Arao AI
Back to search results

Kimi-K3 · IQ3_S

Start a discussion
NVIDIA RTX 5070 Ti16GBvLLM v0.31L2 Cross-framework verified

This page aggregates 1 real-world runs of Kimi-K3 (IQ3_S) on NVIDIA RTX 5070 Ti with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

95.38 tok/s

Decode

Decode speed

330.41 tok/s

Prefill

Prefill speed

1.59 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 · IQ1_S Decode 131.31 · Prefill 466.87 ·
  2. vLLM · IQ3_S (current)Decode 95.38 · Prefill 330.41 ·
  3. Strata · IQ3_S Decode 132.6 · Prefill 715.44 ·
  4. Strata · Q3_K_S Decode 129.13 · Prefill 489.23 ·

Core figures

Decode (avg)
95.38 tok/s
Prefill (avg)
330.41 tok/s
TTFT (avg)
1.59 s
VRAM (avg)
1.05 GB
MTP acceptance rate
—
TTFB
0.76 GB
Power draw
206.3 W

Configuration

Member-level fields are taken from the most recent run

Model
Kimi-K3
Quantization
IQ3_S
Framework
vLLM
Version
v0.31
Context length
8192 tokens
Batch size
32
GPU layers
999
Flash Attention
Off

Hardware

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

GPU
NVIDIA RTX 5070 Ti
Nominal VRAM
16 GB
Measured VRAM (avg)
1.05 GB
OS
Ubuntu 22.04
Driver
570.86
CUDA
12.4
Power draw
206.3 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

    v0.31 · Ubuntu 22.04 · CUDA 12.4 · 8192 ctx

    95.38 tok/s

    Decode

    330.41 tok/s

    Prefill

    1.59 s

    TTFT

    1.05 GB

    VRAM

    —

    MTP

    206.3 W

    Power draw

    L2 batch seed 2026-10-18: vLLM on hw_id=17 running model_id=73 at IQ3_S