Arao AI
Back to search results

Qwen3.8-27B · IQ2_M

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.25L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-27B (IQ2_M) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

200.6 tok/s

Decode

Decode speed

837.93 tok/s

Prefill

Prefill speed

2.39 s

TTFT

Time to first token

30.4 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 · IQ2_M (current)Decode 200.6 · Prefill 837.93 ·
  2. Ollama · IQ3_XS Decode 161.9 · Prefill 672.48 ·
  3. vLLM · Q2_K Decode 248.25 · Prefill 794.5 ·
  4. Ollama · Q2_K Decode 231.24 · Prefill 1001.06 ·
  5. llama.cpp · IQ4_XS Decode 95.7 · Prefill 717.8 ·
  6. llama.cpp · Q5_K_S Decode 76.8 · Prefill 576 ·
  7. llama.cpp · IQ3_XXS Decode 147.1 · Prefill 1103.3 ·
  8. llama.cpp · IQ3_S Decode 128.2 · Prefill 961.5 ·
  9. llama.cpp · Q6_K Decode 59 · Prefill 442.5 ·
  10. vLLM · Q4_K_M Decode 92.9 · Prefill 696.8 ·

Core figures

Decode (avg)
200.6 tok/s
Prefill (avg)
837.93 tok/s
TTFT (avg)
2.39 s
VRAM (avg)
30.4 GB
MTP acceptance rate
—
TTFB
1.6 GB
Power draw
442.4 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-27B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.25
Context length
32768 tokens
Batch size
256
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 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
30.4 GB
OS
Windows 11 23H2
Driver
565.57
CUDA
12.4
Power draw
442.4 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

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

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

    v0.25 · Windows 11 23H2 · CUDA 12.4 · 32768 ctx

    200.6 tok/s

    Decode

    837.93 tok/s

    Prefill

    2.39 s

    TTFT

    30.4 GB

    VRAM

    —

    MTP

    442.4 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=13 running model_id=5 at IQ2_M