Arao AI
Back to search results

DeepSeek-V4-Flash · IQ3_S

Start a discussion
NVIDIA RTX 4080 Super16GBvLLM 0.7.3L0 Self-reported

This page aggregates 1 real-world runs of DeepSeek-V4-Flash (IQ3_S) on NVIDIA RTX 4080 Super with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

5.4 tok/s

Decode

Decode speed

40.5 tok/s

Prefill

Prefill speed

2.47 s

TTFT

Time to first token

16 GB

VRAM

VRAM usage

L0 Self-reported

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. vLLM · Q4_K_M Decode 3.6 · Prefill 27 ·
  2. vLLM · IQ3_S (current)Decode 5.4 · Prefill 40.5 ·
  3. vLLM · IQ4_XS Decode 4.1 · Prefill 30.7 ·
  4. llama.cpp · Q4_K_M Decode 3.3 · Prefill 24.8 ·
  5. llama.cpp · IQ3_S Decode 5 · Prefill 37.5 ·
  6. llama.cpp · IQ4_XS Decode 3.7 · Prefill 27.8 ·

Core figures

Decode (avg)
5.4 tok/s
Prefill (avg)
40.5 tok/s
TTFT (avg)
2.47 s
VRAM (avg)
16 GB
MTP acceptance rate
—
TTFB
— GB
Power draw
— W

Configuration

Member-level fields are taken from the most recent run

Model
DeepSeek-V4-Flash
Quantization
IQ3_S
Framework
vLLM
Version
0.7.3
Context length
4096 tokens
Batch size
512
GPU layers
99
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 4080 Super
Nominal VRAM
16 GB
Measured VRAM (avg)
16 GB
OS
Linux
Driver
570.65
CUDA
12.6

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

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

  1. L0 Self-reportedOtherOriginal link Verified on 2026-10-06

    0.7.3 · Linux · CUDA 12.6 · 4096 ctx

    5.4 tok/s

    Decode

    40.5 tok/s

    Prefill

    2.47 s

    TTFT

    16 GB

    VRAM

    —

    MTP

    — W

    Power draw

    L0 理论估算（add-1000-records-batch3）