News for hackers. Stuff that matters.

A comment on Benchmarking 15 "E-Waste" GPUs with Modern Workloads. Parent comment

Re:Benchmarking 15 "E-Waste" GPUs with Modern Workloads by eso_logic on Monday July 13, 2026 @01:55PM (#48896313)
No problem. Varying context size is a common request I've been getting as well. Personally I'm looking forward to seeing how much we can cram into the ancient K80's 24GB of VRAM :0
[ Parent | on HN ]
Re:Benchmarking 15 "E-Waste" GPUs with Modern Workloads by russianGuy83829 on Tuesday July 14, 2026 @05:49PM (#48913352)
Thank you, looking forward to it.

I just saw this simple patch to enable MTP (potentially 2x performance) on older GPUs (Kepler etc), so maybe it will work for you

https://github.com/ggml-org/llama.cpp/pull/25680

Also, for Qwen, the 4 bit _XL quantization seems to have a good balance of performance to size.

[ Parent | on HN ]