deepseek-v4.1IMAGE
DeepSeek V4.1 Flash hits 25 tok/s at 1M context on dual RTX 3090s
Running DeepSeek-V4.1-Flash-1017 with ISTA IQ2_XXS quantization on dual RTX 3090s in a 2014 DDR4 workstation reaches 25 tok/s at 1M context.
Runner
@RRunner144Archived 6 days agoOriginally posted 2026.10.25
Runner @RRunner144Originally posted 2026.10.25
Back home, got DeepSeek-V4.1-Flash-1017 on a pair of 3090s in a 2014 DDR4 workstation doing 25 tok/s using 1M context. I can't stop laughing and I know Nikos is only going to make this thing faster. ISTA IQ2_XXS Quant using 1M context
This post quotes another post: x.com/rrunner144/status/2114016152351338909
x.com/rrunner144/status/2114162596521607559Archived 2026.10.25 12:00