DirectStorage 1.4 Does Not Make Your SSD Decompress Games: What Zstd and GPU Decompression Actually Do

DirectStorage 1.4 adds Zstandard compression and GPU decompression support for game assets, but that does not mean your SSD is suddenly decompressing textures and models by itself. The storage drive moves compressed data. The CPU or GPU still performs the decompression work.
The simplest way to understand the asset path
What happens when a game loads compressed assets
Why compress game assets at all?
Modern games contain enormous amounts of texture, geometry, audio and other data.
Storing every asset uncompressed would increase installation size and force the storage device to read more bytes for the same content.
Compression reduces how much data must be stored and transferred. The trade-off is that the data has to be decompressed before the game can use it.
DirectStorage moves the bottleneck problem, it does not remove it
A fast NVMe drive can deliver compressed data much faster than older storage pipelines, but faster reads only help if the rest of the path can keep up.
If the CPU has to decompress thousands of small asset chunks while also running game logic, physics and draw submission, decompression itself can become expensive.
That is why DirectStorage supports GPU decompression: some of the work can move away from the CPU and onto massively parallel GPU compute.
What DirectStorage 1.4 adds
DirectStorage 1.4 adds Zstandard, usually written Zstd, as a supported compression format.
Microsoft chose Zstd because it combines strong compression ratios, good decompression performance, broad software and hardware availability, and widespread adoption.
The important change is that Zstd is integrated into DirectStorage's decompression framework with both CPU and GPU paths.
CPU vs GPU decompression
| CPU decompression | GPU decompression | |
|---|---|---|
| Where it runs | ||
| Potential advantage | ||
| Main cost | ||
| Best choice |
GPU decompression still costs GPU resources
GPU decompression is not free.
DirectStorage uses staging buffers in VRAM to coordinate compressed input and decompressed output. Microsoft warns that an oversized staging buffer can take memory away from rendering, while one that is too small can reduce throughput because requests have to wait.
So GPU decompression creates another balancing problem: reduce CPU work without stealing too much GPU memory or compute from rendering.
Why Zstd matters compared with older DirectStorage compression
DirectStorage already supported GDeflate, a compression format designed for parallel GPU decompression.
Zstd adds a different option: a broadly used open compression standard with strong compression ratios and extensive tooling.
Microsoft is also publishing an open-source GPU Zstd decompression compute shader as a baseline that GPU vendors can optimize further.
The initial shader is optimized around chunks of 256 KB or smaller, which matches common streaming-oriented game packaging patterns.
The Asset Streaming Cost Triangle
Compression changes three different costs
| What compression can reduce | What it can increase | |
|---|---|---|
| Installation size | ||
| Storage I/O | ||
| Runtime CPU/GPU work |
What the Game Asset Conditioning Library does
Microsoft introduced the Game Asset Conditioning Library, or GACL, alongside DirectStorage 1.4.
GACL does not replace Zstd. It prepares certain game assets so Zstd can compress them more effectively.
Texture data is a major target because textures often make up a large share of modern game-package size.
What “conditioning” means in plain English
Conditioning means rearranging or slightly modifying the data before compression so the compressor can find more useful patterns.
For block-compressed textures, GACL can shuffle the byte layout before Zstd compression. DirectStorage reverses that transform after decompression.
GACL also includes entropy-reduction techniques that can deliberately simplify some texture information to make it compress better.
Where machine learning enters GACL
Microsoft's Component-Level Entropy Reduction, or CLER, uses machine learning as part of the process for reducing texture entropy.
The purpose is not to generate textures at runtime. The ML work is part of preparing texture data so the final compressed representation can become smaller while keeping acceptable visual quality.
That makes GACL a content-pipeline optimization rather than an in-game generative AI system.
Up to 50% better compression does not mean every game becomes 50% smaller
Microsoft says GACL can deliver up to a 50% improvement in Zstd compression ratios for suitable assets.
That is not the same as saying an entire game installation shrinks by 50%.
The real impact depends on which assets are conditioned, how compressible they are, what proportion of the game they represent and whether lossy techniques are acceptable.
The Compression Claim Test
How to read a game-compression claim correctly
Why replacing DirectStorage DLLs in a game does not enable Zstd
This is an especially important misunderstanding.
Microsoft explicitly answered this after the DirectStorage 1.4 announcement: replacing DirectStorage DLL files in an existing game does not activate the new preview features.
The game's assets have to be authored and compressed with Zstd during the build process, optional GACL conditioning has to happen before shipping, and the game itself has to call the relevant DirectStorage APIs.
Why faster SSDs do not remove decompression cost
An extremely fast SSD can reduce the time needed to read compressed bytes, but it does not eliminate the time required to turn those bytes into usable assets.
As storage becomes faster, the relative importance of decompression, asset preparation, GPU upload and engine-side processing becomes larger.
That is exactly why DirectStorage has evolved beyond raw I/O.
The Load-Time Pipeline Test
How to diagnose whether storage is really the bottleneck
Why this matters for open-world streaming
Open-world games continuously move asset data as the player crosses the map.
That means asset streaming is not only about reducing a loading screen. The pipeline has to deliver, decompress and prepare data while gameplay continues.
A codec with a strong compression ratio can reduce the number of bytes pulled from storage, while fast GPU decompression can help keep CPU time available for simulation.
But if the GPU itself is already saturated, moving decompression work there may need careful scheduling.
DirectStorage 1.4 also improves GPU scheduling visibility
DirectStorage 1.4 adds D3D12 CreatorID support for the internal command queues it manages.
The purpose is to help DirectStorage workloads participate more predictably in D3D12 queue grouping and GPU execution scheduling.
This matters because decompression is now competing for the same GPU ecosystem as rendering and other compute work.
What GPU vendors are doing next
Microsoft says AMD, Intel, NVIDIA and Qualcomm are working on hardware- and driver-specific Zstd optimizations.
AMD and NVIDIA publicly indicated optimized support targeted for the second half of 2026, while Intel and Qualcomm also described ongoing platform work.
That means the public DirectStorage GPU shader is a baseline, not necessarily the final performance path for future hardware.
What would change this answer?
DirectStorage 1.4 is still a public preview. The final API, Zstd shader performance, GACL support and vendor-optimized driver paths can all change.
If future storage devices add standardized hardware decompression integrated into the PC asset pipeline, the clean separation between “SSD reads” and “CPU/GPU decompresses” could become less absolute. That is not the normal DirectStorage 1.4 model today.
Limitations
This article explains Microsoft's current public DirectStorage 1.4 preview architecture. It does not claim that every game using DirectStorage will use Zstd, GPU decompression or GACL.
Microsoft's compression-ratio and performance statements are technology claims from the platform vendor. Real game results depend on the asset mix, compression settings, hardware, drivers and engine integration.
Conclusion
DirectStorage 1.4 is easier to understand once the jobs are separated.
The SSD moves compressed bytes. DirectStorage schedules the transfer. Zstd defines how the data is compressed. The CPU or GPU decompresses it. GACL can prepare some assets so Zstd compresses them more effectively. The game engine then consumes the finished asset.
A faster SSD is only one part of that chain.
FAQ
DirectStorage 1.4 and Zstd in plain English
Does DirectStorage make the SSD decompress game assets?
What is new in DirectStorage 1.4?
Can I replace a game's DirectStorage DLL to enable Zstd?
Is GPU decompression free?
What does GACL do?
Does up to 50% better compression mean a game becomes 50% smaller?
Glossary
Key DirectStorage 1.4 terms
- DirectStorage
- Microsoft's DirectX-family API for high-throughput game-asset I/O and decompression on Windows.
- Zstandard (Zstd)
- An open compression format added to DirectStorage 1.4 for game assets.
- GPU decompression
- Expanding compressed data using GPU compute instead of doing all decompression on the CPU.
- Staging buffer
- Temporary GPU memory used by DirectStorage to coordinate compressed and decompressed data flows.
- GACL
- Microsoft's Game Asset Conditioning Library, which prepares assets to improve compression efficiency.
- Asset Streaming Cost Triangle
- A Figure Rocks model separating storage size, I/O traffic and runtime decompression cost.
- Compression Claim Test
- A Figure Rocks checklist for interpreting compression claims by checking the asset scope, baseline, quality loss and runtime cost.
Primary sources
Microsoft DirectX — DirectStorage 1.4 Adds ZstandardOfficial March 2026 announcement covering Zstd, CPU/GPU decompression, GACL, GPU-vendor optimization plans and the DirectStorage 1.4 preview.
Microsoft DirectX — DirectStorage SDK & APIOfficial release page listing DirectStorage 1.4 preview features and package status.
Microsoft — DirectStorage GitHubOfficial repository containing samples, GPU decompression benchmarks and DirectStorage implementation guidance.
Microsoft — DirectStorage Developer GuidanceOfficial guidance explaining compressed-data flow, GPU decompression, staging buffers and memory trade-offs.
Microsoft — Game Asset Conditioning LibraryOfficial documentation covering texture conditioning, Zstd, shuffle transforms and entropy-reduction techniques.
Related Articles

