Arao AI
Back to search results

Qwen3-8B · Q3_K_S

Start a discussion
NVIDIA RTX 508016GBllama.cpp v0.38L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (Q3_K_S) on NVIDIA RTX 5080 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

206.58 tok/s

Decode

Decode speed

831.19 tok/s

Prefill

Prefill speed

0.14 s

TTFT

Time to first token

13.02 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

17 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_S (current)Decode 206.58 · Prefill 831.19 ·
  2. vLLM · IQ2_M Decode 240.94 · Prefill 1234.91 ·
  3. Ollama · Q5_K_S Decode 128.2 · Prefill 585.03 ·
  4. Strata · IQ1_S Decode 376.58 · Prefill 1681.2 ·
  5. llama.cpp · Q5_K_S Decode 91.75 · Prefill 454.66 ·
  6. Ollama · Q3_K_L Decode 101.88 · Prefill 405.42 ·
  7. Strata · IQ4_XS Decode 144.31 · Prefill 545.08 ·

Core figures

Decode (avg)
206.58 tok/s
Prefill (avg)
831.19 tok/s
TTFT (avg)
0.14 s
VRAM (avg)
13.02 GB
MTP acceptance rate
—
TTFB
1.65 GB
Power draw
274.5 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
Q3_K_S
Framework
llama.cpp
Version
v0.38
Context length
32768 tokens
Batch size
512
GPU layers
28
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 5080
Nominal VRAM
16 GB
Measured VRAM (avg)
13.02 GB
OS
Ubuntu 22.04
Driver
572.83
CUDA
12.6
Power draw
274.5 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.38 · Ubuntu 22.04 · CUDA 12.6 · 32768 ctx

    206.58 tok/s

    Decode

    831.19 tok/s

    Prefill

    0.14 s

    TTFT

    13.02 GB

    VRAM

    —

    MTP

    274.5 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=14 running model_id=53 at Q3_K_S