Zluda CUDA emulator loses funding again; back to hobby project with PhysX gains

<cite index="41-2">Zluda v6 released today with exciting new features but the unfortunate news that commercial funding has been cut from the project</cite>, marking the third cycle of boom-and-bust for the open-source CUDA translation layer. <cite index="42-5">The project was initially started in 2020 to get CUDA running on Intel hardware, but has since turned to AMD cards. After being abandoned in 2021, it was brought back around 2022 thanks to AMD funding. Unfortunately, AMD cut the funding to Zluda in 2024, and in August forced Janik to rebuild the code</cite>. An unnamed sponsor then picked it up in late 2024, only to withdraw support now. Lead developer Andrzej Janik says Zluda is now a weekend project again.

<cite index="42-4">Zluda 6's 32-bit PhysX support is still in pre-alpha stage, but the results are promising. Janik showed off multiple cloth and deformation demos running at speed, and even a screenshot showing a 3x performance uplift of 2010's Mafia II running with PhysX effects turned on</cite>. The release also includes PyTorch-driven compiler fixes and improved Windows support, pivoting from AI/ML workloads to gaming and retro applications now that commercial pressure is gone.

<cite index="41-2">ZLUDA began focusing on multi-GPU CUDA focused on AI after beginning to receive funding from an unspecified party, and that priorities have now shifted to things the developer finds most entertaining</cite>. For architects, this highlights the fragility of ecosystem-bridging projects without sustained corporate ownership. CUDA's lock-in persists: neither AMD (with ROCm/HIP) nor Intel (with oneAPI) have invested enough to make translation layers unnecessary, and without venture backing or captive infrastructure (like xAI at SpaceX), CUDA portability tools cycle between hope and hibernation.