Arao AI
Back to search results

Qwen3.5-0.8B · IQ1_S

Start a discussion
NVIDIA RTX 409024GBllama.cpp v0.28L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-0.8B (IQ1_S) on NVIDIA RTX 4090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

373.87 tok/s

Decode

Decode speed

1,737.06 tok/s

Prefill

Prefill speed

2.08 s

TTFT

Time to first token

2.8 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 373.87 · Prefill 1737.06 ·
  2. llama.cpp · Q3_K_S Decode 364.6 · Prefill 1679.71 ·
  3. vLLM · FP16 Decode 161.03 · Prefill 685.23 ·
  4. vLLM · IQ1_S Decode 328.9 · Prefill 1532.17 ·
  5. llama.cpp · IQ2_XS Decode 280.41 · Prefill 1104.34 ·
  6. llama.cpp · Q4_K_M Decode 278.24 · Prefill 916.88 ·
  7. vLLM · IQ3_XS Decode 302.65 · Prefill 1148.6 ·

Core figures

Decode (avg)
373.87 tok/s
Prefill (avg)
1,737.06 tok/s
TTFT (avg)
2.08 s
VRAM (avg)
2.8 GB
MTP acceptance rate
—
TTFB
2.99 GB
Power draw
385.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-0.8B
Quantization
IQ1_S
Framework
llama.cpp
Version
v0.28
Context length
8192 tokens
Batch size
256
GPU layers
33
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)
2.8 GB
OS
Ubuntu 22.04
Driver
570.86
CUDA
12.4
Power draw
385.9 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-18

    v0.28 · Ubuntu 22.04 · CUDA 12.4 · 8192 ctx

    373.87 tok/s

    Decode

    1,737.06 tok/s

    Prefill

    2.08 s

    TTFT

    2.8 GB

    VRAM

    —

    MTP

    385.9 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=1 running model_id=51 at IQ1_S