Aan de slagLeaderboardDecode calculatorModellenOrganizationsReportsMatrix runsHardwareBenchmarksMarktplaatsVerhuurProAPI Docs
Taal
Actual Computer — Every computer, one endpoint

Speed test matrix

MiMo-V2.6-Pro-RL on H200 SXM 141GB ×8

vllm · 0.29.1rc1.dev533+g1ea7c63f4 · MXFP4+FP88 contexts (8K–1019K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 256K, 512K, 1019K, 1024K
by @keennaylmx dev importMeasured

Hardware

Type
Discrete GPU
GPU
H200 SXM
GPU count
8
VRAM
1128 GB (8 × 141 GB)
OS
linux

Engine

Engine
vllm
Version
0.29.1rc1.dev533+g1ea7c63f4
Quantization
MXFP4+FP8
Backend
cuda
Build
cuda-13.2
Commit
1ea7c63f4af7

Run

Served model
mimo-v2.6-pro-rl
Revision
main
Decode test
Fixed duration
Prefill mode
Measured during decode warm-up
Time per measurement
30 s
Max output
4,096 tokens

Launch command

"$HOME/env_h200-xiaomimimo-flash-vllm-1ea7c63/bin/vllm" serve "${HF_HOME:-$HOME/.cache/huggingface}/mimo-v2.6-pro-rl" --served-model-name mimo-v2.6-pro-rl --trust-remote-code --tensor-parallel-size 8 --reasoning-parser mimo --enable-auto-tool-choice --tool-call-parser mimo --max-model-len 1048576 --gpu-memory-utilization 0.87 --host 0.0.0.0 --port 8000 --api-key <redacted> --generation-config vllm --attention-backend FLASH_ATTN_DIFFKV --linear-backend flashinfer_cutlass --moe-backend marlin

Prefill

prompt tok/s by context length
8K context16.1K509 ms TTFT
16K context15.9K1,022 ms TTFT
32K context15.8K2,043 ms TTFT
64K context15.8K4,082 ms TTFT
128K context14.8K8,674 ms TTFT
256K context12.9K19,919 ms TTFT
512K context9,60453,590 ms TTFT
1019K context6,245163,993 ms TTFT
1024K context6,242164,890 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K145.356 ms TTFT269.268 ms TTFT474.2107 ms TTFT760.6148 ms TTFT1,155.8227 ms TTFT1,649.2351 ms TTFT2,196.3592 ms TTFT2,889.41,087 ms TTFT
16K144.171 ms TTFT267.4105 ms TTFT471.5120 ms TTFT760.0187 ms TTFT1,144.9311 ms TTFT1,619.5531 ms TTFT2,158.4953 ms TTFT2,817.31,789 ms TTFT
32K142.4105 ms TTFT264.2143 ms TTFT465.2181 ms TTFT743.9279 ms TTFT1,122.9468 ms TTFT1,561.6853 ms TTFT2,061.21,610 ms TTFT2,651.73,230 ms TTFT
64K138.9183 ms TTFT257.7240 ms TTFT456.4344 ms TTFT721.3556 ms TTFT1,066.5964 ms TTFT1,480.61,822 ms TTFT1,900.13,494 ms TTFTOver capacity
128K131.8365 ms TTFT244.9425 ms TTFT435.1708 ms TTFT676.41,172 ms TTFT981.22,080 ms TTFT1,308.63,954 ms TTFTOver capacityOver capacity
256K122.5759 ms TTFT228.51,083 ms TTFT403.51,615 ms TTFT603.82,711 ms TTFT844.24,932 ms TTFTOver capacityOver capacityOver capacity
512K105.31,786 ms TTFT198.22,592 ms TTFT352.33,931 ms TTFT496.26,572 ms TTFTOver capacityOver capacityOver capacityOver capacity
1019K82.33,930 ms TTFT156.55,565 ms TTFT279.78,749 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacity

Over capacity: not run because this many requests at this context would exceed the server's KV cache.

TTFT is measured after the benchmark warms the server's prompt cache, so it mostly excludes prefill.

