktop-gpus\u002Frtx-pro-6000\u002F",{"image":935,"title":936,"description":937},{"url":13},"NVIDIA — RTX PRO 6000 Blackwell工作站版","高端独立Blackwell GPU的官方96 GB GDDR7和1,792 GB\u002Fs带宽参考。",{},{"id":940,"data":941,"type":964,"tunes":965},"faq-0138",{"items":942,"title":963},[943,947,951,955,959],{"id":944,"answer":945,"question":946},"faq1","不是。NVIDIA将其定位为个人AI超级计算机。本文以其架构作为思想实验。","DGX Spark是游戏PC吗？",{"id":948,"answer":949,"question":950},"faq2","不会自动。容量、带宽、延迟和软件优化是不同的约束。","128 GB统一内存会胜过32 GB高端GPU吗？",{"id":952,"answer":953,"question":954},"faq3","它可以减少CPU、GPU和本地模型工作负载之间的硬容量边界和重复。","为什么统一内存对游戏AI有吸引力？",{"id":956,"answer":957,"question":958},"faq4","共享内存争用加上减少的模块化可升级性。","最大的风险是什么？",{"id":960,"answer":961,"question":962},"faq5","有可能。大型连贯池加上快速图形本地层可以在为AI和通用工作负载提供更多共享容量的同时保持带宽。","混合型会更好吗？","类似Spark的游戏架构","faq",{},{"id":967,"data":968,"type":986,"tunes":987},"gloss-0139",{"title":969,"entries":970},"架构术语",[971,975,979,983],{"term":972,"anchor":973,"definition":974},"连贯统一内存","coherent-memory","一种内存架构，多个处理器可以访问共享系统内存空间并具有一致性保证。",{"term":976,"anchor":977,"definition":978},"内存层","memory-tier","层次结构中具有不同容量、带宽和延迟特性的层。",{"term":980,"anchor":981,"definition":982},"异构计算","heterogeneous-compute","使用不同类型的处理器或加速器块来处理不同的工作负载类别。",{"term":963,"anchor":984,"definition":985},"spark-like","Figure Rocks的简写，指从DGX Spark等系统借鉴集成CPU\u002FGPU\u002FAI和连贯内存原则；并非已宣布的产品。","glossary",{},"2.31","\u002Fuploads\u002F2026\u002F10\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split-1792340021752-bia7n6.webp","what-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split-1792340021752-bia7n6","PUBLISHED","2026-10-18T12:12:00.000Z","2026-10-18T16:12:11.888Z","2026-10-18T16:16:16.306Z",{"en":996,"de":997,"sr":998,"es":999,"fr":1000,"it":1001,"ru":1002,"zh":1003},"\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Fde\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Fsr\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Fes\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Ffr\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Fit\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Fru\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split","\u002Fzh\u002Fblog\u002Fwhat-would-a-spark-like-gaming-pc-actually-change-unified-memory-ai-compute-and-the-classic-gpu-split",[1005,1008,1012,1016,1020],{"id":1006,"name":291,"slug":1007},144,"tech",{"id":1009,"name":1010,"slug":1011},170,"科技新闻","tech-news",{"id":1013,"name":1014,"slug":1015},280,"显示技术发展史","display-tech-timeline",{"id":1017,"name":1018,"slug":1019},165,"新闻","news",{"id":1021,"name":1022,"slug":1023},427,"新品","new",{"id":283,"login":1025,"email":13,"displayName":1026},"aleksandar","Aleksandar Stajic",[1028,1398],{"lang":8,"title":1029,"content":1030,"contentJson":1031,"excerpt":14},"What Would a Spark-Like Gaming PC Actually Change? Unified Memory, AI Compute and the Classic GPU Split","{\"time\":1792340022841,\"blocks\":[{\"id\":\"p-0111\",\"type\":\"paragraph\",\"data\":{\"text\":\"DGX Spark is built for AI development, not for replacing a GeForce gaming desktop. That is exactly why it is useful as a thought experiment. It packages CPU, GPU, Tensor and RT capability around a large coherent memory pool, making the entire system look less like a classic PC with two major memory islands and more like one heterogeneous compute machine.\"},\"tunes\":{}},{\"id\":\"c-0112\",\"type\":\"callout\",\"data\":{\"variant\":\"info\",\"title\":\"Direct answer\",\"body\":\"\u003Cstrong>A Spark-like gaming PC would change the data topology more than it changes the idea of a GPU.\u003C\u002Fstrong> The potential gain is a large shared pool that CPU, graphics and local AI workloads can access more coherently. The immediate cost is that shared capacity does not automatically provide discrete-GPU-class bandwidth, and tighter integration usually reduces modular upgradeability.\"},\"tunes\":{}},{\"id\":\"c-0113\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"This is an architectural thought experiment\",\"body\":\"DGX Spark specifications are factual. The gaming architecture proposed later is \u003Cstrong>not\u003C\u002Fstrong> an announced NVIDIA product and this article is not predicting a 'DGX Spark Gaming Edition.' Spark is evidence that a commercial Blackwell system can combine Arm CPU cores, GPU\u002FTensor\u002FRT resources and a large coherent memory pool. 'Spark-like' means borrowing those design principles for a gaming thought experiment, not predicting a specific SKU or roadmap.\"},\"tunes\":{}},{\"id\":\"toc-0114\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"h-0115\",\"type\":\"header\",\"data\":{\"text\":\"What Spark actually is\",\"level\":2},\"tunes\":{}},{\"id\":\"t-0116\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"DGX Spark specification\",\"Official value\",\"Why it matters to this thought experiment\"],[\"Architecture\",\"Grace Blackwell \u002F GB10\",\"CPU and Blackwell GPU are designed as one integrated compute platform\"],[\"CPU\",\"20-core Arm\",\"Breaks from the traditional x86 gaming-desktop assumption\"],[\"System memory\",\"64 GB or 128 GB coherent unified LPDDR5X\",\"Large common pool rather than separate CPU RAM + dedicated VRAM\"],[\"Memory bandwidth\",\"273 GB\u002Fs\",\"Strong for an integrated system, but well below top-end discrete graphics memory bandwidth\"],[\"AI compute\",\"Up to 1 PFLOP FP4\",\"Shows how much design priority can move toward matrix\u002FTensor-style workloads\"],[\"RT hardware\",\"4th-gen RT cores\",\"The SoC still includes dedicated real-time graphics acceleration\"]]},\"tunes\":{}},{\"id\":\"h-0117\",\"type\":\"header\",\"data\":{\"text\":\"The classic PC split creates duplication and movement\",\"level\":2},\"tunes\":{}},{\"id\":\"flow-0118\",\"type\":\"processFlow\",\"data\":{\"title\":\"Traditional discrete-gaming data path\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"CPU-side system memory\",\"description\":\"Game logic, streaming buffers and many assets begin in system RAM.