Back to search results
Qwen3-4B · IQ3_S
2x AMD Strix HaloOllama v0.31L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-4B (IQ3_S) on 2x AMD Strix Halo with Ollama, contributed by 1 independent source platforms; metrics are averages of published measurements.
197.15 tok/s
Decode
Decode speed
955.03 tok/s
Prefill
Prefill speed
1.71 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 · FP16 Decode 87.02 · Prefill 443.84 ·
- Ollama · IQ3_S (current)Decode 197.15 · Prefill 955.03 ·
- Ollama · Q4_K_M Decode 165.67 · Prefill 892.04 ·
Core figures
- Decode (avg)
- 197.15 tok/s
- Prefill (avg)
- 955.03 tok/s
- TTFT (avg)
- 1.71 s
- VRAM (avg)
- 7.64 GB
- MTP acceptance rate
- —
- TTFB
- 1.58 GB
- Power draw
- 84.1 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B
- Quantization
- IQ3_S
- Framework
- Ollama
- Version
- v0.31
- Context length
- 32768 tokens
- Batch size
- 256
- GPU layers
- 20
- 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)
- 7.64 GB
- OS
- Windows 11 23H2
- Driver
- 565.57
- CUDA
- N/A
- Power draw
- 84.1 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.31 · Windows 11 23H2 · CUDA N/A · 32768 ctx
197.15 tok/s
Decode
955.03 tok/s
Prefill
1.71 s
TTFT
7.64 GB
VRAM
—
MTP
84.1 W
Power draw
L2 batch seed 2026-10-16: Ollama on hw_id=88 running model_id=54 at IQ3_S