deepseek-v4.1IMAGE
DeepSeek V4.1 Flash IQ2-XXS GGUF test: VRAM beats extra RAM for expert capacity
Testing ISTA-DASLab's IQ2-XXS GGUF on a 32GB Pro4500 and a 16GB 5080 build, the author finds VRAM minus shared weights decides how many experts fit, and more DRAM helps less than more VRAM.
おののき＠意識を低く持て！
@tjm8874Archived 6 days agoOriginally posted 2026.10.25
おののき＠意識を低く持て！ @tjm8874Originally posted 2026.10.25
ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF IQ2-XXS、Strataでテスト Pro4500 32GB + Ryzen9 + DDR5 96GB ＞ 5080 16GB + i5 + DDR4 128GB 「VRAM容量 - 共通重み = VRAMに置けるExpert」が一番大きく違うところで、24GB対7GB DRAM増強効果よりVRAMのほうが効果高い(身も蓋もない)
x.com/tjm8874/status/2114037789369258364Archived 2026.10.25 12:02