mrkeyoor.comAsk

Speedstu/CUDA-for-AMD-Windowstested 03 Oct 2026commit 4a6fb1b

CUDA-for-AMD-Windows review

Use CUDA-for-AMD-Windows for a bounded compatibility trial on one Windows application and AMD GPU.

  • Build 8 s
20 sto install
  • 35 packages
Dev Toolsevaluationupdated 01 Nov 2026
+2stars / 7d

Use CUDA-for-AMD-Windows to qualify a specific CUDA-targeted Windows application on an AMD GPU when you can check its output against a reference. CUDA-for-AMD-Windows provides workload-specific CUDA compatibility, not a complete CUDA implementation. It combines ZLUDA, AMD HIP components, PowerShell installation and validation scripts. Our run on October 3, 2026, at commit 4a6fb1b recorded successful install and build steps in a Debian sandbox; it did not test Windows execution or an AMD GPU workload.

Verdict

Use CUDA-for-AMD-Windows for a bounded compatibility trial on one Windows application and AMD GPU. Our run on October 3, 2026, at commit 4a6fb1b recorded a 20-second install and an 8-second build step; tests were skipped, and no Windows GPU workload was tested. The project's hardware reports justify investigating specific paths, but your decision should depend on matching versions, correct application output and confirmed GPU execution.

We ran it

Lab card: what happened when we ran CUDA-for-AMD-Windows
Lab card from our sandbox run
Screenshot of CUDA-for-AMD-Windows (github.com/Speedstu/CUDA-for-AMD-Windows)
github.com/Speedstu/CUDA-for-AMD-Windows
Install✓ · 20s35 packages · 37 MB
Build✓ · 8s
Testsn/ano test script
Known vulns0(pip-audit)
Repo81 files~6,474 lines of source · 0.7 MB · 4 CI workflows

Answers from our run

Does CUDA-for-AMD-Windows build from source?

Dependencies installed in 20 seconds (35 packages), and the build succeeded in 8 seconds. We cloned commit 4a6fb1b into a clean Debian container with 3 CPUs and no project-specific setup.

Does CUDA-for-AMD-Windows have tests you can run?

Not through a standard command: the project exposes no test script or target that our harness could run.

Does CUDA-for-AMD-Windows have known vulnerabilities in its dependencies?

pip-audit found none in the dependency tree at the time of our run.

Who should not use CUDA-for-AMD-Windows?

Users expecting arbitrary CUDA applications to work after installation.

What are the alternatives to CUDA-for-AMD-Windows?

ZLUDA, TheRock, HIPIFY. Use CUDA-for-AMD-Windows for a bounded compatibility trial on one Windows application and AMD GPU.

Setup2/5Rubric: usable with substantial constraints. Windows, an appropriate AMD driver, architecture-specific HIP and runtime staging remain necessary; our Debian command timings do not measure that setup.
Docs5/5Rubric: unusually complete evidence for this dimension. Supplied documentation distinguishes evidence levels, exact reference versions, numerical failures, candidate status and corrections to earlier diagnoses.
Community3/5Rubric: moderate support evidence. The recorded snapshot lists 307 stars and 3 open issues, while supplied primary documentation includes detailed contributor validation. Adoption and support response times are unknown.
Maturity2/5Rubric: usable with substantial constraints. The recorded page reports no project release, and current primary documentation retains workload-specific support and separate experimental candidates.

Discussed on

  1. hnCUDA for AMD on Windows175 points

Who it’s for

Windows developers evaluating a specific CUDA application on an AMD GPU.
Researchers who can compare numerical results against a CPU or native implementation.
Maintainers who can record the exact GPU, driver, HIP, ZLUDA and application versions when reporting compatibility.
Users whose workloads resemble the project's documented reference paths and who can tolerate a failed qualification attempt.