\"},{\"label\":\"PCIe\u002Fdata movement\",\"description\":\"Resources are copied or mapped toward the GPU as needed.\"},{\"label\":\"Discrete VRAM\",\"description\":\"The GPU keeps its working set in a separate high-bandwidth pool.\"},{\"label\":\"Rendering\",\"description\":\"Shaders, RT and compute consume those resident resources.\"},{\"label\":\"Eviction\u002Fstreaming\",\"description\":\"When the working set changes, data moves again.\"}]},\"tunes\":{}},{\"id\":\"p-0119\",\"type\":\"paragraph\",\"data\":{\"text\":\"That separation is not inherently bad. It allows the GPU to have extremely fast dedicated memory and lets users upgrade components independently. But it also creates boundaries: capacity can be stranded on one side, data can be duplicated, and applications need explicit movement and residency policies.\"},\"tunes\":{}},{\"id\":\"h-0120\",\"type\":\"header\",\"data\":{\"text\":\"A unified design changes the question\",\"level\":2},\"tunes\":{}},{\"id\":\"flow-0121\",\"type\":\"processFlow\",\"data\":{\"title\":\"Spark-like data model\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"One coherent pool\",\"description\":\"CPU and GPU see a much larger common addressable memory resource.\"},{\"label\":\"Specialized engines\",\"description\":\"Graphics, RT and matrix\u002FTensor workloads pull from the same system-level pool.\"},{\"label\":\"Less mandatory duplication\",\"description\":\"Some workloads can avoid maintaining separate CPU and GPU copies.\"},{\"label\":\"New contention problem\",\"description\":\"All compute engines can now compete for the same memory bandwidth.\"},{\"label\":\"Software becomes more important\",\"description\":\"Scheduling and locality determine whether unified memory feels elegant or congested.\"}]},\"tunes\":{}},{\"id\":\"cmp-0122\",\"type\":\"comparison\",\"data\":{\"title\":\"Discrete gaming PC vs Spark-like principle\",\"layout\":\"table\",\"columns\":[{\"id\":\"disc\",\"label\":\"Discrete gaming PC\"},{\"id\":\"unified\",\"label\":\"Spark-like integrated system\"}],\"rows\":[{\"id\":\"capacity\",\"label\":\"Memory capacity\",\"values\":{\"disc\":\"System RAM + separate VRAM pools\",\"unified\":\"Large coherent shared pool\"}},{\"id\":\"bandwidth\",\"label\":\"Peak graphics bandwidth\",\"values\":{\"disc\":\"Can be extremely high with GDDR\u002FHBM-style dedicated memory\",\"unified\":\"Must be designed high enough for graphics + CPU + AI contention\"}},{\"id\":\"upgrade\",\"label\":\"Upgradeability\",\"values\":{\"disc\":\"GPU, RAM and sometimes CPU replaced independently\",\"unified\":\"More platform-level replacement likely\"}},{\"id\":\"ai\",\"label\":\"Local AI fit\",\"values\":{\"disc\":\"AI competes for GPU VRAM or separate NPU\",\"unified\":\"Large shared pool can make models easier to fit\"}},{\"id\":\"efficiency\",\"label\":\"Data movement\",\"values\":{\"disc\":\"Explicit transfers and duplication can matter\",\"unified\":\"Potentially less copying, but locality still matters\"}}]},\"tunes\":{}},{\"id\":\"v3-numbers\",\"type\":\"header\",\"data\":{\"text\":\"Capacity and bandwidth tell radically different stories\",\"level\":2},\"tunes\":{}},{\"id\":\"v3-bandwidth-comparison\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Reference architecture\",\"Memory capacity\",\"Memory bandwidth\",\"Design lesson\"],[\"DGX Spark\",\"128 GB coherent unified LPDDR5X\",\"273 GB\u002Fs\",\"Huge shared capacity, but not discrete-high-end graphics bandwidth\"],[\"RTX PRO 6000 Blackwell Workstation\",\"96 GB GDDR7\",\"1,792 GB\u002Fs\",\"Professional discrete GPU shows how much bandwidth high-end graphics\u002FAI workloads can demand\"],[\"GeForce RTX 5090\",\"32 GB GDDR7, 512-bit\",\"High-end discrete design\",\"Gaming flagship emphasizes bandwidth and dedicated local memory over giant coherent capacity\"]]},\"tunes\":{}},{\"id\":\"v3-bandwidth-ratio\",\"type\":\"paragraph\",\"data\":{\"text\":\"The point is not that one architecture is universally better. Spark demonstrates coherent capacity; the RTX PRO 6000 demonstrates extreme local bandwidth. A gaming architecture that tries to combine Spark-like coherence with flagship-class graphics would need to reconcile those priorities rather than copy one product unchanged.\"},\"tunes\":{}},{\"id\":\"c-0123\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"The bandwidth wall\",\"body\":\"Spark's \u003Cstrong>273 GB\u002Fs\u003C\u002Fstrong> is a useful reminder that memory capacity and memory bandwidth are different specifications. A gaming derivative could not simply copy 128 GB LPDDR5X and assume it replaces a high-end discrete GPU. It would need much more aggressive bandwidth, caching, compression or workload scheduling — possibly all four.\"},\"tunes\":{}},{\"id\":\"v3-compression-path\",\"type\":\"header\",\"data\":{\"text\":\"Compression and prediction could make a lower-bandwidth unified pool more viable\",\"level\":2},\"tunes\":{}},{\"id\":\"v3-compression-path-p\",\"type\":\"paragraph\",\"data\":{\"text\":\"A future integrated design does not have to solve every problem by increasing pin speed. Compression can reduce bytes moved, neural representations can reduce stored detail, predictive prefetch can hide some access latency, and a cache hierarchy can keep hot graphics data near the GPU. Those techniques do not erase the bandwidth wall, but they can reduce how often software hits it.\"},\"tunes\":{}},{\"id\":\"h-0124\",\"type\":\"header\",\"data\":{\"text\":\"Why neural rendering makes unified memory more interesting\",\"level\":2},\"tunes\":{}},{\"id\":\"p-0125\",\"type\":\"paragraph\",\"data\":{\"text\":\"Classic graphics mostly asks whether the GPU can keep textures, geometry and buffers close enough to render the scene. Neural graphics can add model weights, latent representations, history buffers and intermediate tensors. A large coherent pool makes those new data classes easier to share between CPU-side game systems and GPU-side inference, especially when a local game assistant or NPC model is also running.