Skip to content

Benchmark certificate · 2026-09-24.1

Both images passed their complete HTTP E2E suites before benchmarking. Prefill measures reading the prompt; decode measures generating the answer. Rates are tokens per second, with higher values indicating faster inference.

Certified prefill and decode results

Each chart row uses its own AVX512 result as the reference. Compare printed token rates across configurations, rather than bar lengths.

AVX512 certificate · AVX2 certificate · Combined results · SVG chart

Model / configuration AVX512 prefill AVX512 decode AVX2 prefill AVX2 decode
Qwen3.6-35B-A3B-UD-IQ3_S / 2×CPU · auto TP 20.3 10.4 24.8 7.9
Ornith-1.5-35B-Q4_K_M / 2×CPU · auto TP 20.3 13.2 21.8 9.9
Qwen3.5-122B-A10B-UD-Q8_K_XL-00001-of-00004 / 2×CPU + 2×CUDA · auto ExpertOverlay 171.8 13.4 107.2 13.1
Qwen3.5-122B-A10B-UD-Q8_K_XL-00001-of-00004 / 2×CPU + 2×CUDA + 4×ROCm · auto ExpertOverlay 160.1 18.9 156.5 18.6
Qwen3.5-122B-A10B-UD-Q8_K_XL-00001-of-00004 / 2×CPU + 2×ROCm · auto ExpertOverlay 195.0 23.5 106.9 23.3
Qwen3.5-122B-A10B-UD-Q8_K_XL-00001-of-00004 / 2×CPU + 4×ROCm · auto ExpertOverlay 228.8 24.0 165.2 23.7
Ornith-1.5-35B-Q4_K_M / 2×CUDA · auto TP 1,049.8 106.1 1,061.7 106.3
Qwen3.6-35B-A3B-UD-IQ3_S / 1×CUDA · auto single 2,608.5 225.5 2,579.1 224.0
Qwen3.8-27B-IQ4_XS / 2×CUDA · auto TP 741.8 50.8 740.3 50.8
Qwen3.8-27B-IQ4_XS / 2×CUDA · auto PP 1,124.6 50.8 1,121.7 50.8
Qwen3.8-27B-IQ4_XS / 1×CUDA · auto single 913.2 62.7 915.5 62.7
Qwen3.5-122B-A10B-UD-Q8_K_XL-00001-of-00004 / 2×CUDA + 4×ROCm · auto ExpertOverlay 188.6 23.1 197.9 23.9
Qwen3.8-27B-IQ4_XS / 2×CUDA + 2×ROCm · auto PP 499.1 36.7 502.2 36.7
Ornith-1.5-35B-Q4_K_M / 2×ROCm · auto TP 1,042.6 65.9 1,041.5 66.0
Qwen3.6-35B-A3B-UD-IQ3_S / 1×ROCm · auto single 1,160.3 74.7 1,156.3 74.1
Ornith-1.5-35B-Q4_K_M / 1×ROCm · auto single 1,160.5 100.3 1,164.5 100.4
Qwen3.8-27B-IQ4_XS / 2×ROCm · auto TP 373.4 38.6 372.2 38.6
Qwen3.8-27B-IQ4_XS / 2×ROCm · auto PP 271.8 28.0 271.8 28.1
Qwen3.8-27B-IQ4_XS / 1×ROCm · auto single 274.2 36.6 270.8 36.6