deepseek-v4.1IMAGE
DeepSeek V4.1 Flash on 4× CMP 170HX: kernel tweak adds 2–3% decode, 23% faster than 4× DGX Spark at 8 users
A kernel optimization on four CMP 170HX cards speeds up DeepSeek V4.1-Flash-1017 decode by 2–3% over the author's previous best; against 4× DGX Spark it is 23% faster at 8-user decode but loses on cold prefill.
BeyondBacktesting
@BBacktestingArchived 10 days agoOriginally posted 2026.10.22
BeyondBacktesting @BBacktestingOriginally posted 2026.10.22
DeepSeek V4.1-Flash-1017 4× CMP 170HX Kernel Boost We found a small kernel optimization +2% to 3% faster decode vs our previous best. We also ran matched benchmark comparisons vs 4x DGX spark. We were 23% faster on decode at 8 users but Sparks led cold prefill. https://github.com/CurtisAccelerate/DeepSeekv41-Flash-1017-EXL3-CMP170HX
x.com/bbacktesting/status/2112972765729021975Archived 2026.10.22 12:00