AMD Radeon RX 9060 XT Enables CUDA Workloads on Windows via ZLUDA and ROCm
A new reproducible stack lets developers run CUDA‑targeted Windows applications on AMD GPUs using ZLUDA and the AMD HIP/ROCm ecosystem. The project, released by Speedstu, validates the Radeon RX 9060 XT (gfx1200) with ZLUDA v6‑preview.69, HIP SDK 6.4, and LibTorch 2.3.0+cu118. A 2.2‑million‑parameter PPO network completed a full forward/inference cycle, learning, and optimizer step on the AMD…
Key points
- ZLUDA v6‑preview.69 + HIP SDK 6.4 lets CUDA apps run on AMD Radeon RX 9060 XT (gfx1200) in Windows
- LibTorch 2.3.0+cu118 and a 2.2‑M‑parameter PPO network validated on the stack, achieving 65,536 timesteps
- Upstream path is only 3 % slower than a custom overlay, achieving ~13,300 steps/s in benchmark
Performance tests show the public upstream path reaches a median of 13,278 steps per second, only 3.03 % slower than a custom overlay, making the upstream build the default. The stack supports core CUDA libraries such as cuBLAS, cuSPARSE, and cuFFT, but does not yet include cuDNN or MIOpen, limiting convolution‑heavy workloads.
The repository includes scripts for installation, staging, diagnostics, and GPU scanning, and it is MIT‑licensed. Users are encouraged to report compatibility results for other AMD GPUs, which are currently unverified.
The story so far
3 episodes →- AMD Radeon RX 9060 XT Enables CUDA Workloads on Windows via ZLUDA and ROCmthis story
CUDA for AMD on Windows
github.com · 13 September 2026Loading the full article…
This text was published by github.com. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- Hacker News discussion · 48 pointsnews.ycombinator.com
- Solo dev enables running CUDA on AMD hardware in Windows, getting multiple CUDA libraries running on a gaming Radeon RX 9060 XT GPU in Windows — CUDA-exclusive workloads on AMD hardware in Windows possible without virtualization or dual-bootingPress · Tom's Hardware ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Mindbeam announces collaboration with NVIDIA on Litespark Nemotron 3.5 Super VL model and inference engine · 4 src
- Arista Networks launches 7060XE7 Series for AI data center networking · 1 src
- NVIDIA adds cuPotential library to CUDA-X for AI chip simulations · 5 src
- Nvidia, AMD, Micron drop premarket as AI financing fears rise · 3 src
- Amazon’s Panay envisions AI wearables replacing apps for daily tasks · 1 src
Comments
via GitHub Discussions