VRAM Usage Is Not VRAM Requirement: Why a Full Memory Meter Does Not Tell the Whole Story
Seeing 7.8 GB used on an 8 GB graphics card can look like proof that a game has run out of VRAM. It is not that simple. This guide explains VRAM capacity, residency budgets, working sets, shared memory and how to tell whether memory pressure is actually causing stutter.
Storage and Streaming Stutter: When SSD, Decompression, and VRAM Matter
Streaming stutter is asset loading: storage speed, decompression, and memory behavior. Use this triage to stop spikes when entering new areas.
Storage and Streaming: Reduce Load Times Without Creating Stutter
Fast storage helps only when streaming behavior is stable. This guide explains how IO affects stutter and what to change first.

RTX Neural Texture Compression Is Not Upscaling: How AI Can Trade Texture Memory for GPU Compute
NVIDIA RTX Neural Texture Compression changes how game materials can be stored. Instead of keeping every texture channel only as conventional texels, a material can be compressed into compact latent data and a small neural decoder, then reconstructed by the GPU when needed.

Lowered Graphics Settings but FPS Didn't Improve? You're Probably Tuning the Wrong Bottleneck
You lower shadows, effects and resolution, but the FPS barely changes. This guide explains why graphics settings only help when they reduce the workload that is actually limiting the frame—and how to identify CPU, GPU, memory, streaming and frame-cap bottlenecks.
Storage Streaming Stutter Fixes: When Assets Can’t Keep Up
Streaming stutter happens when new areas load: storage, decompression, or asset streaming limits. Use this fix order before you drop every graphics setting.