Arao AI
Back to search results

Qwen3-8B · IQ2_XS

Start a discussion
NVIDIA RTX 507012GBllama.cpp v0.53L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (IQ2_XS) on NVIDIA RTX 5070 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

93.24 tok/s

Decode

Decode speed

441.9 tok/s

Prefill

Prefill speed

1.39 s

TTFT

Time to first token

10.16 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ2_XS (current)Decode 93.24 · Prefill 441.9 ·
  2. llama.cpp · Q3_K_S Decode 83.47 · Prefill 327.27 ·
  3. Strata · IQ3_XS Decode 128.34 · Prefill 480.77 ·
  4. llama.cpp · Q8_0 Decode 60.9 · Prefill 456.8 ·
  5. llama.cpp · Q5_K_M Decode 88.8 · Prefill 666 ·
  6. llama.cpp · Q4_K_M Decode 107 · Prefill 802.5 ·

Core figures

Decode (avg)
93.24 tok/s
Prefill (avg)
441.9 tok/s
TTFT (avg)
1.39 s
VRAM (avg)
10.16 GB
MTP acceptance rate
—
TTFB
0.75 GB
Power draw
276.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
IQ2_XS
Framework
llama.cpp
Version
v0.53
Context length
8192 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 5070
Nominal VRAM
12 GB
Measured VRAM (avg)
10.16 GB
OS
Ubuntu 24.04
Driver
565.57
CUDA
12.8
Power draw
276.7 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

    v0.53 · Ubuntu 24.04 · CUDA 12.8 · 8192 ctx

    93.24 tok/s

    Decode

    441.9 tok/s

    Prefill

    1.39 s

    TTFT

    10.16 GB

    VRAM

    —

    MTP

    276.7 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=18 running model_id=53 at IQ2_XS