Is it possible to combine in a tensor parallel mode 2 different radeons, but same rdna versions
7900 xtx and W7900
It could be possible that combining different Radeon GPUs of the same RDNA3 generation (e.g., 7900 XTX and W7900) for tensor parallelism in vLLM might work, but the data sources do not explicitly confirm support for mixed consumer/professional cards. The documentation and issues mention support for the Radeon RX 7900 series (gfx1100/1101) and suggest using ROCm 6.3, but do not detail mixed-GPU setups or guarantee compatibility between different models within the same architecture. There are also warnings about device detection and the need to set environment variables like CUDA_DEVICE_ORDER to avoid unexpected behavior when multiple device types are present, but no clear statement about mixing 7900 XTX and W7900 specifically is provided.
Though I’m not entirely certain, it might be that as long as both cards are recognized by ROCm and have the same architecture (gfx1100), tensor parallelism could function, but you may encounter issues related to driver, memory, or performance mismatches. You may want to test this setup and monitor for errors or unexpected behavior. Would you like more detail on ROCm setup or vLLM’s device detection and configuration process?
Sources:
- https://github.com/vllm-project/vllm/issues/19628
- https://github.com/vllm-project/vllm/issues/11249
Note: I wasn’t able to find highly relevant data sources, but above are a few potentially relevant links.
Hint: Mention @RunLLM in the post for followups.