\"},\"tunes\":{}},{\"id\":\"ref-0126\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Ffigure.rocks\u002Fblog\u002Fdirectx-is-becoming-an-ml-platform-what-linear-algebra-and-neural-shaders-mean-for-future-games\",\"title\":\"DirectX Is Becoming an ML Platform: What Linear Algebra and Neural Shaders Mean for Future Games\",\"excerpt\":\"The software side of this transition: machine-learning math is entering the graphics programming model itself.\",\"ctaLabel\":\"Read the DirectX ML analysis\"},\"tunes\":{}},{\"id\":\"h-0127\",\"type\":\"header\",\"data\":{\"text\":\"What would a gaming version need that Spark does not optimize for?\",\"level\":2},\"tunes\":{}},{\"id\":\"t-0128\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Gaming requirement\",\"Why Spark-like integration helps\",\"What still must be solved\"],[\"High-refresh graphics\",\"Close CPU\u002FGPU coordination\",\"Much higher sustained graphics bandwidth and low latency\"],[\"Large neural models\",\"Shared capacity reduces fit\u002Fduplication problems\",\"Inference cannot steal frame time unpredictably\"],[\"Asset streaming\",\"One pool simplifies residency relationships\",\"Storage-to-memory path and decompression must stay fast\"],[\"Competitive latency\",\"Fewer transfers can help\",\"Unified scheduling must not add contention or long queues\"],[\"Long product life\",\"Integrated platform can be power efficient\",\"Repair\u002Fupgrade path can become worse than a modular PC\"]]},\"tunes\":{}},{\"id\":\"h-0129\",\"type\":\"header\",\"data\":{\"text\":\"The most interesting version is probably hybrid\",\"level\":2},\"tunes\":{}},{\"id\":\"c-0130\",\"type\":\"callout\",\"data\":{\"variant\":\"success\",\"title\":\"Figure Rocks hypothesis\",\"body\":\"A future high-end gaming system may not choose between \u003Cstrong>fully discrete\u003C\u002Fstrong> and \u003Cstrong>fully unified\u003C\u002Fstrong>. A hybrid could combine a large coherent memory pool with a high-bandwidth local cache or graphics-memory tier, allowing AI and game-state workloads to share capacity while latency-sensitive rendering keeps a very fast local path.\"},\"tunes\":{}},{\"id\":\"flow-0131\",\"type\":\"processFlow\",\"data\":{\"title\":\"A possible hybrid gaming memory hierarchy\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"Large coherent pool\",\"description\":\"Holds game assets, AI models, CPU state and less latency-sensitive data.\"},{\"label\":\"Predictive residency manager\",\"description\":\"Moves likely-needed resources toward faster memory before demand.\"},{\"label\":\"High-bandwidth graphics tier\",\"description\":\"Protects the current render working set.\"},{\"label\":\"Neural compression\",\"description\":\"Keeps selected assets or representations smaller until needed.\"},{\"label\":\"Engine scheduler\",\"description\":\"Balances CPU, graphics, RT and neural workloads against frame deadlines.\"}]},\"tunes\":{}},{\"id\":\"h-0132\",\"type\":\"header\",\"data\":{\"text\":\"What the buyer gains — and gives up\",\"level\":2},\"tunes\":{}},{\"id\":\"cmp-0133\",\"type\":\"comparison\",\"data\":{\"title\":\"Buyer trade-off\",\"layout\":\"table\",\"columns\":[{\"id\":\"gain\",\"label\":\"Potential gain\"},{\"id\":\"cost\",\"label\":\"Potential cost\"}],\"rows\":[{\"id\":\"memory\",\"label\":\"Very large useful pool\",\"values\":{\"gain\":\"Fewer hard walls between RAM, VRAM and model memory\",\"cost\":\"Shared contention and less obvious performance specs\"}},{\"id\":\"ai\",\"label\":\"Local AI headroom\",\"values\":{\"gain\":\"Bigger models can coexist with games\",\"cost\":\"AI workloads can compete with rendering\"}},{\"id\":\"power\",\"label\":\"Integration\",\"values\":{\"gain\":\"Potential efficiency and compactness\",\"cost\":\"Thermal limits may cap peak graphics performance\"}},{\"id\":\"upgrade\",\"label\":\"Platform simplicity\",\"values\":{\"gain\":\"Fewer compatibility boundaries\",\"cost\":\"Whole-platform upgrade instead of GPU-only upgrade\"}},{\"id\":\"software\",\"label\":\"Optimization\",\"values\":{\"gain\":\"Engine can coordinate more of the system\",\"cost\":\"Poor software support can waste expensive hardware\"}},{\"id\":\"predictive\",\"label\":\"Predictive memory management\",\"values\":{\"gain\":\"Fewer late asset\u002Fpage arrivals when prediction is accurate\",\"cost\":\"Wrong predictions can waste bandwidth, residency and cache\"}},{\"id\":\"platform\",\"label\":\"Platform-level optimization\",\"values\":{\"gain\":\"Engine, driver and hardware can coordinate data placement and accelerators more deeply\",\"cost\":\"Greater ecosystem lock-in and more opaque performance behavior\"}}]},\"tunes\":{}},{\"id\":\"h-0134\",\"type\":\"header\",\"data\":{\"text\":\"What would falsify the idea?\",\"level\":2},\"tunes\":{}},{\"id\":\"p-0135\",\"type\":\"paragraph\",\"data\":{\"text\":\"If high-end discrete GPUs continue to deliver much better cost, bandwidth and latency while local gaming AI remains optional, there may be little reason to accept the lock-in and contention of a unified architecture. The hypothesis becomes stronger only if games increasingly need large shared data sets and neural workloads that are awkward to fit inside a traditional CPU-RAM + discrete-VRAM split.\"},\"tunes\":{}},{\"id\":\"h-0136\",\"type\":\"header\",\"data\":{\"text\":\"Primary source\",\"level\":2},\"tunes\":{}},{\"id\":\"src-0137\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fproducts\u002Fworkstations\u002Fdgx-spark\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"NVIDIA DGX Spark — Official specifications\",\"description\":\"Official GB10 architecture, unified memory, bandwidth, CPU, GPU and AI-compute specifications.