DeepSeek V4 Flash (98GB) on 1x 4060ti + CPU got 300% faster this week [ 2->7t/s]
英文标题：DeepSeek V4 Flash (98GB) on 1x 4060ti + CPU got 300% faster this week [ 2->7t/s]
This is an an insane budget box that I've been using to test out a 98GB model using cpu generation on a 6 core CPU, 16gb vram.. for science. This week it went from 2t/s -> 7t/s on DeepSeek-V4-Flash-UD-Q2_K_XL, which has a 98GB vram requirement. Somewhere between b9986 and b10034 the llamacpp
打开原文相关 2 · 机会 0
中文
This is an an insane budget box that I've been using to test out a 98GB model using cpu generation on a 6 core CPU, 16gb vram.. for science. This week it went from 2t/s -> 7t/s on DeepSeek-V4-Flash-UD-Q2_K_XL, which has a 98GB vram requirement. Somewhere between b9986 and b10034 the llamacpp
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
This is an an insane budget box that I've been using to test out a 98GB model using cpu generation on a 6 core CPU, 16gb vram.. for science. This week it went from 2t/s -> 7t/s on DeepSeek-V4-Flash-UD-Q2_K_XL, which has a 98GB vram requirement. Somewhere between b9986 and b10034 the llamacpp