Who it’s NOT for

Users expecting arbitrary CUDA applications to work after installation.
Teams treating device detection, library loading or a successful process exit as proof of correct GPU computation.
Workloads that cannot tolerate numerical errors, process failures or Windows crashes during evaluation.
Developers seeking a complete stable cuDNN replacement or guaranteed support for NVIDIA-specific kernels.
Users seeking a native HIP application stack rather than a CUDA compatibility layer.

Setup reality

Our run on October 3, 2026, used commit 4a6fb1b in the supplied Python 3.12 Debian Bookworm sandbox with 3 CPUs and 8 GB RAM. The stored harness summary records installation at 20 seconds and the build step at 8 seconds, both with exit code 0. It records 35 installed packages occupying 37 MB.

The harness skipped tests with the reason "no script". That describes its test discovery, not the absence of validation scripts in the repository. The build log tail is empty, so the recorded success does not establish that Windows components were compiled or exercised.

Real setup requires Windows x64, an AMD driver, the architecture-appropriate HIP SDK and staged ZLUDA assets. Numerical checks also require a compatible Python/PyTorch environment. The stable documented reference uses RX 9060 XT, ZLUDA v6-preview.69, HIP SDK 6.4 and LibTorch 2.3.0+cu118. Current source documentation and our older sandbox run describe different evidence and should be read separately.

Should you try it for your application?

Choose this project when you need to evaluate an existing CUDA-targeted Windows application on AMD hardware and can verify the results. CUDA-for-AMD-Windows provides workload-specific CUDA compatibility, not a complete CUDA implementation. A detected GPU is only the start: the useful result is a workload that executes on that GPU, produces correct output and exits cleanly.

The project's stable reference is the Radeon RX 9060 XT with ZLUDA v6-preview.69, HIP SDK 6.4 and LibTorch 2.3.0+cu118. Its compatibility documentation lists RX 9070 XT as externally validated, RX 7900 XT and Radeon 890M as community partial, and other recognized GPUs as unverified candidates. Those labels do not promise support for every application on a listed GPU.

What our sandbox established

Our run on October 3, 2026, used commit 4a6fb1b. Our sandbox recorded installation at 20 seconds and the build step at 8 seconds. The attached harness values round the raw durations of 19,900 ms and 7,527 ms. Both steps returned exit code 0 without a timeout; the test step was skipped because the harness found "no script".

These are stored command outcomes from a Python 3.12 Debian Bookworm container, not a Windows compatibility test. The harness records status, elapsed time, exit code, timeout state and a log tail for each command step. Here, the install tail identifies the Python environment and the build tail is empty. Neither establishes that PowerShell installation, AMD drivers, HIP, ZLUDA or an application ran successfully.

We did not test GPU throughput, numerical correctness, Windows launch behavior or application compatibility. The repository contains Windows validation scripts even though this harness did not discover a runnable test target. Skipped tests are not failed tests, and there is no passing-test total from our run.

The stored pip-audit result reports zero advisories. Scan coverage is unknown because the supplied record has no scannedPackages field; its packages value of zero does not establish coverage of the 35 installed packages. This is an advisory record, not a security assessment.

The MrKeyoor testing methodology explains the publisher's approach and limits. The supplied methodology page is newer than this run; linking it does not establish that its current procedure governed the October 3 execution.

What the project tested on hardware

The project's validation record documents a September 13, 2026, Windows RX 9060 XT run using the stable reference stack. It reports CUDA-facing library checks and a LibTorch PPO workload that completed a learning/update iteration. Those are project-reported hardware results, separate from our sandbox observations.

That evidence answers a narrower question than whether a complete training stack will work. The same document records a later reference-machine probe where math attention passed but memory-efficient attention returned incorrect output. A successful PPO update therefore does not establish convolution or fused-attention correctness.

Current documentation describes additional fixes and candidate builds developed after our checkout. Their results must retain their own hardware, version and workload scope; they do not extend our October 3 sandbox result.

Check the launch path before trusting output

The README says the installer detects the GPU, checks HIP, downloads pinned assets, stages the runtime and runs smoke checks. It performs numerical checks when a compatible Python/PyTorch environment is available. Install completion alone does not establish that those optional numerical checks ran.

