The ZLUDA open-source project for “CUDA on non-NVIDIA GPUs” continues being developed for enabling CUDA like GeekBench and AI workloads on AMD GPUs and other hardware vendors. The ZLUDA project hopes to have PyTorch up and running on it this year along with eyeing 32-bit PhysX support since NVIDIA has dropped support upstream for the 32-bit PhysX libraries with the recent RTX 50 Blackwell launch.
ZLUDA lead developer Andrzej Janik is out with a new blog post to summarize the development efforts during the first quarter of the year. PyTorch remains a top priority for ZLUDA development and there is a goal of getting it working during the second half of this year.
PhysX on the modern ZLUDA hasn’t been a big focus but since NVIDIA dropping support for 32-bit PhysX with the GeForce RTX 50 / Blackwell GPUs, that thinking changed slightly. There’s a belief that ZLUDA can “rescue” the 32-bit PhysX support to get it working for both AMD and NVIDIA GPUs. This isn’t a main priority though so ideally more open-source developers interested in PhysX will need to get involved.
ZLUDA has also been working on improving the LLVM unit tests and other improvements.
Those interested in the current state of ZLUDA can find the Q1’2025 progress report via the GitHub project site.
