入門指南排行榜Decode calculator模型OrganizationsReportsMatrix runs硬體基準評測市集租用ProAPI文件
語言
Actual Computer — Every computer, one endpoint

Speed test matrix

gemma-4-E2B-it on H200 NVL 141GB

vllm · 0.27.2rc1.dev113+g5cecfc013 · BF165 contexts (8K–123K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 123K, 128K
by @keennaylmx dev importMeasured

Hardware

Type
Discrete GPU
GPU
H200 NVL
VRAM
141 GB
OS
linux

Engine

Engine
vllm
Version
0.27.2rc1.dev113+g5cecfc013
Quantization
BF16
Backend
cuda
Build
cuda-13.2
Commit
5cecfc013750

Run

Served model
gemma
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-redhatai-vllm/bin/vllm" serve RedHatAI/gemma-4-E2B-it --served-model-name gemma --trust-remote-code --tensor-parallel-size 1 --reasoning-parser gemma4 --enable-auto-tool-choice --tool-call-parser gemma4 --max-model-len 131072 --gpu-memory-utilization 0.854015 --host 0.0.0.0 --port 8000 --api-key <redacted> --reasoning-parser gemma4 --limit-mm-per-prompt '{"image":4,"audio":1}' --async-scheduling

Prefill

prompt tok/s by context length
8K context62K132 ms TTFT
16K context56.8K286 ms TTFT
32K context51.4K629 ms TTFT
64K context41.1K1,571 ms TTFT
123K context27.6K4,492 ms TTFT
128K context26.4K4,884 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K260.371 ms TTFT514.087 ms TTFT1,001.5113 ms TTFT1,903.3134 ms TTFT3,529.7180 ms TTFT5,982.0248 ms TTFT9,049.3310 ms TTFT13.1K1,051 ms TTFT
16K256.197 ms TTFT507.0116 ms TTFT975.3118 ms TTFT1,809.1207 ms TTFT3,243.9205 ms TTFT5,150.8268 ms TTFT7,255.1869 ms TTFT9,854.81,600 ms TTFT
32K250.1145 ms TTFT489.5180 ms TTFT923.6233 ms TTFT1,650.8247 ms TTFT2,761.0208 ms TTFT4,029.2302 ms TTFT5,330.11,784 ms TTFT6,468.13,480 ms TTFT
64K242.1211 ms TTFT463.6289 ms TTFT839.3268 ms TTFT1,394.8279 ms TTFT2,077.7302 ms TTFT2,875.22,018 ms TTFT3,419.33,670 ms TTFT3,796.17,401 ms TTFT
123K227.1473 ms TTFT415.8541 ms TTFT709.4517 ms TTFT1,063.6573 ms TTFT1,497.92,692 ms TTFT1,859.24,997 ms TTFT2,047.68,685 ms TTFT2,151.016,966 ms TTFT

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
Prefill8K1Valid62K—132 ms——1——
Prefill16K1Valid56.8K—286 ms——1——
Prefill32K1Valid51.4K—629 ms——1——
Prefill64K1Valid41.1K—1,571 ms——1——
Prefill123K1Valid27.6K—4,492 ms——1——
Prefill128K1Valid26.4K—4,884 ms——1——
Sustained decode8K1Valid260.3263.971 ms3.8 ms1.027,80930.0 s
Sustained decode8K2Valid514.0260.987 ms3.8 ms2.0415,41930.0 s
Sustained decode8K4Valid1,001.5256.7113 ms3.9 ms4.0830,04330.0 s
Sustained decode8K8Valid1,903.3243.4134 ms4.1 ms8.01657,09630.0 s
Sustained decode8K16Valid3,529.7223.6180 ms4.5 ms15.716105,88930.0 s
Sustained decode8K32Valid5,982.0192.7248 ms5.2 ms32.032179,46230.0 s
Sustained decode8K64Valid9,049.3149.9310 ms6.7 ms63.464271,49530.0 s
Sustained decode8K128Valid13.1K104.21,051 ms9.6 ms128.00393,46630.0 s
Sustained decode16K1Valid256.1258.497 ms3.9 ms1.027,68730.0 s
Sustained decode16K2Valid507.0256.7116 ms3.9 ms2.0415,21030.0 s
Sustained decode16K4Valid975.3246.2118 ms4.1 ms4.0829,26030.0 s
Sustained decode16K8Valid1,809.1228.4207 ms4.4 ms8.0854,27430.0 s
Sustained decode16K16Valid3,243.9205.5205 ms4.9 ms16.01697,31630.0 s
Sustained decode16K32Valid5,150.8164.1268 ms6.1 ms31.732154,52630.0 s
Sustained decode16K64Valid7,255.1114.9869 ms8.7 ms64.00217,61730.0 s
Sustained decode16K128Valid9,854.877.51,600 ms12.9 ms128.00295,68030.0 s
Sustained decode32K1Valid250.1255.5145 ms3.9 ms1.027,50730.0 s
Sustained decode32K2Valid489.5247.8180 ms4.0 ms2.0414,68830.0 s
Sustained decode32K4Valid923.6233.9233 ms4.3 ms3.9427,70730.0 s
Sustained decode32K8Valid1,650.8209.1247 ms4.8 ms8.0849,52430.0 s
Sustained decode32K16Valid2,761.0173.7208 ms5.8 ms16.01682,82430.0 s
Sustained decode32K32Valid4,029.2127.1302 ms7.9 ms31.832120,87930.0 s
Sustained decode32K64Valid5,330.183.11,784 ms12.0 ms64.00159,87230.0 s
Sustained decode32K128Valid6,468.150.83,480 ms19.7 ms128.00194,04830.0 s
Sustained decode64K1Valid242.1245.3211 ms4.1 ms1.017,26230.0 s
Sustained decode64K2Valid463.6234.3289 ms4.3 ms2.0213,90730.0 s
Sustained decode64K4Valid839.3211.9268 ms4.7 ms3.9425,17730.0 s
Sustained decode64K8Valid1,394.8175.3279 ms5.7 ms7.9841,84030.0 s
Sustained decode64K16Valid2,077.7131.2302 ms7.6 ms15.91662,33230.0 s
Sustained decode64K32Valid2,875.289.62,018 ms11.2 ms32.0086,24030.0 s
Sustained decode64K64Valid3,419.353.23,670 ms18.8 ms64.00102,59230.0 s
Sustained decode64K128Valid3,796.129.87,401 ms33.6 ms128.00113,92030.0 s
Sustained decode123K1Valid227.1231.5473 ms4.3 ms1.016,81230.0 s
Sustained decode123K2Valid415.8212.8541 ms4.7 ms2.0212,47230.0 s
Sustained decode123K4Valid709.4180.4517 ms5.5 ms4.0421,28230.0 s
Sustained decode123K8Valid1,063.6135.6573 ms7.4 ms8.0831,90530.0 s
Sustained decode123K16Valid1,497.993.52,692 ms10.7 ms16.0044,92830.0 s
Sustained decode123K32Valid1,859.257.64,997 ms17.4 ms32.0055,77630.0 s
Sustained decode123K64Valid2,047.631.88,685 ms31.5 ms64.0061,43930.0 s
Sustained decode123K128Valid2,151.016.916,966 ms59.2 ms128.0064,51330.0 s