llamaperf

DeepSeek V4.1 Flash 552B (16B active)

on 4× NVIDIA RTX PRO 6000 Max-Q · SGLang

Tone: positiveNVIDIA hardware
Nov 11, 2026
Throughput
1280.0 t/s gen
Quant
FP8
System RAM
768 GB

Use cases

agentic

Summary

User reports DeepSeek V4.1 Flash at 1280 tok/s on four RTX PRO 6000 Max-Q Workstation Edition GPUs in a Dell Precision 9 T6 workstation with a Xeon 654 and 768GB RAM. Setup is SGLang 0.5.23 with an FP8 config, tensor parallel across the four cards, plus patches ported from Sybil's build and added patches for DeepGEMM, HiCache, page width and SWA. User says the tuning gave around a 20% performance increase and believes this is the fastest build example of DS4.1 Flash on four RTX 6000 cards.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

DeepSeek V4.1 Flash 552B · 16B active at 3-bit by hardware

Bars are medians on one scale, 0 to 73.4 tokens/s. Press ▶ to watch text arrive at a row's speed.

Models most run on the NVIDIA RTX PRO 6000 Max-Q

Bars are medians on one scale, 0 to 291.0 tokens/s. Press ▶ to watch text arrive at a row's speed.

Everything reported on the NVIDIA RTX PRO 6000 Max-Q →Every DeepSeek V4.1 Flash report →Estimate DeepSeek V4.1 Flash on this card →