Make Volta Fast Again

A developer community post highlights 1Cat-vLLM, a vLLM fork optimizing serving for V100 cards. The post compares performance for Qwen3.6-35b, showing raw numbers from llama-benchy for a V100 with 1Cat against a Strix Halo with a highly optimized llama.cpp fork. While not a direct comparison, the results suggest good performance for the 10-year-old GPUs, and the author welcomes further optimization suggestions.