Use the documented launcher and diagnostics for the selected stack. The current README says the PyTorch launcher defaults to math-only scaled dot-product attention on stable v6 and latest v7. It describes this as a workaround for finite but incorrect fused-attention results. Running the upstream executable directly bypasses that Python safety hook.

The current compatibility guide also corrects the interpretation of an older Radeon 890M convolution timeout: a missing MIOpen DLL triggered a Windows modal dialog before a CUDA kernel launched. The no-cuDNN fallback has separate passing evidence. This does not establish that the cuDNN compute path works on that machine.

Where compatibility ends

The stable Windows HIP path does not provide a complete cuDNN equivalent. Experimental bridges cover declared subsets on specified configurations. The current documentation reports bounded direct-HIP/MIOpen GPU checks for additional cuDNN v8 operations on RX 9060 XT, while explicitly distinguishing them from CUDA-targeted PyTorch application validation.

NCCL, TensorRT, unsupported PTX behavior, custom CUDA extensions and architecture-specific kernels may fail. For your application, check the operations it actually needs rather than inferring support from a library name or a simpler passing probe.

A practical qualification sequence is to record the exact stack, run the relevant numerical probes, compare application output against a trusted reference, confirm GPU execution and check clean shutdown. An explicit unsupported result is a reason to stop that path, not to accept plausible output from a fallback.

Choose the layer you need

ZLUDA is the underlying CUDA compatibility layer. Evaluate it directly if you want to assemble and validate the surrounding Windows stack yourself. CUDA-for-AMD-Windows adds installation, diagnostics and workload-specific evidence around that layer. The existing ZLUDA review addresses the underlying runtime rather than this Windows qualification workflow.

TheRock is a build and release system for HIP and ROCm. It can supply components used by a compatibility stack, or serve a native HIP/ROCm application workflow. It is not itself a replacement for CUDA translation. Its documentation recommends prebuilt releases for supported configurations and warns that source builds require substantial disk space and compilation time.

HIPIFY translates CUDA source code into portable HIP C++. Consider it when you control the source and can maintain a port. It addresses source migration rather than running an unchanged CUDA binary.

How to read the ratings

The retained ratings use a five-point editorial rubric: 1 means substantial documented obstacles, 2 means usable with substantial constraints, 3 means mixed evidence or moderate support, 4 means strong evidence with limited gaps, and 5 means unusually complete evidence for that dimension. They are judgments, not benchmark results.

Setup scores 2 because the Windows hardware and runtime prerequisites outweigh the short sandbox command times. Documentation scores 5 for explicit support levels, versioned validation, known failures and corrections to earlier diagnoses. Community scores 3 on the recorded snapshot of 307 stars and 3 open issues, together with the supplied contributor reports; that does not measure adoption or response times. Maturity scores 2 because the recorded page had no project release and current documentation still separates stable paths from experimental candidates.

Alternatives

ProjectWhat it isPick it when
ZLUDA gh↗The underlying compatibility layer for running CUDA applications on non-NVIDIA GPUs.Choose it directly when you want to assemble the surrounding runtime and validation workflow yourself; this repository packages that work for Windows.
TheRockA build and release system for HIP and ROCm components.Choose it for a native HIP/ROCm workflow or to obtain supporting components. It can complement this stack but does not replace CUDA translation.
HIPIFYTools that translate CUDA source into portable HIP C++.Choose it when you control the CUDA source and can maintain a HIP port rather than needing to run an unchanged binary.

What people are saying

  1. [velocity-scout] Speedstu/CUDA-for-AMD-Windows
  2. [hackernews] CUDA for AMD on Windows

Sources

  1. CUDA for AMD on Windows README
  2. Compatibility matrix
  3. Validation record
  4. Architecture guide
  5. RX 7900 XT compatibility report
  6. Radeon 890M compatibility report

More dev tools reviews