Arao AI
Back to search results

Qwen3-8B · Q6_K

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.32L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (Q6_K) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

179.01 tok/s

Decode

Decode speed

827.24 tok/s

Prefill

Prefill speed

0.83 s

TTFT

Time to first token

18.9 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ4_XS Decode 253.29 · Prefill 909.82 ·
  2. llama.cpp · Q6_K (current)Decode 179.01 · Prefill 827.24 ·
  3. vLLM · IQ3_S Decode 284.83 · Prefill 1483.72 ·
  4. vLLM · IQ4_XS Decode 252.51 · Prefill 829.1 ·
  5. Ollama · Q5_K_M Decode 172.9 · Prefill 940.17 ·
  6. llama.cpp · Q8_0 Decode 162.5 · Prefill 1218.8 ·
  7. llama.cpp · Q5_K_M Decode 236.9 · Prefill 1776.8 ·
  8. llama.cpp · Q4_K_M Decode 285.3 · Prefill 2139.8 ·

Core figures

Decode (avg)
179.01 tok/s
Prefill (avg)
827.24 tok/s
TTFT (avg)
0.83 s
VRAM (avg)
18.9 GB
MTP acceptance rate
—
TTFB
1.43 GB
Power draw
444.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
Q6_K
Framework
llama.cpp
Version
v0.32
Context length
32768 tokens
Batch size
512
GPU layers
33
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)
18.9 GB
OS
Ubuntu 22.04
Driver
565.57
CUDA
12.6
Power draw
444.9 W

Sources & evidence

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

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

    v0.32 · Ubuntu 22.04 · CUDA 12.6 · 32768 ctx

    179.01 tok/s

    Decode

    827.24 tok/s

    Prefill

    0.83 s

    TTFT

    18.9 GB

    VRAM

    —

    MTP

    444.9 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=13 running model_id=53 at Q6_K