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

Speed test matrix

gemma-4-31B-it on H200 NVL 141GB

sglang · 0.5.6.post3.dev5791+gf7d4e44d8 · BF168 contexts (8K–1024K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K
by @keennaylmx dev importMeasured

Hardware

Type
Discrete GPU
GPU
H200 NVL
VRAM
141 GB
OS
linux

Engine

Engine
sglang
Version
0.5.6.post3.dev5791+gf7d4e44d8
Quantization
BF16
Backend
cuda
Build
cuda-13.2
Commit
f7d4e44d82ac

Run

Served model
gemma
Revision
main
Decode test
Fixed duration
Prefill mode
Measured during decode warm-up
Time per measurement
30 s
Max output
8,192 tokens

Launch command

"$HOME/env_h200-gemma-sglang/bin/sglang" serve --model-path google/gemma-4-31B-it --served-model-name gemma --trust-remote-code --tp 1 --reasoning-parser gemma4 --tool-call-parser gemma4 --context-length 262144 --mem-fraction-static 0.848736 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length
8K context4,7651,721 ms TTFT
16K context3,7584,327 ms TTFT
32K context2,70511,965 ms TTFT
64K context1,74137,091 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K44.2Requests queued115 ms TTFT87.8153 ms TTFT175.0209 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacity
16K40.9157 ms TTFT81.9212 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity
32K35.9228 ms TTFT72.5306 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity
64K28.6389 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity
128KOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity
256KOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity
512KOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity
1024KOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity

Requests queued: the server could not run all requests at once, so some waited in a queue. 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
Prefill8K1Valid4,765.0—1,721 ms——1——
Prefill16K1Valid3,758.0—4,327 ms——1——
Prefill32K1Valid2,705.0—11,965 ms——1——
Prefill64K1Valid1,741.0—37,091 ms——1——
Sustained decode8K1Requests queued44.244.2115 ms22.6 ms1.001,32430.0 s
Sustained decode8K2Valid87.843.9153 ms22.8 ms2.002,63230.0 s
Sustained decode8K4Valid175.043.8209 ms22.9 ms4.005,24630.0 s
Sustained decode8K8Over capacity————————
Sustained decode8K16Over capacity————————
Sustained decode8K32Over capacity————————
Sustained decode8K64Over capacity————————
Sustained decode8K128Over capacity————————
Sustained decode16K1Valid40.940.9157 ms24.5 ms1.001,22630.0 s
Sustained decode16K2Valid81.940.9212 ms24.4 ms2.002,45630.0 s
Sustained decode16K4Over capacity————————
Sustained decode16K8Over capacity————————
Sustained decode16K16Over capacity————————
Sustained decode16K32Over capacity————————
Sustained decode16K64Over capacity————————
Sustained decode16K128Over capacity————————
Sustained decode32K1Valid35.935.9228 ms27.8 ms1.001,07629.9 s
Sustained decode32K2Valid72.536.2306 ms27.6 ms2.002,17330.0 s
Sustained decode32K4Over capacity————————
Sustained decode32K8Over capacity————————
Sustained decode32K16Over capacity————————
Sustained decode32K32Over capacity————————
Sustained decode32K64Over capacity————————
Sustained decode32K128Over capacity————————
Sustained decode64K1Valid28.628.5389 ms35.0 ms1.0085630.0 s
Sustained decode64K2Over capacity————————
Sustained decode64K4Over capacity————————
Sustained decode64K8Over capacity————————
Sustained decode64K16Over capacity————————
Sustained decode64K32Over capacity————————
Sustained decode64K64Over capacity————————
Sustained decode64K128Over capacity————————
Sustained decode128K1Over capacity————————
Sustained decode128K2Over capacity————————
Sustained decode128K4Over capacity————————
Sustained decode128K8Over capacity————————
Sustained decode128K16Over capacity————————
Sustained decode128K32Over capacity————————
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode256K1Over capacity————————
Sustained decode256K2Over capacity————————
Sustained decode256K4Over capacity————————
Sustained decode256K8Over capacity————————
Sustained decode256K16Over capacity————————
Sustained decode256K32Over capacity————————
Sustained decode256K64Over capacity————————
Sustained decode256K128Over capacity————————
Sustained decode512K1Over capacity————————
Sustained decode512K2Over capacity————————
Sustained decode512K4Over capacity————————
Sustained decode512K8Over capacity————————
Sustained decode512K16Over capacity————————
Sustained decode512K32Over capacity————————
Sustained decode512K64Over capacity————————
Sustained decode512K128Over capacity————————
Sustained decode1024K1Over capacity————————
Sustained decode1024K2Over capacity————————
Sustained decode1024K4Over capacity————————
Sustained decode1024K8Over capacity————————
Sustained decode1024K16Over capacity————————
Sustained decode1024K32Over capacity————————
Sustained decode1024K64Over capacity————————
Sustained decode1024K128Over capacity————————

Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.