Primeros pasosClasificaciónDecode calculatorModelosOrganizationsReportsMatrix runsHardwareEvaluaciones comparativasMarketplaceAlquileresProDocs de API
Idioma
Actual Computer — Every computer, one endpoint

Speed test matrix

NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 on H200 NVL 141GB

sglang · 0.5.6.post3.dev9335+gdd15fb57b · BF168 contexts (8K–1019K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 256K, 512K, 1019K, 1024K
by @keennaylmx dev importMeasured

Hardware

Type
Discrete GPU
GPU
H200 NVL
VRAM
141 GB
OS
linux

Engine

Engine
sglang
Version
0.5.6.post3.dev9335+gdd15fb57b
Quantization
BF16
Backend
cuda
Build
cuda-13.2
Commit
dd15fb57b5ef

Run

Served model
nemotron
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

env SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 "$HOME/env_h200-nvidia-sglang/bin/sglang" serve --model-path nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 --served-model-name nemotron --trust-remote-code --tp 1 --reasoning-parser nemotron_3 --tool-call-parser qwen3_coder --context-length 1048576 --mem-fraction-static 0.862866 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted> --mamba-ssm-dtype float16 --cuda-graph-max-bs-decode 16

Prefill

prompt tok/s by context length
8K context18.5K443 ms TTFT
16K context19.7K823 ms TTFT
32K context19.8K1,636 ms TTFT
64K context19.3K3,341 ms TTFT
128K context17.7K7,291 ms TTFT
256K context14.8K17,342 ms TTFT
512K context11.1K46,527 ms TTFT
1019K context7,259141,030 ms TTFT
1024K context7,233142,240 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K282.148 ms TTFT439.161 ms TTFT729.092 ms TTFT1,023.6132 ms TTFT956.1185 ms TTFT1,792.4320 ms TTFT2,959.6627 ms TTFT4,501.91,134 ms TTFT
16K279.774 ms TTFT70.5121 ms TTFT155.6132 ms TTFT888.1237 ms TTFT1,556.8344 ms TTFT1,807.3655 ms TTFT2,816.01,251 ms TTFT4,224.62,498 ms TTFT
32K276.898 ms TTFT66.3147 ms TTFT147.6225 ms TTFT724.2376 ms TTFT1,300.5768 ms TTFT1,734.61,404 ms TTFT2,487.82,761 ms TTFT3,523.45,521 ms TTFT
64K272.6140 ms TTFT57.2265 ms TTFT117.3431 ms TTFT315.5783 ms TTFT927.21,515 ms TTFT1,498.92,965 ms TTFT2,147.55,989 ms TTFTOver capacity
128K263.2228 ms TTFT67.4411 ms TTFT144.2749 ms TTFT643.01,462 ms TTFT1,074.42,843 ms TTFT1,283.65,583 ms TTFTOver capacityOver capacity
256K246.6405 ms TTFT87.0798 ms TTFT292.51,480 ms TTFT788.92,838 ms TTFT1,134.95,600 ms TTFTOver capacityOver capacityOver capacity
512K224.1798 ms TTFT165.11,548 ms TTFT505.62,948 ms TTFT521.35,908 ms TTFTOver capacityOver capacityOver capacityOver capacity
1019K183.61,647 ms TTFT165.53,183 ms TTFT410.96,102 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
Prefill8K1Valid18.5K—443 ms——1——
Prefill16K1Valid19.7K—823 ms——1——
Prefill32K1Valid19.8K—1,636 ms——1——
Prefill64K1Valid19.3K—3,341 ms——1——
Prefill128K1Valid17.7K—7,291 ms——1——
Prefill256K1Valid14.8K—17,342 ms——1——
Prefill512K1Valid11.1K—46,527 ms——1——
Prefill1019K1Valid7,259.0—141,030 ms——1——
Prefill1024K1Valid7,233.0—142,240 ms——1——
Sustained decode8K1Valid282.1283.148 ms3.5 ms1.028,46330.0 s
Sustained decode8K2Valid439.1215.561 ms4.7 ms2.2213,17130.0 s
Sustained decode8K4Valid729.0180.092 ms5.6 ms4.0421,87130.0 s
Sustained decode8K8Valid1,023.6141.4132 ms7.2 ms9.3830,70630.0 s
Sustained decode8K16Valid956.159.6185 ms16.8 ms24.0028,67230.0 s
Sustained decode8K32Valid1,792.456.1320 ms17.8 ms52.5053,76030.0 s
Sustained decode8K64Valid2,959.646.6627 ms21.4 ms96.0088,76830.0 s
