Primeiros passosRankingDecode calculatorModelosOrganizationsReportsMatrix runsHardwareAvaliações comparativasMarketplaceAluguéisProDocs da API
Idioma
Actual Computer — Every computer, one endpoint

Speed test matrix

Nanbeige4.2-3B on H200 NVL 141GB

sglang · 0.0.0.dev17517+g3e59d89e5 · 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
sglang
Version
0.0.0.dev17517+g3e59d89e5
Quantization
BF16
Backend
cuda
Build
cuda-13.2
Commit
3e59d89e5349

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-sglang/bin/sglang" serve --model-path Nanbeige/Nanbeige4.2-3B --served-model-name nanbeige --tp 1 --reasoning-parser nanbeige --tool-call-parser nanbeige --context-length 262144 --mem-fraction-static 0.855920 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted> --speculative-algorithm DSPARK --speculative-draft-model-path Nanbeige/Nanbeige4.2-3B-DSpark --speculative-dspark-block-size 7

Prefill

prompt tok/s by context length
8K context26.6K308 ms TTFT
16K context22.8K712 ms TTFT
32K context17K1,900 ms TTFT
64K context11K5,868 ms TTFT
128K context6,36820,158 ms TTFT
251K context3,52571,347 ms TTFT
256K context3,48773,578 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K310.848 ms TTFT153.1126 ms TTFT997.884 ms TTFT1,823.672 ms TTFT2,153.4155 ms TTFT3,195.7371 ms TTFTOver capacityOver capacity
16K291.374 ms TTFT318.095 ms TTFT908.599 ms TTFT1,515.1159 ms TTFT1,857.5365 ms TTFTOver capacityOver capacityOver capacity
32K267.3100 ms TTFT397.6124 ms TTFT794.8162 ms TTFT1,114.5256 ms TTFTOver capacityOver capacityOver capacityOver capacity
64K218.7149 ms TTFT177.6275 ms TTFT566.8308 ms TTFT484.7754 ms TTFTOver capacityOver capacityOver capacityOver capacity
128K182.9308 ms TTFT122.6561 ms TTFT251.51,034 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacity
251K120.7723 ms TTFT98.81,414 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
Prefill8K1Valid26.6K—308 ms——1——
Prefill16K1Valid22.8K—712 ms——1——
Prefill32K1Valid17K—1,900 ms——1——
Prefill64K1Valid11K—5,868 ms——1——
Prefill128K1Valid6,368.0—20,158 ms——1——
Prefill251K1Valid3,525.0—71,347 ms——1——
Prefill256K1Valid3,487.0—73,578 ms——1——
Sustained decode8K1Valid310.8307.748 ms3.2 ms1.029,32230.0 s
Sustained decode8K2Valid153.177.3126 ms13.0 ms3.004,59230.0 s
Sustained decode8K4Valid997.8233.984 ms4.3 ms4.5429,92730.0 s
Sustained decode8K8Valid1,823.6228.872 ms4.4 ms8.01154,70930.0 s
Sustained decode8K16Valid2,153.4149.7155 ms6.7 ms17.41664,58430.0 s
Sustained decode8K32Valid3,195.795.0371 ms10.5 ms33.83095,85030.0 s
Sustained decode8K64Over capacity————————
Sustained decode8K128Over capacity————————
Sustained decode16K1Valid291.3299.174 ms3.3 ms1.028,73630.0 s
Sustained decode16K2Valid318.0215.395 ms5.5 ms19.529,54030.0 s
Sustained decode16K4Valid908.5228.699 ms4.4 ms4.5427,25130.0 s
Sustained decode16K8Valid1,515.1189.0159 ms5.3 ms7.9845,44330.0 s
Sustained decode16K16Valid1,857.5112.0365 ms8.9 ms18.1355,71130.0 s
Sustained decode16K32Over capacity————————
Sustained decode16K64Over capacity————————
Sustained decode16K128Over capacity————————
Sustained decode32K1Valid267.3268.4100 ms3.7 ms1.028,01630.0 s
Sustained decode32K2Valid397.6204.4124 ms5.0 ms3.5211,92630.0 s
Sustained decode32K4Valid794.8193.3162 ms5.2 ms4.3423,83730.0 s
Sustained decode32K8Valid1,114.5143.1256 ms7.0 ms8.7833,43530.0 s
Sustained decode32K16Over capacity————————
Sustained decode32K32Over capacity————————
Sustained decode32K64Over capacity————————
Sustained decode32K128Over capacity————————
Sustained decode64K1Valid218.7221.2149 ms4.5 ms1.016,56130.0 s
Sustained decode64K2Valid177.690.6275 ms11.0 ms9.405,32830.0 s
Sustained decode64K4Valid566.8148.8308 ms6.8 ms4.3417,00030.0 s
Sustained decode64K8Valid484.761.7754 ms16.2 ms12.0014,52930.0 s
Sustained decode64K16Over capacity————————
Sustained decode64K32Over capacity————————
Sustained decode64K64Over capacity————————
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid182.9183.0308 ms5.5 ms1.015,48630.0 s
Sustained decode128K2Valid122.661.9561 ms16.2 ms10.003,67630.0 s
Sustained decode128K4Valid251.563.31,034 ms15.8 ms5.407,54030.0 s
Sustained decode128K8Over capacity————————
Sustained decode128K16Over capacity————————
Sustained decode128K32Over capacity————————
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode251K1Valid120.7174.2723 ms6.2 ms1.013,62030.0 s
Sustained decode251K2Valid98.851.81,414 ms19.4 ms5.902,96330.0 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.