CUDA Goes Wild on AMD! Official Release of “ZLUDA 6” Breaks the GPU Barrier with GPU Support for PhysX & Blender!

What Happened? Overview of the News

• ZLUDA Version 6 Released: A major update has arrived for the middleware that allows CUDA applications to run unmodified on non-NVIDIA GPUs (primarily AMD).
• Expanded Key Workloads: The pre-release version of 32-bit PhysX and Blender (with texture support) are now operational. Old games with PhysX support now run smoothly on AMD GPUs.
• Change in Development Structure: The commercial funding has ended, transitioning the project to its developer vosen’s “weekend project.” The focus has shifted from profit to prioritizing features that “developers enjoy.”

Why Is This Important? Key Points to Note

• Liberation of Legacy Assets: With PhysX support, AMD GPU users can fully enjoy physical effects (like fire and debris) in classic games.
• Steady Improvements in ML Support: Based on user feedback regarding PyTorch traces, numerous compiler bugs have been fixed, and compatibility with performance libraries (cuBLAS/cuDNN, etc.) has been enhanced.
• Improved Usability in Windows Environments: In the complex world of Windows, the introduction of ROCm has been made more user-friendly with enhanced notifications for missing libraries and automatic loading features.

🦈 Shark’s Eye (Curator’s Perspective)

The magic of running CUDA unmodified has finally reached Version 6! What’s thrilling about this update is the shift from “commercial” to “hobby” project, fully dedicating itself to our playground, like PhysX! Watching Mafia II’s PhysX benchmark run flawlessly on AMD GPUs is simply breathtaking. The implementation of Blender’s texture support is a game-changer, making it clear that practicality is beginning to challenge NVIDIA’s stronghold. The improved compatibility with PyTorch is also incredibly exciting, thanks to a relentless approach of analyzing user execution logs (traces) for implementation!

What’s Next?

With development now primarily on weekends, the update frequency may decrease, but we can expect more eccentric and challenging features to be added. Particularly, full support for PhysX and further optimizations for ML frameworks are anticipated to progress under community leadership.

A Word from Haru Shark

“Seeing CUDA run on AMD is as exhilarating as a shark flying through the sky! 🦈🔥 Let’s smash the GPU barriers and enjoy the latest AI along with classic games on our favorite hardware!”

Terminology Explained

• CUDA: A parallel computing platform for GPUs provided by NVIDIA. It typically only runs on NVIDIA’s own GPUs.

• PhysX: A technology that processes physics calculations in games (like explosions and cloth movements). Some older titles depend on CUDA.

• ROCm: An open-source GPU computing platform provided by AMD. ZLUDA uses this to convert CUDA instructions.

• Source: Zluda 6 release (run unmodified CUDA applications on non-Nvidia GPUs)