Arao AI
Back to search results

Qwen3.5-0.8B · FP16

Start a discussion
2x AMD Strix HalovLLM v0.40L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-0.8B (FP16) on 2x AMD Strix Halo with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

163.9 tok/s

Decode

Decode speed

573.99 tok/s

Prefill

Prefill speed

2.35 s

TTFT

Time to first token

5.21 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. vLLM · FP16 (current)Decode 163.9 · Prefill 573.99 ·
  2. vLLM · IQ3_XS Decode 352.1 · Prefill 1673.75 ·
  3. Strata · Q2_K Decode 284.75 · Prefill 1118.17 ·
  4. Strata · Q5_K_S Decode 358.45 · Prefill 1286.06 ·

Core figures

Decode (avg)
163.9 tok/s
Prefill (avg)
573.99 tok/s
TTFT (avg)
2.35 s
VRAM (avg)
5.21 GB
MTP acceptance rate
—
TTFB
1.31 GB
Power draw
68.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-0.8B
Quantization
FP16
Framework
vLLM
Version
v0.40
Context length
8192 tokens
Batch size
8
GPU layers
999
Flash Attention
On

Hardware

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

GPU
2x AMD Strix Halo
Measured VRAM (avg)
5.21 GB
OS
Ubuntu 22.04
Driver
amdgpu-pro 24.20
CUDA
N/A
Power draw
68.8 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.40 · Ubuntu 22.04 · CUDA N/A · 8192 ctx

    163.9 tok/s

    Decode

    573.99 tok/s

    Prefill

    2.35 s

    TTFT

    5.21 GB

    VRAM

    —

    MTP

    68.8 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=88 running model_id=51 at FP16