\"}},\"tunes\":{}},{\"id\":\"v3-src-pro6000\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.nvidia.com\u002Fde-de\u002Fproducts\u002Fworkstations\u002Fprofessional-desktop-gpus\u002Frtx-pro-6000\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"NVIDIA — RTX PRO 6000 Blackwell Workstation Edition\",\"description\":\"Official 96 GB GDDR7 and 1,792 GB\u002Fs bandwidth reference for a high-end discrete Blackwell GPU.\"}},\"tunes\":{}},{\"id\":\"faq-0138\",\"type\":\"faq\",\"data\":{\"title\":\"Spark-like gaming architecture\",\"items\":[{\"id\":\"faq1\",\"question\":\"Is DGX Spark a gaming PC?\",\"answer\":\"No. NVIDIA positions it as a personal AI supercomputer. This article uses its architecture as a thought experiment.\"},{\"id\":\"faq2\",\"question\":\"Would 128 GB unified memory beat a 32 GB high-end GPU?\",\"answer\":\"Not automatically. Capacity, bandwidth, latency and software optimization are different constraints.\"},{\"id\":\"faq3\",\"question\":\"Why is unified memory attractive for gaming AI?\",\"answer\":\"It can reduce hard capacity boundaries and duplication between CPU, GPU and local-model workloads.\"},{\"id\":\"faq4\",\"question\":\"What is the biggest risk?\",\"answer\":\"Shared-memory contention plus reduced modular upgradeability.\"},{\"id\":\"faq5\",\"question\":\"Could a hybrid be better?\",\"answer\":\"Potentially. A large coherent pool plus a fast graphics-local tier could preserve bandwidth while giving AI and general workloads more shared capacity.\"}]},\"tunes\":{}},{\"id\":\"gloss-0139\",\"type\":\"glossary\",\"data\":{\"title\":\"Architecture terms\",\"entries\":[{\"term\":\"Coherent unified memory\",\"definition\":\"A memory architecture where multiple processors can access a shared system memory space with coherence guarantees.\",\"anchor\":\"coherent-memory\"},{\"term\":\"Memory tier\",\"definition\":\"A layer in a hierarchy with different capacity, bandwidth and latency characteristics.\",\"anchor\":\"memory-tier\"},{\"term\":\"Heterogeneous compute\",\"definition\":\"Using different processor types or accelerator blocks for different workload classes.\",\"anchor\":\"heterogeneous-compute\"},{\"term\":\"Spark-like gaming architecture\",\"definition\":\"Figure Rocks shorthand for borrowing integrated CPU\u002FGPU\u002FAI and coherent-memory principles from systems such as DGX Spark; not an announced product.\",\"anchor\":\"spark-like\"}]},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":1032,"blocks":1033,"version":1397},1792340022841,[1034,1038,1043,1048,1052,1056,1085,1089,1109,1113,1117,1137,1172,1176,1193,1197,1202,1206,1210,1214,1218,1225,1229,1257,1261,1266,1286,1290,1335,1339,1343,1347,1354,1361,1381],{"id":541,"data":1035,"type":544,"tunes":1037},{"text":1036},"DGX Spark is built for AI development, not for replacing a GeForce gaming desktop. That is exactly why it is useful as a thought experiment. It packages CPU, GPU, Tensor and RT capability around a large coherent memory pool, making the entire system look less like a classic PC with two major memory islands and more like one heterogeneous compute machine.",{},{"id":547,"data":1039,"type":552,"tunes":1042},{"body":1040,"title":1041,"variant":551},"\u003Cstrong>A Spark-like gaming PC would change the data topology more than it changes the idea of a GPU.\u003C\u002Fstrong> The potential gain is a large shared pool that CPU, graphics and local AI workloads can access more coherently. The immediate cost is that shared capacity does not automatically provide discrete-GPU-class bandwidth, and tighter integration usually reduces modular upgradeability.","Direct answer",{},{"id":555,"data":1044,"type":552,"tunes":1047},{"body":1045,"title":1046,"variant":559},"DGX Spark specifications are factual. The gaming architecture proposed later is \u003Cstrong>not\u003C\u002Fstrong> an announced NVIDIA product and this article is not predicting a 'DGX Spark Gaming Edition.' Spark is evidence that a commercial Blackwell system can combine Arm CPU cores, GPU\u002FTensor\u002FRT resources and a large coherent memory pool. 'Spark-like' means borrowing those design principles for a gaming thought experiment, not predicting a specific SKU or roadmap.","This is an architectural thought experiment",{},{"id":562,"data":1049,"type":566,"tunes":1051},{"title":1050,"maxLevel":565,"minLevel":47},"Contents",{},{"id":569,"data":1053,"type":572,"tunes":1055},{"text":1054,"level":47},"What Spark actually is",{},{"id":575,"data":1057,"type":607,"tunes":1084},{"content":1058,"stretched":606,"withHeadings":15},[1059,1063,1066,1069,1073,1076,1080],[1060,1061,1062],"DGX Spark specification","Official value","Why it matters to this thought experiment",[1064,584,1065],"Architecture","CPU and Blackwell GPU are designed as one integrated compute platform",[587,1067,1068],"20-core Arm","Breaks from the traditional x86 gaming-desktop assumption",[1070,1071,1072],"System memory","64 GB or 128 GB coherent unified LPDDR5X","Large common pool rather than separate CPU RAM + dedicated VRAM",[1074,596,1075],"Memory bandwidth","Strong for an integrated system, but well below top-end discrete graphics memory bandwidth",[1077,1078,1079],"AI compute","Up to 1 PFLOP FP4","Shows how much design priority can move toward matrix\u002FTensor-style workloads",[1081,1082,1083],"RT hardware","4th-gen RT cores","The SoC still includes dedicated real-time graphics acceleration",{},{"id":610,"data":1086,"type":572,"tunes":1088},{"text":1087,"level":47},"The classic PC split creates duplication and movement",{},{"id":615,"data":1090,"type":635,"tunes":1108},{"steps":1091,"title":1107,"orientation":634},[1092,1095,1098,1101,1104],{"label":1093,"description":1094},"CPU-side system memory","Game logic, streaming buffers and many assets begin in system RAM.",{"label":1096,"description":1097},"PCIe\u002Fdata movement","Resources are copied or mapped toward the GPU as needed.",{"label":1099,"description":1100},"Discrete VRAM","The GPU keeps its working set in a separate high-bandwidth pool.",{"label":1102,"description":1103},"Rendering","Shaders, RT and compute