The third-party solution gives AMD cards support for Nvidia's CUDA

Nvidia's CUDA technology is one of the main reasons for the company's dominant position. Many professional programs have been developed for CUDA and nothing else, although AMD has long had a competing technology called ROCM that is also open source. For the past two or three years, AMD has been quietly funding a project to develop technology that would allow CUDA code to run via ROCM without any special modifications, much like how Molten VK runs Vulkan code on Apple's bare metal framework, according to reports. Voronix.

AMD has now withdrawn, but the code has been released as open source. The lead developer of the project Andrzej Janik has ported the old ZLUDA tool from Intel to AMD with great care, and it is already possible to run many CUDA programs directly in Windows on AMD cards using ROCM.

Some features aren't fully functional yet, like Optix, and the CUDA code must use Nvidia's PTX Assembly language, but according to Phoronix's own tests, it actually works surprisingly well. The site was tested, among other things, using Blender.

While the development was an internal secret at AMD, the tool presented the graphics card for CUDA programs as a “graphics device” to avoid stumbling into logs or performance databases. Now that it's been released as open source, the developers will soon change the code to show exactly which Radeon card it's running on.