llamaperf

DeepSeek V4 Flash 284B (13B active)

on NVIDIA RTX 3090 · llama.cpp · 8,192 ctx

Tone: mixedNVIDIA hardware
Sep 7, 2026
Throughput
6.6 t/s gen · 37.5 tokens/s prompt processing (prefill)Prompt processing measures input prefill speed; it does not tell you the time to first token.
Quant
Q4_K_XL (GGUF)
System RAM
192 GB
VRAM reported
24 GB

Summary

User reports DeepSeek V4 Flash at 37.47 t/s pp and 6.62 t/s tg on a dual Xeon E5-2660 v2 with an RTX 3090. Setup is the Q4_K_XL quant at 8k context. The user also tested Cydonia v4.3, a Mistral Small 3.2 finetune, at 308.92 t/s pp and 7.8 t/s tg.