Back to search results
Qwen3.5-9B · Q5_K_S
AMD Strix Halo (Radeon 8060S)llama.cpp v0.48L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3.5-9B (Q5_K_S) on AMD Strix Halo (Radeon 8060S) with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
60.47 tok/s
Decode
Decode speed
213.4 tok/s
Prefill
Prefill speed
0.26 s
TTFT
Time to first token
19.11 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 · IQ1_S Decode 116.24 · Prefill 433.01 ·
- llama.cpp · Q5_K_S (current)Decode 60.47 · Prefill 213.4 ·
- Strata · Q4_K_S Decode 105.67 · Prefill 508.86 ·
- Strata · Q6_K Decode 73.08 · Prefill 234.71 ·
Core figures
- Decode (avg)
- 60.47 tok/s
- Prefill (avg)
- 213.4 tok/s
- TTFT (avg)
- 0.26 s
- VRAM (avg)
- 19.11 GB
- MTP acceptance rate
- —
- TTFB
- 0.48 GB
- Power draw
- 84.2 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-9B
- Quantization
- Q5_K_S
- Framework
- llama.cpp
- Version
- v0.48
- Context length
- 2048 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
- AMD Strix Halo (Radeon 8060S)
- Measured VRAM (avg)
- 19.11 GB
- OS
- Ubuntu 22.04
- Driver
- amdgpu 6.11
- CUDA
- N/A
- Power draw
- 84.2 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.48 · Ubuntu 22.04 · CUDA N/A · 2048 ctx
60.47 tok/s
Decode
213.4 tok/s
Prefill
0.26 s
TTFT
19.11 GB
VRAM
—
MTP
84.2 W
Power draw
L2 batch seed 2026-10-16: llama.cpp on hw_id=87 running model_id=49 at Q5_K_S