AMD has added the Radeon RX 7900 XT (Test) with 20GB of graphics memory to the list of models compatible with PyTorch via ROCm Version 5.7. Alongside the AMD Radeon Pro W7900 and Radeon RX 7900 XTX, the current RDNA-3 high-end graphics cards are now compatible with the primary AI framework. However, comprehensive support for all GPUs doesn’t seem to be arriving anytime soon.
The manufacturer only began establishing compatibility between RDNA-3 GPUs and PyTorch a few weeks ago. Apparently, the graphics cards need to be individually integrated due to their varying specifications. It’s likely that long-term support for RDNA-3 is intended, even if the required work is ongoing.
ROCm 5.7 is AMD’s open-source initiative and software collection enabling GPU-accelerated AI applications on Ubuntu Version 22.04.3. AMD is increasingly promoting it as an alternative to pricier or less available competitor graphics cards.
Nvidia still holds a significant lead
Especially in contrast to Cuda, general support for operating systems, software, and graphics cards remains patchy. Alongside other GPUs using RDNA-3 architecture, older GPUs like the Radeon RX 6800 and RX 6900 XT are also absent. With their 16GB memory, they would also be suitable for AI applications.
With Nvidia, it’s common for practically all models to come with Cuda support, automatically qualifying them for all applications based on it. Ultimately, it only depends on whether the computational power and available memory are sufficient.
READ MORE: Google Maps’ New Design Sparks Criticism Over Visibility and Navigation Clarity