← sources

Make Volta Fast Again

source: reddit (reddit:LocalLLaMA)published 2026-09-25 17:02ingested 2026-09-26 02:38

retained text

For those who have V100 cards, I wanted to point you to 1Cat-vLLM, a vLLM fork that enables optimized serving for these cards. Showing stats for Qwen3.6-35b comparing a Strix Halo with a hughly  
optimized llama.cpp fork (pwilkin) and the V100 with 1Cat. It’s not apples to apples, but I decided to show the raw numbers from llama-benchy so folks get an idea of the performance. IMO this is still very good for 10 year old GPUs.

Welcome any other suggestions for optimization!

document fetched 2026-09-26 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1wq1rmf/make_volta_fast_again