Arao AI
Back to search results

DeepSeek-V4-Flash-Vision-Exp · UD-Q8_K_XL

2x NVIDIA RTX 509064GBllama.cpp moecachev1.6-rc0L0 Self-reported

This page aggregates 1 real-world runs of DeepSeek-V4-Flash-Vision-Exp (UD-Q8_K_XL) on 2x NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

32.08 tok/s

Decode

Decode speed

671.43 tok/s

Prefill

Prefill speed

0.12 s

TTFT

Time to first token

60.7 GB

VRAM

VRAM usage

L0 Self-reported

1

Measured runs

1

Independent sources

GitHub

Source platforms

Today

Last verified

Performance

  1. llama.cpp · UD-Q8_K_XL (current)Decode 32.08 · Prefill 671.43 ·

Core figures

Decode (avg)
32.08 tok/s
Prefill (avg)
671.43 tok/s
TTFT (avg)
0.12 s
VRAM (avg)
60.7 GB
MTP acceptance rate
—
TTFB
— GB
Power draw
— W

Configuration

Member-level fields are taken from the most recent run

Model
DeepSeek-V4-Flash-Vision-Exp
Quantization
UD-Q8_K_XL
Framework
llama.cpp
Version
moecachev1.6-rc0
Context length
204800 tokens
Batch size
4096
GPU layers
99

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
2x NVIDIA RTX 5090
Nominal VRAM
64 GB
Measured VRAM (avg)
60.7 GB
OS
Ubuntu 24.04.4 LTS
Driver
610.57.04
CUDA
13.3

Get started

Startup command

Command from the latest benchmark run
model=/models/deepseek-v4-flash-vision-exp-gguf/unsloth-ud-q8kxl/UD-Q8_K_XL/DeepSeek-V4-Flash-Vision-Exp-UD-Q8_K_XL-00001-of-00005.gguf offload=blk\.[0-9]+\.ffn_(gate|up|down)_exps\.weight=CPU ubatch=4096 ngl=99 split=1,1

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L0 Self-reportedGitHubOriginal link Verified on 2026-09-20

    moecachev1.6-rc0 · Ubuntu 24.04.4 LTS · CUDA 13.3 · 204800 ctx

    32.08 tok/s

    Decode

    671.43 tok/s

    Prefill

    0.12 s

    TTFT

    60.7 GB

    VRAM

    —

    MTP

    — W

    Power draw

    [bench] narrative run 1/5: 1000 tok in 31.2s (32.23 decode TPS, ttft 145ms) [bench] narrative run 5/5: 1000 tok in 31.3s (32.03 decode TPS, ttft 113ms) [bench] prefill-10k run 1/3: 10005 prompt tok, 675.4 prefill tok/s, ttft 14812ms decode_TPS mean= 32.08 std= 0.11 CV= 0.3% min=31.98 max=32.23 [soak] boot_vram_mib 62157