शुरुआत करेंलीडरबोर्डDecode calculatorमॉडलOrganizationsReportsMatrix runsहार्डवेयरबेंचमार्कमार्केटप्लेसरेंटलProAPI दस्तावेज़
भाषा
Actual Computer — Every computer, one endpoint

Speed test matrix

gemma-4-12B-it on H200 NVL 141GB

vllm · 0.27.2rc1.dev113+g5cecfc013 · 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
0.27.2rc1.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-4-12B-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 262144 --gpu-memory-utilization 0.849406 --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 context19.4K423 ms TTFT
16K context18.3K887 ms TTFT
32K context16.7K1,933 ms TTFT
64K context13.7K4,721 ms TTFT
128K context9,78513,182 ms TTFT
251K context6,33739,894 ms TTFT
256K context6,27341,104 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K106.166 ms TTFT210.0111 ms TTFT408.1152 ms TTFT778.4186 ms TTFT1,426.5277 ms TTFT2,370.3402 ms TTFT3,491.4779 ms TTFTOver capacity
16K105.5106 ms TTFT208.9154 ms TTFT403.5202 ms TTFT762.3241 ms TTFT1,375.3326 ms TTFT2,229.4605 ms TTFT3,186.71,099 ms TTFTOver capacity
32K104.6151 ms TTFT206.2205 ms TTFT394.7274 ms TTFT730.6398 ms TTFT1,279.8547 ms TTFT1,988.91,027 ms TTFTOver capacityOver capacity
64K103.3216 ms TTFT201.4319 ms TTFT378.1378 ms TTFT676.7694 ms TTFT1,126.11,201 ms TTFTOver capacityOver capacityOver capacity
128K100.5445 ms TTFT192.7597 ms TTFT349.8911 ms TTFT588.41,365 ms TTFTOver capacityOver capacityOver capacityOver capacity
251K95.8998 ms TTFT178.21,393 ms TTFT305.02,110 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
Prefill8K1Valid19.4K—423 ms——1——
Prefill16K1Valid18.3K—887 ms——1——
Prefill32K1Valid16.7K—1,933 ms——1——
Prefill64K1Valid13.7K—4,721 ms——1——
Prefill128K1Valid9,785.0—13,182 ms——1——
Prefill251K1Valid6,337.0—39,894 ms——1——
Prefill256K1Valid6,273.0—41,104 ms——1——
Sustained decode8K1Valid106.1106.266 ms9.4 ms1.003,18430.0 s
Sustained decode8K2Valid210.0105.1111 ms9.5 ms2.006,29930.0 s
Sustained decode8K4Valid408.1102.2152 ms9.8 ms4.0012,24430.0 s
Sustained decode8K8Valid778.497.6186 ms10.2 ms8.0023,36830.0 s
Sustained decode8K16Valid1,426.589.6277 ms11.2 ms16.0042,78630.0 s
Sustained decode8K32Valid2,370.374.4402 ms13.4 ms32.0071,10230.0 s
Sustained decode8K64Valid3,491.454.9779 ms18.2 ms64.00104,78630.0 s
Sustained decode8K128Over capacity————————
Sustained decode16K1Valid105.5105.6106 ms9.5 ms1.003,16430.0 s
Sustained decode16K2Valid208.9104.6154 ms9.6 ms2.006,26630.0 s
Sustained decode16K4Valid403.5101.0202 ms9.9 ms4.0012,10130.0 s
Sustained decode16K8Valid762.395.5241 ms10.5 ms8.0022,86430.0 s
Sustained decode16K16Valid1,375.386.2326 ms11.6 ms16.0041,26030.0 s
Sustained decode16K32Valid2,229.469.9605 ms14.3 ms32.0066,88030.0 s
Sustained decode16K64Valid3,186.750.11,099 ms20.0 ms64.0095,61630.0 s
Sustained decode16K128Over capacity————————
Sustained decode32K1Valid104.6104.8151 ms9.5 ms1.003,13730.0 s
Sustained decode32K2Valid206.2103.3205 ms9.7 ms2.006,18630.0 s
Sustained decode32K4Valid394.798.8274 ms10.1 ms4.0011,84030.0 s
Sustained decode32K8Valid730.691.5398 ms10.9 ms8.0021,91730.0 s
Sustained decode32K16Valid1,279.880.0547 ms12.5 ms16.0038,38430.0 s
Sustained decode32K32Valid1,988.962.11,027 ms16.1 ms32.0059,64930.0 s
Sustained decode32K64Over capacity————————
Sustained decode32K128Over capacity————————
Sustained decode64K1Valid103.3103.5216 ms9.7 ms1.003,09730.0 s
Sustained decode64K2Valid201.4100.9319 ms9.9 ms2.006,04430.0 s
Sustained decode64K4Valid378.194.6378 ms10.6 ms4.0011,34430.0 s
Sustained decode64K8Valid676.784.4694 ms11.8 ms8.0020,29830.0 s
Sustained decode64K16Valid1,126.169.91,201 ms14.3 ms16.0033,77530.0 s
Sustained decode64K32Over capacity————————
Sustained decode64K64Over capacity————————
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid100.5100.8445 ms9.9 ms1.003,01530.0 s
Sustained decode128K2Valid192.796.6597 ms10.4 ms2.005,78030.0 s
Sustained decode128K4Valid349.887.4911 ms11.4 ms4.0010,49630.0 s
Sustained decode128K8Valid588.473.41,365 ms13.6 ms8.0017,66430.0 s
Sustained decode128K16Over capacity————————
Sustained decode128K32Over capacity————————
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode251K1Valid95.896.3998 ms10.4 ms1.002,87430.0 s
Sustained decode251K2Valid178.289.51,393 ms11.2 ms2.005,34630.0 s
Sustained decode251K4Valid305.076.42,110 ms13.1 ms4.009,14830.0 s
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.