llamaperf

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

on 4× NVIDIA RTX 3080 20GB · Strata · 1,000,000 ctx

Tone: positiveNVIDIA hardware
Nov 5, 2026
Throughput
37.0 t/s gen
Quant
IQ2-XXS (GGUF)
System RAM
128 GB
VRAM reported
20 GB

Summary

User reports DeepSeek V4.1 Flash at 37 t/s decode on four RTX 3080 20GB cards with 128GB DDR4 system RAM. Setup is Strata with an IQ2-XXS GGUF at 1M context; prefill is given as a 150-300 t/s range. User compares it with Qwen3.8-Flash-Next (IQ3-S, 256K context) on the same machine, which reached 100 t/s decode and 1400-1500 t/s prefill but failed twice and took long to finish, while DeepSeek completed the task in around 15 minutes.

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 3080 20GB

Bars are medians on one scale, 0 to 31.3 tokens/s. Each dot is one run and opens its report. Press ▶ to watch text arrive at a row's speed.

Everything reported on the NVIDIA RTX 3080 20GB →Every DeepSeek V4.1 Flash report →Estimate DeepSeek V4.1 Flash on this card →