consume those resident resources.",{"label":1105,"description":1106},"Eviction\u002Fstreaming","When the working set changes, data moves again.","Traditional discrete-gaming data path",{},{"id":638,"data":1110,"type":544,"tunes":1112},{"text":1111},"That separation is not inherently bad. It allows the GPU to have extremely fast dedicated memory and lets users upgrade components independently. But it also creates boundaries: capacity can be stranded on one side, data can be duplicated, and applications need explicit movement and residency policies.",{},{"id":643,"data":1114,"type":572,"tunes":1116},{"text":1115,"level":47},"A unified design changes the question",{},{"id":648,"data":1118,"type":635,"tunes":1136},{"steps":1119,"title":1135,"orientation":634},[1120,1123,1126,1129,1132],{"label":1121,"description":1122},"One coherent pool","CPU and GPU see a much larger common addressable memory resource.",{"label":1124,"description":1125},"Specialized engines","Graphics, RT and matrix\u002FTensor workloads pull from the same system-level pool.",{"label":1127,"description":1128},"Less mandatory duplication","Some workloads can avoid maintaining separate CPU and GPU copies.",{"label":1130,"description":1131},"New contention problem","All compute engines can now compete for the same memory bandwidth.",{"label":1133,"description":1134},"Software becomes more important","Scheduling and locality determine whether unified memory feels elegant or congested.","Spark-like data model",{},{"id":669,"data":1138,"type":710,"tunes":1171},{"rows":1139,"title":1165,"layout":607,"columns":1166},[1140,1145,1150,1155,1160],{"id":673,"label":1141,"values":1142},"Memory capacity",{"disc":1143,"unified":1144},"System RAM + separate VRAM pools","Large coherent shared pool",{"id":679,"label":1146,"values":1147},"Peak graphics bandwidth",{"disc":1148,"unified":1149},"Can be extremely high with GDDR\u002FHBM-style dedicated memory","Must be designed high enough for graphics + CPU + AI contention",{"id":685,"label":1151,"values":1152},"Upgradeability",{"disc":1153,"unified":1154},"GPU, RAM and sometimes CPU replaced independently","More platform-level replacement likely",{"id":691,"label":1156,"values":1157},"Local AI fit",{"disc":1158,"unified":1159},"AI competes for GPU VRAM or separate NPU","Large shared pool can make models easier to fit",{"id":697,"label":1161,"values":1162},"Data movement",{"disc":1163,"unified":1164},"Explicit transfers and duplication can matter","Potentially less copying, but locality still matters","Discrete gaming PC vs Spark-like principle",[1167,1169],{"id":705,"label":1168},"Discrete gaming PC",{"id":708,"label":1170},"Spark-like integrated system",{},{"id":713,"data":1173,"type":572,"tunes":1175},{"text":1174,"level":47},"Capacity and bandwidth tell radically different stories",{},{"id":718,"data":1177,"type":607,"tunes":1192},{"content":1178,"stretched":606,"withHeadings":15},[1179,1182,1185,1188],[1180,1141,1074,1181],"Reference architecture","Design lesson",[725,1183,596,1184],"128 GB coherent unified LPDDR5X","Huge shared capacity, but not discrete-high-end graphics bandwidth",[1186,730,731,1187],"RTX PRO 6000 Blackwell Workstation","Professional discrete GPU shows how much bandwidth high-end graphics\u002FAI workloads can demand",[734,1189,1190,1191],"32 GB GDDR7, 512-bit","High-end discrete design","Gaming flagship emphasizes bandwidth and dedicated local memory over giant coherent capacity",{},{"id":740,"data":1194,"type":544,"tunes":1196},{"text":1195},"The point is not that one architecture is universally better. Spark demonstrates coherent capacity; the RTX PRO 6000 demonstrates extreme local bandwidth. A gaming architecture that tries to combine Spark-like coherence with flagship-class graphics would need to reconcile those priorities rather than copy one product unchanged.",{},{"id":745,"data":1198,"type":552,"tunes":1201},{"body":1199,"title":1200,"variant":749},"Spark's \u003Cstrong>273 GB\u002Fs\u003C\u002Fstrong> is a useful reminder that memory capacity and memory bandwidth are different specifications. A gaming derivative could not simply copy 128 GB LPDDR5X and assume it replaces a high-end discrete GPU. It would need much more aggressive bandwidth, caching, compression or workload scheduling — possibly all four.","The bandwidth wall",{},{"id":752,"data":1203,"type":572,"tunes":1205},{"text":1204,"level":47},"Compression and prediction could make a lower-bandwidth unified pool more viable",{},{"id":757,"data":1207,"type":544,"tunes":1209},{"text":1208},"A future integrated design does not have to solve every problem by increasing pin speed. Compression can reduce bytes moved, neural representations can reduce stored detail, predictive prefetch can hide some access latency, and a cache hierarchy can keep hot graphics data near the GPU. Those techniques do not erase the bandwidth wall, but they can reduce how often software hits it.",{},{"id":762,"data":1211,"type":572,"tunes":1213},{"text":1212,"level":47},"Why neural rendering makes unified memory more interesting",{},{"id":767,"data":1215,"type":544,"tunes":1217},{"text":1216},"Classic graphics mostly asks whether the GPU can keep textures, geometry and buffers close enough to render the scene. Neural graphics can add model weights, latent representations, history buffers and intermediate tensors. A large coherent pool makes those new data classes easier to share between CPU-side game systems and GPU-side inference, especially when a local game assistant or NPC model is also running.",{},{"id":772,"data":1219,"type":778,"tunes":1224},{"url":1220,"title":1221,"excerpt":1222,"ctaLabel":1223},"https:\u002F\u002Ffigure.rocks\u002Fblog\u002Fdirectx-is-becoming-an-ml-platform-what-linear-algebra-and-neural-shaders-mean-for-future-games","DirectX Is Becoming an ML Platform: What Linear Algebra and Neural Shaders Mean for Future Games","The software side of this transition: