vLLM issue #37242: CUDA graphs DO work on RTX 5090 (sm_120) under WSL2 2.7.0, contrary to community consensus that it is a permanent WSL2/Blackwell limitation

Fields

id

sources/comm-github-vllm-37242-5090-wsl2-cudagraphs

URL

https://github.com/vllm-project/vllm/issues/37242

Publisher

vLLM GitHub issues

Kind

forum

Accessed

2026-10-04

notes

Reporter Kyzcreig: RTX 5090 32GB, Windows 11 + WSL2 2.7.0 pre-release, kernel 6.6.114.1-microsoft-standard-WSL2, CUDA 12.8, driver 581.80, vLLM 0.17.1, Qwen/Qwen3-14B-AWQ (awq_marlin). INDEPENDENT CORROBORATION: commenter SoloRyuken confirms CUDA graph capture working on RTX 5090 + WSL2 2.6.3 stable with ComfyUI 0.19.x (different stack, same conclusion). Issue was auto-closed by stale bot, not by a fix. Reported fixes are all WSL2/driver-service ordering, not GPU-specific.

Notes

Reporter Kyzcreig: RTX 5090 32GB, Windows 11 + WSL2 2.7.0 pre-release, kernel 6.6.114.1-microsoft-standard-WSL2, CUDA 12.8, driver 581.80, vLLM 0.17.1, Qwen/Qwen3-14B-AWQ (awq_marlin). INDEPENDENT CORROBORATION: commenter SoloRyuken confirms CUDA graph capture working on RTX 5090 + WSL2 2.6.3 stable with ComfyUI 0.19.x (different stack, same conclusion). Issue was auto-closed by stale bot, not by a fix. Reported fixes are all WSL2/driver-service ordering, not GPU-specific.

Cross-references

This record references no other records.