All measurements

Latencies are medians. Total tok/s is output across all simultaneous requests.
PhaseContextRequests at onceStatusTotal tok/sPer request tok/sTTFTTime between tokensAverage active requestsCompleted requestsOutput tokensMeasured for
Prefill8K1Valid16.1K—509 ms——1——
Prefill16K1Valid15.9K—1,022 ms——1——
Prefill32K1Valid15.8K—2,043 ms——1——
Prefill64K1Valid15.8K—4,082 ms——1——
Prefill128K1Valid14.8K—8,674 ms——1——
Prefill256K1Valid12.9K—19,919 ms——1——
Prefill512K1Valid9,604.0—53,590 ms——1——
Prefill1019K1Valid6,245.0—163,993 ms——1——
Prefill1024K1Valid6,242.0—164,890 ms——1——
Sustained decode8K1Valid145.3145.756 ms6.9 ms1.014,35730.0 s
Sustained decode8K2Valid269.2135.168 ms7.4 ms2.028,07630.0 s
Sustained decode8K4Valid474.2118.7107 ms8.4 ms4.0014,22430.0 s
Sustained decode8K8Valid760.695.2148 ms10.5 ms8.0022,81930.0 s
Sustained decode8K16Valid1,155.872.5227 ms13.8 ms16.0034,67230.0 s
Sustained decode8K32Valid1,649.251.8351 ms19.3 ms32.0049,47230.0 s
Sustained decode8K64Valid2,196.334.6592 ms28.9 ms64.0065,85630.0 s
Sustained decode8K128Valid2,889.422.81,087 ms43.9 ms128.0086,65630.0 s
Sustained decode16K1Valid144.1144.671 ms6.9 ms1.014,32430.0 s
Sustained decode16K2Valid267.4134.2105 ms7.5 ms2.028,02230.0 s
Sustained decode16K4Valid471.5117.9120 ms8.5 ms4.0014,14430.0 s
Sustained decode16K8Valid760.095.0187 ms10.5 ms8.0022,80030.0 s
Sustained decode16K16Valid1,144.971.8311 ms13.9 ms16.0034,33630.0 s
Sustained decode16K32Valid1,619.550.9531 ms19.7 ms32.0048,57630.0 s
Sustained decode16K64Valid2,158.434.0953 ms29.4 ms64.0064,70430.0 s
Sustained decode16K128Valid2,817.322.31,789 ms44.8 ms128.0084,48030.0 s
Sustained decode32K1Valid142.4143.2105 ms7.0 ms1.014,27230.0 s
Sustained decode32K2Valid264.2132.9143 ms7.5 ms2.027,92530.0 s
Sustained decode32K4Valid465.2116.4181 ms8.6 ms4.0013,95230.0 s
Sustained decode32K8Valid743.993.0279 ms10.8 ms8.0022,31230.0 s
Sustained decode32K16Valid1,122.970.3468 ms14.2 ms16.0033,68030.0 s
Sustained decode32K32Valid1,561.649.2853 ms20.3 ms32.0046,84830.0 s
Sustained decode32K64Valid2,061.232.51,610 ms30.7 ms64.0061,82530.0 s
Sustained decode32K128Valid2,651.721.13,230 ms47.3 ms128.0079,53530.0 s
Sustained decode64K1Valid138.9140.1183 ms7.1 ms1.014,16630.0 s
Sustained decode64K2Valid257.7129.8240 ms7.7 ms2.027,73130.0 s
Sustained decode64K4Valid456.4114.1344 ms8.8 ms4.0013,69230.0 s
Sustained decode64K8Valid721.390.2556 ms11.1 ms8.0021,64030.0 s
Sustained decode64K16Valid1,066.566.8964 ms15.0 ms16.0031,99530.0 s
Sustained decode64K32Valid1,480.646.51,822 ms21.5 ms32.0044,41630.0 s
Sustained decode64K64Valid1,900.130.03,494 ms33.3 ms64.0056,95930.0 s
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid131.8355.3365 ms4.5 ms1.013,95430.0 s
Sustained decode128K2Valid244.9124.8425 ms8.0 ms2.027,34830.0 s
Sustained decode128K4Valid435.1108.9708 ms9.2 ms4.0013,05230.0 s
Sustained decode128K8Valid676.484.51,172 ms11.8 ms8.0020,28830.0 s
Sustained decode128K16Valid981.261.42,080 ms16.3 ms16.0029,42430.0 s
Sustained decode128K32Valid1,308.641.03,954 ms24.4 ms32.0039,23230.0 s
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode256K1Valid122.5122.8759 ms8.1 ms1.013,61829.5 s
Sustained decode256K2Valid228.5114.41,083 ms8.7 ms2.006,85430.0 s
Sustained decode256K4Valid403.5100.81,615 ms9.9 ms4.0012,10430.0 s
Sustained decode256K8Valid603.875.52,711 ms13.2 ms8.0018,11230.0 s
Sustained decode256K16Valid844.252.94,932 ms18.9 ms16.0025,31230.0 s
Sustained decode256K32Over capacity————————
Sustained decode256K64Over capacity————————
Sustained decode256K128Over capacity————————
Sustained decode512K1Valid105.3105.71,786 ms9.5 ms1.003,15930.0 s
Sustained decode512K2Valid198.299.32,592 ms10.1 ms2.005,94430.0 s
Sustained decode512K4Valid352.388.03,931 ms11.4 ms4.0010,56830.0 s
Sustained decode512K8Valid496.262.46,572 ms16.0 ms8.0014,88030.0 s
Sustained decode512K16Over capacity————————
Sustained decode512K32Over capacity————————
Sustained decode512K64Over capacity————————
Sustained decode512K128Over capacity————————
Sustained decode1019K1Valid82.383.03,930 ms12.1 ms1.002,47030.0 s
Sustained decode1019K2Valid156.578.65,565 ms12.7 ms2.004,69630.0 s
Sustained decode1019K4Valid279.770.18,749 ms14.3 ms4.008,38830.0 s
Sustained decode1019K8Over capacity————————
Sustained decode1019K16Over capacity————————
Sustained decode1019K32Over capacity————————
Sustained decode1019K64Over capacity————————
Sustained decode1019K128Over capacity————————

Over capacity: not run because this many requests at this context would exceed the server's KV cache.