Can vLLM built for old GPU (GT 630M) ? It may use CUDA 9.1.85
I want to use old GPU (GT 630M) to run inference by vLLM.
current environment
Nvidia GT 630M
CUDA 9.1.85
Ubuntu Desktop 16.04.7
vLLM does not support GPUs with compute capability below 7.0, and there are no official pre-built wheels or documented support for CUDA versions as old as 9.1.85. The minimum supported CUDA version for vLLM is 11.8, and attempts to build or run vLLM on older CUDA versions or GPUs (like GT 630M, compute capability 2.1) are not supported and will fail due to missing kernel images and incompatibility with required CUDA features. See the official requirements and related discussions for details: docs.vllm.ai - GPU requirements, issue #19542, issue #3812, issue #4616.
Would you like more detail on supported hardware or alternative solutions?
Sources: