Arao AI
Back to search results

Kimi-K3 · Q4_K_S

Start a discussion
NVIDIA RTX 509032GBvLLM v0.22L2 Cross-framework verified

This page aggregates 1 real-world runs of Kimi-K3 (Q4_K_S) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

223.8 tok/s

Decode

Decode speed

1,049.81 tok/s

Prefill

Prefill speed

2.15 s

TTFT

Time to first token

1.05 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. vLLM · Q3_K_L Decode 253.52 · Prefill 1057.1 ·
  2. vLLM · Q4_K_S (current)Decode 223.8 · Prefill 1049.81 ·
  3. Strata · IQ3_S Decode 330.44 · Prefill 1792.04 ·
  4. SGLang · Q5_K_M Decode — · Prefill — ·
  5. SGLang · Q4_K_M Decode — · Prefill — ·
  6. vLLM · Q5_K_M Decode — · Prefill — ·
  7. vLLM · Q4_K_M Decode — · Prefill — ·

Core figures

Decode (avg)
223.8 tok/s
Prefill (avg)
1,049.81 tok/s
TTFT (avg)
2.15 s
VRAM (avg)
1.05 GB
MTP acceptance rate
—
TTFB
1.79 GB
Power draw
438.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Kimi-K3
Quantization
Q4_K_S
Framework
vLLM
Version
v0.22
Context length
4096 tokens
Batch size
32
GPU layers
999
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
1.05 GB
OS
Windows 11 24H2
Driver
560.94
CUDA
13.0
Power draw
438.9 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-18

    v0.22 · Windows 11 24H2 · CUDA 13.0 · 4096 ctx

    223.8 tok/s

    Decode

    1,049.81 tok/s

    Prefill

    2.15 s

    TTFT

    1.05 GB

    VRAM

    —

    MTP

    438.9 W

    Power draw

    L2 batch seed 2026-10-18: vLLM on hw_id=13 running model_id=73 at Q4_K_S