Community
Community reports DeepSeek-V4.1-Flash and Qwen3.8-Flash-Next running fast on a single RTX 3090 Ti
The poster reports about 35 tok/s decode for DeepSeek-V4.1-Flash-1017 and about 130 to 140 tok/s for Qwen3.8-Flash-Next on a 128GB RAM Gentoo system, using a gpu-unlock kernel module and community quantized weights from Hugging Face.