Back to search results
Qwen3-4B-Thinking-2507 · IQ3_S
AMD Strix Halo (Radeon 8060S)Strata v0.28L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (IQ3_S) on AMD Strix Halo (Radeon 8060S) with Strata, contributed by 1 independent source platforms; metrics are averages of published measurements.
166.26 tok/s
Decode
Decode speed
669.87 tok/s
Prefill
Prefill speed
0.77 s
TTFT
Time to first token
7.64 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- llama.cpp · Q3_K_S Decode 117.04 · Prefill 404.22 ·
- llama.cpp · Q5_K_M Decode 77.99 · Prefill 423.17 ·
- Strata · IQ2_XS Decode 206.18 · Prefill 1005.37 ·
- Strata · IQ3_S (current)Decode 166.26 · Prefill 669.87 ·
Core figures
- Decode (avg)
- 166.26 tok/s
- Prefill (avg)
- 669.87 tok/s
- TTFT (avg)
- 0.77 s
- VRAM (avg)
- 7.64 GB
- MTP acceptance rate
- 0.53
- TTFB
- 1.33 GB
- Power draw
- 92.7 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B-Thinking-2507
- Quantization
- IQ3_S
- Framework
- Strata
- Version
- v0.28
- Context length
- 4096 tokens
- Batch size
- 1024
- GPU layers
- 8
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- AMD Strix Halo (Radeon 8060S)
- Measured VRAM (avg)
- 7.64 GB
- OS
- Ubuntu 24.04
- Driver
- amdgpu-pro 24.20
- CUDA
- N/A
- Power draw
- 92.7 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.28 · Ubuntu 24.04 · CUDA N/A · 4096 ctx
166.26 tok/s
Decode
669.87 tok/s
Prefill
0.77 s
TTFT
7.64 GB
VRAM
0.53
MTP
92.7 W
Power draw
L2 batch seed 2026-10-16: Strata on hw_id=87 running model_id=57 at IQ3_S