Arao AI
Back to search results

Qwen3.8-9B · IQ2_XS

Start a discussion
NVIDIA RTX 309024GBllama.cpp v0.53L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-9B (IQ2_XS) on NVIDIA RTX 3090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

98.31 tok/s

Decode

Decode speed

360.63 tok/s

Prefill

Prefill speed

1.75 s

TTFT

Time to first token

11.17 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ2_XS (current)Decode 98.31 · Prefill 360.63 ·
  2. llama.cpp · Q3_K_L Decode 79.17 · Prefill 416.87 ·
  3. Ollama · Q5_K_S Decode 58.86 · Prefill 223.43 ·
  4. Ollama · Q8_0 Decode 46.15 · Prefill 154.63 ·
  5. llama.cpp · Q3_K_S Decode 130.89 · Prefill 575.17 ·
  6. Ollama · IQ4_XS Decode 98.13 · Prefill 496.33 ·

Core figures

Decode (avg)
98.31 tok/s
Prefill (avg)
360.63 tok/s
TTFT (avg)
1.75 s
VRAM (avg)
11.17 GB
MTP acceptance rate
—
TTFB
0.79 GB
Power draw
252.4 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-9B
Quantization
IQ2_XS
Framework
llama.cpp
Version
v0.53
Context length
2048 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 3090
Nominal VRAM
24 GB
Measured VRAM (avg)
11.17 GB
OS
Ubuntu 22.04
Driver
570.86
CUDA
12.4
Power draw
252.4 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

    v0.53 · Ubuntu 22.04 · CUDA 12.4 · 2048 ctx

    98.31 tok/s

    Decode

    360.63 tok/s

    Prefill

    1.75 s

    TTFT

    11.17 GB

    VRAM

    —

    MTP

    252.4 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=2 running model_id=52 at IQ2_XS