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.
Matching runs 693
Submitted by @keennay. Newest first. Page 1 of 28.- Peak decode
- 752.8 tok/s
- Peak prefill
- 10.2K tok/s
- Peak decode
- 1,330.3 tok/s
- Peak prefill
- 11.4K tok/s
- GLM-5.3-Flash-NVFP4H200 NVL 141GB ×2 · vllm 0.1.dev1+g3b5189828 · NVFP48 contexts × 8 concurrency levels
- Peak decode
- 959.4 tok/s
- Peak prefill
- 13.2K tok/s
- GLM-5.3-Flash-NVFP4H200 NVL 141GB ×2 · vllm 0.1.dev1+g3b5189828 · NVFP48 contexts × 8 concurrency levels
- Peak decode
- 1,234.2 tok/s
- Peak prefill
- 13K tok/s
- GLM-5.3-Flash-NVFP4H200 SXM 141GB ×4 · vllm 0.1.dev1+g3b5189828 · NVFP48 contexts × 8 concurrency levels
- Peak decode
- 2,214.4 tok/s
- Peak prefill
- 18.4K tok/s
- Peak decode
- 2,156.0 tok/s
- Peak prefill
- 18.3K tok/s
- NVIDIA-Nemotron-3-Super-120B-A12B-FP8H200 NVL 141GB ×2 · sglang 0.5.6.post3.dev9335+gdd15fb57b · FP88 contexts × 8 concurrency levels
- Peak decode
- 2,815.6 tok/s
- Peak prefill
- 15.5K tok/s
- GLM-5.3-Flash-NVFP4RTX PRO 6000 Blackwell Server Edition 96GB ×4 · vllm 0.32.1rc1.dev96+g8c60cf07b · NVFP48 contexts × 8 concurrency levels
- Peak decode
- 1,335.7 tok/s
- Peak prefill
- 13.3K tok/s
- GLM-5.3-Flash-NVFP4RTX PRO 6000 Blackwell Server Edition 96GB ×4 · sglang 0.0.0.dev0 · NVFP48 contexts × 8 concurrency levels
- Peak decode
- 1,357.7 tok/s
- Peak prefill
- 12.8K tok/s
- Gemma-4-31B-IT-NVFP4H200 NVL 141GB ×2 · sglang 0.5.6.post3.dev9335+gdd15fb57b · NVFP46 contexts × 8 concurrency levels
- Peak decode
- 170.5 tok/s
- Peak prefill
- 6,589 tok/s
- DeepSeek-V4.1-Flash-1017H200 NVL 141GB ×2 · vllm 0.32.0 · MXFP4+FP86 contexts × 8 concurrency levels
- Peak decode
- 303.7 tok/s
- Peak prefill
- 2,844 tok/s
- Qwen3.8-27BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 1,322.3 tok/s
- Peak prefill
- 8,440 tok/s
- Qwen3.8-27BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 2,447.8 tok/s
- Peak prefill
- 8,591 tok/s
- Qwen3.6-35B-A3BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 3,960.1 tok/s
- Peak prefill
- 32K tok/s
- Qwen3.6-35B-A3BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 4,551.0 tok/s
- Peak prefill
- 31.4K tok/s
- Qwen3.6-27BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 2,108.1 tok/s
- Peak prefill
- 8,398 tok/s
- Qwen3.6-27BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 2,450.7 tok/s
- Peak prefill
- 8,599 tok/s
- Qwen3.5-35B-A3BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 3,972.8 tok/s
- Peak prefill
- 32K tok/s
- Qwen3.5-35B-A3BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 4,517.0 tok/s
- Peak prefill
- 31.4K tok/s
- Qwen3.5-35B-A3B-BaseRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 3,367.7 tok/s
- Peak prefill
- 31.9K tok/s
- Qwen3.5-35B-A3B-BaseRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 4,240.3 tok/s
- Peak prefill
- 32K tok/s
- Qwen3.5-27BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 2,208.6 tok/s
- Peak prefill
- 8,403 tok/s
- Qwen3.5-27BRTX PRO 6000 Blackwell Server Edition 96GB ×2 · vllm 0.30.1rc1.dev452+g73a583112 · BF166 contexts × 8 concurrency levels
- Peak decode
- 2,451.6 tok/s
- Peak prefill
- 8,620 tok/s
- Kimi-K2.6-NVFP4RTX PRO 6000 Blackwell Server Edition 96GB ×8 · vllm 0.1.dev1+g9fedd24fa · NVFP46 contexts × 8 concurrency levels
- Peak decode
- 362.0 tok/s
- Peak prefill
- 6,216 tok/s
- Qwen3.8-27B-FP8RTX PRO 6000 Blackwell Server Edition 96GB · vllm 0.30.1rc1.dev452+g73a583112 · FP86 contexts × 8 concurrency levels
- Peak decode
- 1,150.0 tok/s
- Peak prefill
- 9,767 tok/s