inference/research
verifiedhigh · 0.90updated 2026-10-03

vLLM issue #12529: V1 engine has higher memory usage, context length collapses from 12K to 3K on 4x RTX 3070

forum

Fields

id
sources/gotcha-vllm-issue-12529-v1-higher-memory
URL
https://github.com/vllm-project/vllm/issues/12529
Publisher
vLLM GitHub issues
Kind
forum
Published
2025-01-28
Accessed
2026-10-03
notes
Same model and flags under 0.6.x vs 0.7.0 with VLLM_USE_V1=1; closed after maintainer discussion.

Notes

Same model and flags under 0.6.x vs 0.7.0 with VLLM_USE_V1=1; closed after maintainer discussion.

Cross-references

This record references no other records.