machine-learning math is entering the graphics programming model itself.","Read the DirectX ML analysis",{},{"id":781,"data":1226,"type":572,"tunes":1228},{"text":1227,"level":47},"What would a gaming version need that Spark does not optimize for?",{},{"id":786,"data":1230,"type":607,"tunes":1256},{"content":1231,"stretched":606,"withHeadings":15},[1232,1236,1240,1244,1248,1252],[1233,1234,1235],"Gaming requirement","Why Spark-like integration helps","What still must be solved",[1237,1238,1239],"High-refresh graphics","Close CPU\u002FGPU coordination","Much higher sustained graphics bandwidth and low latency",[1241,1242,1243],"Large neural models","Shared capacity reduces fit\u002Fduplication problems","Inference cannot steal frame time unpredictably",[1245,1246,1247],"Asset streaming","One pool simplifies residency relationships","Storage-to-memory path and decompression must stay fast",[1249,1250,1251],"Competitive latency","Fewer transfers can help","Unified scheduling must not add contention or long queues",[1253,1254,1255],"Long product life","Integrated platform can be power efficient","Repair\u002Fupgrade path can become worse than a modular PC",{},{"id":815,"data":1258,"type":572,"tunes":1260},{"text":1259,"level":47},"The most interesting version is probably hybrid",{},{"id":820,"data":1262,"type":552,"tunes":1265},{"body":1263,"title":1264,"variant":824},"A future high-end gaming system may not choose between \u003Cstrong>fully discrete\u003C\u002Fstrong> and \u003Cstrong>fully unified\u003C\u002Fstrong>. A hybrid could combine a large coherent memory pool with a high-bandwidth local cache or graphics-memory tier, allowing AI and game-state workloads to share capacity while latency-sensitive rendering keeps a very fast local path.","Figure Rocks hypothesis",{},{"id":827,"data":1267,"type":635,"tunes":1285},{"steps":1268,"title":1284,"orientation":634},[1269,1272,1275,1278,1281],{"label":1270,"description":1271},"Large coherent pool","Holds game assets, AI models, CPU state and less latency-sensitive data.",{"label":1273,"description":1274},"Predictive residency manager","Moves likely-needed resources toward faster memory before demand.",{"label":1276,"description":1277},"High-bandwidth graphics tier","Protects the current render working set.",{"label":1279,"description":1280},"Neural compression","Keeps selected assets or representations smaller until needed.",{"label":1282,"description":1283},"Engine scheduler","Balances CPU, graphics, RT and neural workloads against frame deadlines.","A possible hybrid gaming memory hierarchy",{},{"id":848,"data":1287,"type":572,"tunes":1289},{"text":1288,"level":47},"What the buyer gains — and gives up",{},{"id":853,"data":1291,"type":710,"tunes":1334},{"rows":1292,"title":1328,"layout":607,"columns":1329},[1293,1298,1303,1308,1313,1318,1323],{"id":857,"label":1294,"values":1295},"Very large useful pool",{"cost":1296,"gain":1297},"Shared contention and less obvious performance specs","Fewer hard walls between RAM, VRAM and model memory",{"id":691,"label":1299,"values":1300},"Local AI headroom",{"cost":1301,"gain":1302},"AI workloads can compete with rendering","Bigger models can coexist with games",{"id":868,"label":1304,"values":1305},"Integration",{"cost":1306,"gain":1307},"Thermal limits may cap peak graphics performance","Potential efficiency and compactness",{"id":685,"label":1309,"values":1310},"Platform simplicity",{"cost":1311,"gain":1312},"Whole-platform upgrade instead of GPU-only upgrade","Fewer compatibility boundaries",{"id":879,"label":1314,"values":1315},"Optimization",{"cost":1316,"gain":1317},"Poor software support can waste expensive hardware","Engine can coordinate more of the system",{"id":885,"label":1319,"values":1320},"Predictive memory management",{"cost":1321,"gain":1322},"Wrong predictions can waste bandwidth, residency and cache","Fewer late asset\u002Fpage arrivals when prediction is accurate",{"id":891,"label":1324,"values":1325},"Platform-level optimization",{"cost":1326,"gain":1327},"Greater ecosystem lock-in and more opaque performance behavior","Engine, driver and hardware can coordinate data placement and accelerators more deeply","Buyer trade-off",[1330,1332],{"id":899,"label":1331},"Potential gain",{"id":902,"label":1333},"Potential cost",{},{"id":906,"data":1336,"type":572,"tunes":1338},{"text":1337,"level":47},"What would falsify the idea?",{},{"id":911,"data":1340,"type":544,"tunes":1342},{"text":1341},"If high-end discrete GPUs continue to deliver much better cost, bandwidth and latency while local gaming AI remains optional, there may be little reason to accept the lock-in and contention of a unified architecture. The hypothesis becomes stronger only if games increasingly need large shared data sets and neural workloads that are awkward to fit inside a traditional CPU-RAM + discrete-VRAM split.",{},{"id":916,"data":1344,"type":572,"tunes":1346},{"text":1345,"level":47},"Primary source",{},{"id":921,"data":1348,"type":928,"tunes":1353},{"link":923,"meta":1349},{"image":1350,"title":1351,"description":1352},{"url":13},"NVIDIA DGX Spark — Official specifications","Official GB10 architecture, unified memory, bandwidth, CPU, GPU and AI-compute specifications.",{},{"id":931,"data":1355,"type":928,"tunes":1360},{"link":933,"meta":1356},{"image":1357,"title":1358,"description":1359},{"url":13},"NVIDIA — RTX PRO 6000 Blackwell Workstation Edition","Official 96 GB GDDR7 and 1,792 GB\u002Fs bandwidth reference for a high-end discrete Blackwell GPU.",{},{"id":940,"data":1362,"type":964,"tunes":1380},{"items":1363,"title":1379},[1364,1367,1370,1373,1376],{"id":944,"answer":1365,"question":1366},"No. NVIDIA positions it as a personal AI supercomputer. This article uses its architecture as a thought experiment.","Is DGX Spark a gaming PC?",{"id":948,"answer":1368,"question":1369},"Not automatically. Capacity, bandwidth, latency and software optimization are different constraints.","Would 