Sustained decode8K128Valid4,501.935.01,134 ms28.6 ms199.50135,04030.0 s
Sustained decode16K1Valid279.7281.174 ms3.6 ms1.028,39230.0 s
Sustained decode16K2Valid70.535.2121 ms28.4 ms189.702,11430.0 s
Sustained decode16K4Valid155.638.9132 ms25.7 ms134.004,66830.0 s
Sustained decode16K8Valid888.1106.5237 ms9.4 ms11.8026,64030.0 s
Sustained decode16K16Valid1,556.896.6344 ms10.4 ms16.0046,70430.0 s
Sustained decode16K32Valid1,807.356.2655 ms17.8 ms35.2054,22430.0 s
Sustained decode16K64Valid2,816.044.31,251 ms22.6 ms96.0084,48030.0 s
Sustained decode16K128Valid4,224.632.82,498 ms30.5 ms203.70126,72030.0 s
Sustained decode32K1Valid276.8278.698 ms3.6 ms1.028,30430.0 s
Sustained decode32K2Valid66.333.2147 ms30.2 ms194.001,98830.0 s
Sustained decode32K4Valid147.636.6225 ms27.3 ms134.004,42830.0 s
Sustained decode32K8Valid724.283.9376 ms11.9 ms34.4021,72830.0 s
Sustained decode32K16Valid1,300.578.7768 ms12.7 ms18.9039,00830.0 s
Sustained decode32K32Valid1,734.654.61,404 ms18.3 ms41.6052,03230.0 s
Sustained decode32K64Valid2,487.840.02,761 ms25.0 ms96.0074,62430.0 s
Sustained decode32K128Valid3,523.428.95,521 ms34.6 ms210.00105,72830.0 s
Sustained decode64K1Valid272.6275.0140 ms3.6 ms1.028,17730.0 s
Sustained decode64K2Valid57.228.8265 ms34.7 ms194.001,71630.0 s
Sustained decode64K4Valid117.329.2431 ms34.2 ms166.403,51630.0 s
Sustained decode64K8Valid315.538.5783 ms26.0 ms121.809,46430.0 s
Sustained decode64K16Valid927.258.71,515 ms17.0 ms27.0027,80830.0 s
Sustained decode64K32Valid1,498.948.22,965 ms20.8 ms48.8044,96030.0 s
Sustained decode64K64Valid2,147.535.95,989 ms27.8 ms96.0064,38430.0 s
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid263.2267.0228 ms3.7 ms1.027,89630.0 s
Sustained decode128K2Valid67.433.7411 ms29.7 ms92.802,02230.0 s
Sustained decode128K4Valid144.236.2749 ms27.7 ms70.004,32430.0 s
Sustained decode128K8Valid643.076.11,462 ms13.1 ms17.0019,28830.0 s
Sustained decode128K16Valid1,074.467.52,843 ms14.8 ms20.3032,22430.0 s
Sustained decode128K32Valid1,283.643.05,583 ms23.2 ms47.8038,49630.0 s
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode256K1Valid246.6253.5405 ms3.9 ms0.927,39930.0 s
Sustained decode256K2Valid87.043.7798 ms22.9 ms34.002,60830.0 s
Sustained decode256K4Valid292.570.41,480 ms14.2 ms23.008,77630.0 s
Sustained decode256K8Valid788.999.92,838 ms10.0 ms9.3023,66430.0 s
Sustained decode256K16Valid1,134.972.85,600 ms13.7 ms16.1034,04830.0 s
Sustained decode256K32Over capacity————————
Sustained decode256K64Over capacity————————
Sustained decode256K128Over capacity————————
Sustained decode512K1Valid224.1230.1798 ms4.3 ms1.016,72230.0 s
Sustained decode512K2Valid165.179.61,548 ms12.6 ms13.904,95230.0 s
Sustained decode512K4Valid505.6131.22,948 ms7.6 ms4.0414,91429.5 s
Sustained decode512K8Valid521.370.75,908 ms14.2 ms12.0015,64030.0 s
Sustained decode512K16Over capacity————————
Sustained decode512K32Over capacity————————
Sustained decode512K64Over capacity————————
Sustained decode512K128Over capacity————————
Sustained decode1019K1Valid183.6194.31,647 ms5.1 ms1.015,50830.0 s
Sustained decode1019K2Valid165.580.63,183 ms12.4 ms8.104,96430.0 s
Sustained decode1019K4Valid410.9106.76,102 ms9.4 ms4.0012,32830.0 s
Sustained decode1019K8Over capacity————————
Sustained decode1019K16Over capacity————————
Sustained decode1019K32Over capacity————————
Sustained decode1019K64Over capacity————————
Sustained decode1019K128Over capacity————————

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