deepseek-v4.1IMAGE

4× CMP 170HX 跑 V4.1 Flash：内核优化提速 2–3%，8 用户解码比 4× DGX Spark 快

作者在 4 张 CMP 170HX 上为 V4.1-Flash-1017 做了内核优化，解码比此前最好成绩快 2–3%；与 4× DGX Spark 同题对比，8 用户解码快 23%，但冷启动 prefill 落后。

BeyondBacktesting

@BBacktesting10 天前收录原推发布于 2026.10.22

4× CMP 170HX 跑 V4.1 Flash：内核优化提速 2–3%，8 用户解码比 4× DGX Spark 快

BeyondBacktesting @BBacktesting原推发布于 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/2112972765729021975采集于 2026.10.22 12:00