Premiers pasClassementDecode calculatorModèlesOrganizationsReportsMatrix runsMatérielÉvaluations comparativesMarketplaceLocationsProDocs API
Langue
Actual Computer — Every computer, one endpoint

Matrix runs

A matrix run measures prompt reading (prefill) and generation (decode) speed in one go, across several context lengths and numbers of simultaneous requests. Run one with lmx speed-test matrix run.

All runs 1,102

Newest first. Page 14 of 45.
  • GLM-5.3-Flash-W4A16-MTPCMP 170HX 64GB ×4 · vllm 0.32.0rc1.dev215+g819146023 · W4A167 contexts × 5 concurrency levels
    Peak decode
    431.5 tok/s
    Peak prefill
    5,652 tok/s
    by @Sandberg
  • granite-4.2-30b-nvfp4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP45 contexts × 8 concurrency levels
    Peak decode
    565.1 tok/s
    Peak prefill
    1,654 tok/s
    by @Sandberg
  • Qwen3-14B-AWQCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AWQ3 contexts × 8 concurrency levels
    Peak decode
    1,564.7 tok/s
    Peak prefill
    3,523 tok/s
    by @Sandberg
  • NVIDIA-Nemotron-Nano-12B-v2-AWQ-4bitCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AWQ 4-bit5 contexts × 8 concurrency levels
    Peak decode
    1,831.9 tok/s
    Peak prefill
    4,169 tok/s
    by @Sandberg
  • Ornith-1.5-9B-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP46 contexts × 8 concurrency levels
    Peak decode
    2,410.9 tok/s
    Peak prefill
    5,592 tok/s
    by @Sandberg
  • NVIDIA-Nemotron-Nano-9B-v2-quantized.w4a16CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · W4A165 contexts × 8 concurrency levels
    Peak decode
    2,026.4 tok/s
    Peak prefill
    4,921 tok/s
    by @Sandberg
  • Tess-4-27B-NVFP4RTX 3090 24GB · vllm 0.32.0 · NVFP44 contexts × 8 concurrency levels
    Peak decode
    180.2 tok/s
    Peak prefill
    1,784 tok/s
    by @Sandberg
  • gemma-4-26B-A4B-it-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP46 contexts × 8 concurrency levels
    Peak decode
    2,994.2 tok/s
    Peak prefill
    6,995 tok/s
    by @Sandberg
  • Qwen3-8B-AWQCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AWQ3 contexts × 8 concurrency levels
    Peak decode
    1,989.9 tok/s
    Peak prefill
    6,083 tok/s
    by @Sandberg
  • Qwen3.8-27B-INT4RTX 3090 24GB · vllm 0.32.0 · INT4 W4A16 g1285 contexts × 8 concurrency levels
    Peak decode
    348.9 tok/s
    Peak prefill
    1,816 tok/s
    by @Sandberg
  • Qwen3.5-4B-AWQ-4bitCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AWQ 4-bit6 contexts × 8 concurrency levels
    Peak decode
    3,451.6 tok/s
    Peak prefill
    9,663 tok/s
    by @Sandberg
  • deepseek-r1-distill-qwen-14b-awqCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AWQ5 contexts × 8 concurrency levels
    Peak decode
    1,487.7 tok/s
    Peak prefill
    3,427 tok/s
    by @Sandberg
  • NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4RTX 3090 24GB · vllm 0.32.0 · NVFP44 contexts × 8 concurrency levels
    Peak decode
    264.2 tok/s
    Peak prefill
    8,160 tok/s
    by @Sandberg
  • Qwen3.6-35B-A3B-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP46 contexts × 8 concurrency levels
    Peak decode
    2,916.2 tok/s
    Peak prefill
    6,639 tok/s
    by @Sandberg
  • Phi-4-reasoning-plus-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP43 contexts × 8 concurrency levels
    Peak decode
    1,489.8 tok/s
    Peak prefill
    3,267 tok/s
    by @Sandberg
  • granite-4.2-3b-nvfp4RTX 3090 24GB · vllm 0.32.0 · NVFP45 contexts × 8 concurrency levels
    Peak decode
    1,775.8 tok/s
    Peak prefill
    12.1K tok/s
    by @Sandberg
  • Ornith-1.5-9B-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP46 contexts × 8 concurrency levels
    Peak decode
    3,578.1 tok/s
    Peak prefill
    5,826 tok/s
    by @Sandberg
  • Qwen3.6-35B-A3B-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP46 contexts × 8 concurrency levels
    Peak decode
    3,006.5 tok/s
    Peak prefill
    7,043 tok/s
    by @Sandberg
  • granite-4.2-8b-nvfp4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP45 contexts × 8 concurrency levels
    Peak decode
    1,825.1 tok/s
    Peak prefill
    5,123 tok/s
    by @Sandberg
  • Hy-MT2-30B-A3B-FP8CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · FP86 contexts × 8 concurrency levels
    Peak decode
    1,258.7 tok/s
    Peak prefill
    7,809 tok/s
    by @Sandberg
  • Fara1.5-4B-AutoRound-W4A16-RTNCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AutoRound W4A166 contexts × 8 concurrency levels
    Peak decode
    4,682.7 tok/s
    Peak prefill
    9,791 tok/s
    by @Sandberg
  • NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · NVFP48 contexts × 8 concurrency levels
    Peak decode
    2,385.6 tok/s
    Peak prefill
    7,570 tok/s
    by @Sandberg
  • NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4RTX 3090 24GB · vllm 0.32.0 · NVFP48 contexts × 8 concurrency levels
    Peak decode
    1,604.3 tok/s
    Peak prefill
    9,619 tok/s
    by @Sandberg
  • Qwen3.5-4B-AWQ-4bitCMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · AWQ 4-bit6 contexts × 8 concurrency levels
    Peak decode
    4,810.5 tok/s
    Peak prefill
    9,988 tok/s
    by @Sandberg
  • Qwen3.8-27B-FP8CMP 170HX 64GB · vllm 0.32.0rc1.dev249+g18305948a · FP86 contexts × 8 concurrency levels
    Peak decode
    652.0 tok/s
    Peak prefill
    1,688 tok/s
    by @Sandberg