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

Speed test matrix

Nanbeige4.2-3B on H200 NVL 141GB

vllm · 0.1.dev20429+g62f6de733 · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K
by @keennaylmx dev importMeasured

Hardware

Type
Discrete GPU
GPU
H200 NVL
VRAM
141 GB
OS
linux

Engine

Engine
vllm
Version
0.1.dev20429+g62f6de733
Quantization
BF16
Backend
cuda
Build
cuda-13.2
Commit
62f6de733d7a

Run

Served model
nanbeige
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-nanbeige-vllm/bin/vllm" serve Nanbeige/Nanbeige4.2-3B --served-model-name nanbeige --tensor-parallel-size 1 --reasoning-parser nanbeige --enable-auto-tool-choice --tool-call-parser nanbeige --max-model-len 262144 --gpu-memory-utilization 0.868238 --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length
8K context29.5K179 ms TTFT
16K context24.2K432 ms TTFT
32K context20K1,034 ms TTFT
64K context14.3K2,874 ms TTFT
128K context9,0739,061 ms TTFT
251K context5,23230,778 ms TTFT
256K context5,20731,543 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K164.648 ms TTFT316.976 ms TTFT603.792 ms TTFT1,065.1110 ms TTFT1,658.8202 ms TTFT2,476.2324 ms TTFTOver capacityOver capacity
16K157.783 ms TTFT300.397 ms TTFT557.782 ms TTFT921.5254 ms TTFT1,385.0291 ms TTFTOver capacityOver capacityOver capacity
32K146.4116 ms TTFT271.3111 ms TTFT477.5178 ms TTFT741.7325 ms TTFT1,015.0534 ms TTFTOver capacityOver capacityOver capacity
64K128.7148 ms TTFT229.9210 ms TTFT378.2306 ms TTFT519.8440 ms TTFTOver capacityOver capacityOver capacityOver capacity
128K104.5255 ms TTFT176.8303 ms TTFT255.9512 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacity
251K75.9562 ms TTFT120.6763 ms TTFTOver capacityOver 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
Prefill8K1Valid29.5K—179 ms——1——
Prefill16K1Valid24.2K—432 ms——1——
Prefill32K1Valid20K—1,034 ms——1——
Prefill64K1Valid14.3K—2,874 ms——1——
Prefill128K1Valid9,073.0—9,061 ms——1——
Prefill251K1Valid5,232.0—30,778 ms——1——
Prefill256K1Valid5,207.0—31,543 ms——1——
Sustained decode8K1Valid164.6165.748 ms6.0 ms1.014,93930.0 s
Sustained decode8K2Valid316.9160.376 ms6.2 ms1.929,50630.0 s
Sustained decode8K4Valid603.7153.892 ms6.5 ms4.0418,11030.0 s
Sustained decode8K8Valid1,065.1138.1110 ms7.2 ms8.0831,95230.0 s
Sustained decode8K16Valid1,658.8105.8202 ms9.4 ms16.0049,76030.0 s
Sustained decode8K32Valid2,476.279.1324 ms12.6 ms32.0074,26730.0 s
Sustained decode8K64Over capacity————————
Sustained decode8K128Over capacity————————
Sustained decode16K1Valid157.7159.083 ms6.3 ms1.014,73130.0 s
Sustained decode16K2Valid300.3151.997 ms6.6 ms2.029,01330.0 s
Sustained decode16K4Valid557.7141.882 ms7.1 ms4.0416,73030.0 s
Sustained decode16K8Valid921.5116.3254 ms8.6 ms8.0027,63730.0 s
Sustained decode16K16Valid1,385.087.7291 ms11.4 ms16.0041,55230.0 s
Sustained decode16K32Over capacity————————
Sustained decode16K64Over capacity————————
Sustained decode16K128Over capacity————————
Sustained decode32K1Valid146.4148.3116 ms6.7 ms1.014,39330.0 s
Sustained decode32K2Valid271.3137.9111 ms7.3 ms2.028,13830.0 s
Sustained decode32K4Valid477.5127.8178 ms7.9 ms4.0414,32330.0 s
Sustained decode32K8Valid741.793.2325 ms10.7 ms8.0022,24830.0 s
Sustained decode32K16Valid1,015.064.1534 ms15.6 ms16.0030,44430.0 s
Sustained decode32K32Over capacity————————
Sustained decode32K64Over capacity————————
Sustained decode32K128Over capacity————————
Sustained decode64K1Valid128.7132.1148 ms7.6 ms1.013,86030.0 s
Sustained decode64K2Valid229.9115.3210 ms8.7 ms2.006,89630.0 s
Sustained decode64K4Valid378.294.9306 ms10.5 ms4.0011,34430.0 s
Sustained decode64K8Valid519.865.2440 ms15.3 ms8.0015,59230.0 s
Sustained decode64K16Over capacity————————
Sustained decode64K32Over capacity————————
Sustained decode64K64Over capacity————————
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid104.5104.9255 ms9.5 ms1.003,13530.0 s
Sustained decode128K2Valid176.888.6303 ms11.3 ms2.005,30630.0 s
Sustained decode128K4Valid255.964.2512 ms15.6 ms4.007,67630.0 s
Sustained decode128K8Over capacity————————
Sustained decode128K16Over capacity————————
Sustained decode128K32Over capacity————————
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode251K1Valid75.976.2562 ms13.1 ms1.002,27730.0 s
Sustained decode251K2Valid120.660.6763 ms16.5 ms2.003,61830.0 s
Sustained decode251K4Over capacity————————
Sustained decode251K8Over capacity————————
Sustained decode251K16Over capacity————————
Sustained decode251K32Over capacity————————
Sustained decode251K64Over capacity————————
Sustained decode251K128Over capacity————————

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