128 GB unified memory beat a 32 GB high-end GPU?",{"id":952,"answer":1371,"question":1372},"It can reduce hard capacity boundaries and duplication between CPU, GPU and local-model workloads.","Why is unified memory attractive for gaming AI?",{"id":956,"answer":1374,"question":1375},"Shared-memory contention plus reduced modular upgradeability.","What is the biggest risk?",{"id":960,"answer":1377,"question":1378},"Potentially. A large coherent pool plus a fast graphics-local tier could preserve bandwidth while giving AI and general workloads more shared capacity.","Could a hybrid be better?","Spark-like gaming architecture",{},{"id":967,"data":1382,"type":986,"tunes":1396},{"title":1383,"entries":1384},"Architecture terms",[1385,1388,1391,1394],{"term":1386,"anchor":973,"definition":1387},"Coherent unified memory","A memory architecture where multiple processors can access a shared system memory space with coherence guarantees.",{"term":1389,"anchor":977,"definition":1390},"Memory tier","A layer in a hierarchy with different capacity, bandwidth and latency characteristics.",{"term":1392,"anchor":981,"definition":1393},"Heterogeneous compute","Using different processor types or accelerator blocks for different workload classes.",{"term":1379,"anchor":984,"definition":1395},"Figure Rocks shorthand for borrowing integrated CPU\u002FGPU\u002FAI and coherent-memory principles from systems such as DGX Spark; not an announced product.",{},"2.31.6",{"lang":7,"title":534,"content":536,"contentJson":1399,"excerpt":14},{"time":538,"blocks":1400,"version":988},[1401,1404,1407,1410,1413,1416,1427,1430,1439,1442,1445,1454,1471,1474,1482,1485,1488,1491,1494,1497,1500,1503,1506,1516,1519,1522,1531,1534,1555,1558,1561,1564,1569,1574,1583],{"id":541,"data":1402,"type":544,"tunes":1403},{"text":543},{},{"id":547,"data":1405,"type":552,"tunes":1406},{"body":549,"title":550,"variant":551},{},{"id":555,"data":1408,"type":552,"tunes":1409},{"body":557,"title":558,"variant":559},{},{"id":562,"data":1411,"type":566,"tunes":1412},{"title":564,"maxLevel":565,"minLevel":47},{},{"id":569,"data":1414,"type":572,"tunes":1415},{"text":571,"level":47},{},{"id":575,"data":1417,"type":607,"tunes":1426},{"content":1418,"stretched":606,"withHeadings":15},[1419,1420,1421,1422,1423,1424,1425],[579,580,581],[583,584,585],[587,588,589],[591,592,593],[595,596,597],[599,600,601],[603,604,605],{},{"id":610,"data":1428,"type":572,"tunes":1429},{"text":612,"level":47},{},{"id":615,"data":1431,"type":635,"tunes":1438},{"steps":1432,"title":633,"orientation":634},[1433,1434,1435,1436,1437],{"label":619,"description":620},{"label":622,"description":623},{"label":625,"description":626},{"label":628,"description":629},{"label":631,"description":632},{},{"id":638,"data":1440,"type":544,"tunes":1441},{"text":640},{},{"id":643,"data":1443,"type":572,"tunes":1444},{"text":645,"level":47},{},{"id":648,"data":1446,"type":635,"tunes":1453},{"steps":1447,"title":666,"orientation":634},[1448,1449,1450,1451,1452],{"label":652,"description":653},{"label":655,"description":656},{"label":658,"description":659},{"label":661,"description":662},{"label":664,"description":665},{},{"id":669,"data":1455,"type":710,"tunes":1470},{"rows":1456,"title":702,"layout":607,"columns":1467},[1457,1459,1461,1463,1465],{"id":673,"label":674,"values":1458},{"disc":676,"unified":677},{"id":679,"label":680,"values":1460},{"disc":682,"unified":683},{"id":685,"label":686,"values":1462},{"disc":688,"unified":689},{"id":691,"label":692,"values":1464},{"disc":694,"unified":695},{"id":697,"label":698,"values":1466},{"disc":700,"unified":701},[1468,1469],{"id":705,"label":706},{"id":708,"label":709},{},{"id":713,"data":1472,"type":572,"tunes":1473},{"text":715,"level":47},{},{"id":718,"data":1475,"type":607,"tunes":1481},{"content":1476,"stretched":606,"withHeadings":15},[1477,1478,1479,1480],[722,674,595,723],[725,726,596,727],[729,730,731,732],[734,735,736,737],{},{"id":740,"data":1483,"type":544,"tunes":1484},{"text":742},{},{"id":745,"data":1486,"type":552,"tunes":1487},{"body":747,"title":748,"variant":749},{},{"id":752,"data":1489,"type":572,"tunes":1490},{"text":754,"level":47},{},{"id":757,"data":1492,"type":544,"tunes":1493},{"text":759},{},{"id":762,"data":1495,"type":572,"tunes":1496},{"text":764,"level":47},{},{"id":767,"data":1498,"type":544,"tunes":1499},{"text":769},{},{"id":772,"data":1501,"type":778,"tunes":1502},{"url":774,"title":775,"excerpt":776,"ctaLabel":777},{},{"id":781,"data":1504,"type":572,"tunes":1505},{"text":783,"level":47},{},{"id":786,"data":1507,"type":607,"tunes":1515},{"content":1508,"stretched":606,"withHeadings":15},[1509,1510,1511,1512,1513,1514],[790,791,792],[794,795,796],[798,799,800],[802,803,804],[806,807,808],[810,811,812],{},{"id":815,"data":1517,"type":572,"tunes":1518},{"text":817,"level":47},{},{"id":820,"data":1520,"type":552,"tunes":1521},{"body":822,"title":823,"variant":824},{},{"id":827,"data":1523,"type":635,"tunes":1530},{"steps":1524,"title":845,"orientation":634},[1525,1526,1527,1528,1529],{"label":831,"description":832},{"label":834,"description":835},{"label":837,"description":838},{"label":840,"description":841},{"label":843,"description":844},{},{"id":848,"data":1532,"type":572,"tunes":1533},{"text":850,"level":47},{},{"id":853,"data":1535,"type":710,"tunes":1554},{"rows":1536,"title":896,"layout":607,"columns":1551},[1537,1539,1541,1543,1545,1547,1549],{"id":857,"label":858,"values":1538},{"cost":860,"gain":861},{"id":691,"label":863,"values":1540},{"cost":865,"gain":866},{"id":868,"label":869,"values":1542},{"cost":871,"gain":872},{"id":685,"label":874,"values":1544},{"cost":876,"gain":877},{"id":879,"label":880,"values":1546},{"cost":882,"gain":883},{"id":885,"label":886,"values":1548},{"cost":888,"gain":889},{"id":891,"label":892,"values":1550},{"cost":894,"gain":895},[1552,1553],{"id":899,"label":900},{"id":902,"label":903},{},{"id":906,"data":1556,"type":572,"tunes":1557},{"text":908,"level":47},{},{"id":911,"data":1559,"type":544,"tunes":1560},{"text":913},{},{"id":916,"data":1562,"type":572,"tunes":1563},{"text":918,"level":47},{},{"id":921,"data":1565,"type":928,"tunes":1568},{"link":923,"meta":1566},{"image":1567,"title":926,"description":927},{"url":13},{},{"id":931,"data":1570,"type":928,"tunes":1573},{"link":933,"meta":1571},{"image":1572,"title":936,"description":937},{"url":13},{},{"id":940,"data":1575,"type":964,"tunes":1582},{"items":1576,"title":963},[1577,1578,1579,1580,1581],{"id":944,"answer":945,"question":946},{"id":948,"answer":949,"question":950},{"id":952,"answer":953,"question":954},{"id":956,"answer":957,"question":958},{"id":960,"answer":961,"question":962},{},{"id":967,"data":1584,"type":986,"tunes":1590},{"title":969,"entries":1585},[1586,1587,1588,1589],{"term":972,"anchor":973,"definition":974},{"term":976,"anchor":977,"definition":978},{"term":980,"anchor":981,"definition":982},{"term":963,"anchor":984,"definition":985},{},"Post erfolgreich abgerufen",{"items":1593,"source":1667,"manualIds":1668,"manualMatchedIds":1669},[1594,1600,1606,1613,1620,1627,1634,1641,1648,1655,1661],{"id":1595,"slug":1596,"title":1597,"excerpt":1598,"featuredImage":14,"publishedAt":1599},"185","hdr-tone-mapping-traps-when-dynamic-settings-destroy-detail","HDR色调映射陷阱：当“动态”设置破坏细节时","许多HDR问题实为色调映射问题。“动态”模式可能压暗高光或将黑色提亮成灰雾。以此基准保持细节清晰可辨。","2026-02-20T16:00:00.000Z",{"id":1601,"slug":1602,"title":1603,"excerpt":1604,"featuredImage":14,"publishedAt":1605},"402","ll-amiibo-compatible-games-complete-overview-of-bonuses-and-gameplay-integration","所有兼容amiibo的游戏——奖励与游戏玩法整合完整概览","amiibo手办可与Wii U、任天堂3DS及任天堂Switch平台上的众多任天堂游戏互动。不同游戏中的附加价值各有差异：有时奖励仅为外观装饰，有时则会直接影响游戏玩法。以下将按游戏平台分类，系统梳理兼容作品及扫描手办可解锁的核心内容。","2026-02-27T18:13:00.000Z",{"id":1607,"slug":1608,"title":1609,"excerpt":1610,"featuredImage":1611,"publishedAt":1612},"434","google-io-2026-gaming-ai-xr-and-the-next-wave-of-collectibles","Google I\u002FO 2026 与游戏：AI、XR 以及下一波收藏品浪潮","Google I\u002FO 2026并非一场典型的游戏活动，但其AI、XR、Gemini、Google Play和创意工具方面的公告指向了游戏文化的一次明显转变。本文解释了AI发现、XR伴侣层、Gemini驱动的媒体以及更快的角色可见性可能如何塑造游戏、手办、收藏品和未来的粉丝生态系统。","\u002Fuploads\u002F2026\u002F05\u002Fgoogle-io-2026-gaming-ai-xr-and-the-next-wave-of-collectibles-1779257974958-jjj07f.webp","2026-05-20T00:17:00.000Z",{"id":1614,"slug":1615,"title":1616,"excerpt":1617,"featuredImage":1618,"publishedAt":1619},"455","what-if-a-game-could-predict-its-next-memory-access-ai-prefetching-and-the-future-of-stutter","如果一款游戏能预测它的下一次内存访问会怎样？AI预取与卡顿的未来","游戏引擎已经知道玩家、摄像机和世界将去向何处。研究表明，学习式GPU预取能够预测内存页。如果将这些想法结合起来会发生什么？","\u002Fuploads\u002F2026\u002F10\u002Fwhat-if-a-game-could-predict-its-next-memory-access-ai-prefetching-and-the-future-of-stutter-1792339100514-lyx0yw.webp","2026-10-18T11:57:00.000Z",{"id":1621,"slug":1622,"title":1623,"excerpt":1624,"featuredImage":1625,"publishedAt":1626},"457","what-does-future-proof-gaming-hardware-mean-in-the-neural-rendering-era","在神经渲染时代，“面向未来”的游戏硬件意味着什么？","过去，面向未来意味着购买比当下所需更多的 CPU、GPU 和 VRAM。神经渲染让这条规则变得不那么可靠。这里有一个更好的决策框架。","\u002Fuploads\u002F2026\u002F10\u002Fwhat-does-future-proof-gaming-hardware-mean-in-the-neural-rendering-era-1792340303825-012uhb.webp","2026-10-18T12:16:00.000Z",{"id":1628,"slug":1629,"title":1630,"excerpt":1631,"featuredImage":1632,"publishedAt":1633},"458","the-game-engine-is-becoming-part-renderer-part-scheduler-part-ai-system","游戏引擎正逐渐演变为部分渲染器、部分调度器、部分AI系统","游戏引擎过去用于编排模拟与渲染。神经图形、预测性内存和游戏智能体正在增加第三项职责：决定推理在帧和游戏状态中的位置。","\u002Fuploads\u002F2026\u002F10\u002Fthe-game-engine-is-becoming-part-renderer-part-scheduler-part-ai-system-1792340569647-63jv23.webp","2026-10-18T12:21:00.000Z",{"id":1635,"slug":1636,"title":1637,"excerpt":1638,"featuredImage":1639,"publishedAt":1640},"433","pre-order-alert-good-smile-company-figma-doom-the-dark-ages-doom-slayer-dx-edition","Pre-Order Alert: Good Smile Company Figma Doom: The Dark Ages – Doom Slayer DX Edition","The new Good Smile Company Figma Doom: The Dark Ages Doom Slayer DX Edition is more than a routine figure drop. It connects collector demand, franchise identity, and the wider appeal of Doom as one of gaming’s most durable icons.","\u002Fuploads\u002F2026\u002F05\u002Fpre-order-alert-good-smile-company-figma-doom-the-dark-ages-doom-slayer-dx-edition-1779227794098-251pwj.webp","2026-04-22T04:37:00.000Z",{"id":1642,"slug":1643,"title":1644,"excerpt":1645,"featuredImage":1646,"publishedAt":1647},"453","render-less-infer-more-is-ai-breaking-the-traditional-gaming-hardware-upgrade-cycle","渲染更少，推理更多：AI正在打破传统游戏硬件升级周期吗？","人工智能正使高性能芯片和存储器在游戏之外更具价值，同时也为游戏提供了新的渲染、压缩和预测方式。本分析探讨这两种趋势是否会汇聚成一种新的硬件模式。","\u002Fuploads\u002F2026\u002F10\u002Frender-less-infer-more-is-ai-breaking-the-traditional-gaming-hardware-upgrade-cycle-1792323493963-l2x8qb.webp","2026-10-18T03:34:00.000Z",{"id":1649,"slug":1650,"title":1651,"excerpt":1652,"featuredImage":1653,"publishedAt":1654},"454","why-ai-is-reshaping-gaming-memory-hbm-gddr-vram-and-the-fight-for-capacity","为什么AI正在重塑游戏内存：HBM、GDDR、VRAM与容量之争","现代游戏背后的内存问题不再仅仅是GPU需要多少千兆字节。AI基础设施正在改变供应商优先考虑哪些内存产品，以及消费者买家能负担多少。","\u002Fuploads\u002F2026\u002F10\u002Fwhy-ai-is-reshaping-gaming-memory-hbm-gddr-vram-and-the-fight-for-capacity-1792335495408-ymehaw.webp","2026-10-18T10:56:00.000Z",{"id":1656,"slug":1657,"title":1658,"excerpt":1659,"featuredImage":14,"publishedAt":1660},"399","beyond-scanning-creative-uses-and-modifications-of-amiibo","\"超越扫描——amiibo的创意用途与改造\"","amiibo是任天堂自2014年起推出的搭载NFC技术的角色手办。从技术层面看，它们是封装在塑胶模型内的微型数据载体。在实际应用中，这些产品介于玩具、收藏品与交互设备之间。将其扫描接入游戏主机仅是它们生命周期中的一环。多年观察可见，许多拥有者将其视为具备独立创作潜能的实体物件。","2026-02-27T16:36:00.000Z",{"id":1662,"slug":1663,"title":1664,"excerpt":1665,"featuredImage":14,"publishedAt":1666},"398","upcoming-nintendo-amiibo-new-figures-release-windows-and-functional-significance","即将推出的任天堂amiibo——新模型、发售窗口与功能意义","amiibo是配备NFC芯片的实体手办，可与任天堂游戏机进行通信。在兼容游戏中，它们能解锁游戏内物品、装饰元素或小型游戏功能。本目录概览主要关注官方已公布的即将推出的amiibo，包括其预计发布时间窗口、功能价值以及在更广泛的产品批次中的定位。","2026-02-27T15:54:00.000Z","fallback",[],[]]