Arao AI
Back to search results

Qwen3-8B · IQ1_S

Start a discussion
NVIDIA RTX 409024GBllama.cpp v0.36L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (IQ1_S) on NVIDIA RTX 4090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

187.88 tok/s

Decode

Decode speed

983.15 tok/s

Prefill

Prefill speed

0.64 s

TTFT

Time to first token

9.16 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ1_S (current)Decode 187.88 · Prefill 983.15 ·
  2. llama.cpp · Q8_0 Decode 87.18 · Prefill 465.1 ·
  3. vLLM · Q2_K Decode 164.91 · Prefill 671.93 ·
  4. llama.cpp · Q6_K Decode 88.33 · Prefill 428.17 ·
  5. vLLM · Q4_K_S Decode 126.59 · Prefill 467.91 ·
  6. vLLM · Q8_0 Decode 76.65 · Prefill 513.21 ·
  7. Ollama · Q6_K Decode 82.66 · Prefill 332.47 ·
  8. llama.cpp · IQ4_XS Decode 112 · Prefill 1323 ·
  9. vLLM · IQ4_XS Decode 104 · Prefill 1176 ·
  10. vLLM · Q4_K_M Decode 96 · Prefill 1061 ·
  11. vLLM · Q5_K_M Decode 95 · Prefill 815 ·
  12. Ollama · Q4_K_M Decode 113 · Prefill 922 ·
  13. Ollama · Q5_K_M Decode 95 · Prefill 1090 ·
  14. llama.cpp · Q5_K_M Decode 133.3 · Prefill 999.8 ·
  15. llama.cpp · Q4_K_M Decode 160.5 · Prefill 1203.8 ·

Core figures

Decode (avg)
187.88 tok/s
Prefill (avg)
983.15 tok/s
TTFT (avg)
0.64 s
VRAM (avg)
9.16 GB
MTP acceptance rate
—
TTFB
1.5 GB
Power draw
315.2 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
IQ1_S
Framework
llama.cpp
Version
v0.36
Context length
2048 tokens
Batch size
256
GPU layers
99
Flash Attention
Off

Hardware

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

GPU
NVIDIA RTX 4090
Nominal VRAM
24 GB
Measured VRAM (avg)
9.16 GB
OS
Windows 11 23H2
Driver
560.94
CUDA
12.6
Power draw
315.2 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

    v0.36 · Windows 11 23H2 · CUDA 12.6 · 2048 ctx

    187.88 tok/s

    Decode

    983.15 tok/s

    Prefill

    0.64 s

    TTFT

    9.16 GB

    VRAM

    —

    MTP

    315.2 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=1 running model_id=53 at IQ1_S