inference/research
verifiedhigh · 0.85updated 2026-10-03

vLLM issue 40980: TP=2 deadlock on dual AMD R9700 (gfx1201/RDNA4) - GPUs spin at 100%

repo

Fields

id
sources/mgpu-vllm-40980-rocm-tp2-deadlock
URL
https://github.com/vllm-project/vllm/issues/40980
Publisher
vllm-project/vllm
Kind
repo
Published
2026-04-27
Accessed
2026-10-03
notes
v0.19.1+rocm721 / v0.19.2rc1.dev215+g32e45636e, PyTorch 2.11.0 built for gfx1201, ROCm 7.2.1, RCCL 2.27.7, Fedora 43. Reports rocm-smi at 100% on both GPUs with 0% VRAM, and RCCL working in isolation at world_size=1.

Notes

v0.19.1+rocm721 / v0.19.2rc1.dev215+g32e45636e, PyTorch 2.11.0 built for gfx1201, ROCm 7.2.1, RCCL 2.27.7, Fedora 43. Reports rocm-smi at 100% on both GPUs with 0% VRAM, and RCCL working in isolation at world_size=1.

Cross-references

This record references no other records.