Premiers pasClassementDecode calculatorModèlesOrganizationsReportsMatrix runsMatérielÉvaluations comparativesMarketplaceLocationsProDocs API
Langue
Actual Computer — Every computer, one endpoint

Speed test matrix

Qwen3.8-Flash-Next-FP8 on H200 NVL 141GB ×2

vllm · 0.1.1.dev5+ge77daef89 · FP86 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
GPU count
2
VRAM
282 GB (2 × 141 GB)
OS
linux

Engine

Engine
vllm
Version
0.1.1.dev5+ge77daef89
Quantization
FP8
Backend
cuda
Build
cuda-13.2
Commit
e77daef89e18

Run

Served model
qwen
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 VLLM_PLE_CPU_OFFLOAD=1 "$HOME/env_h200-qwen-flash-next-vllm/bin/vllm" serve Qwen/Qwen3.8-Flash-Next-FP8 --served-model-name qwen --tensor-parallel-size 2 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.858381 --host 0.0.0.0 --port 8000 --api-key <redacted> --max-num-seqs 256 --enable-prefix-caching --no-enable-flashinfer-autotune --enable-expert-parallel --moe-backend triton

Prefill

prompt tok/s by context length
8K context19.6K420 ms TTFT
16K context21K773 ms TTFT
32K context21.3K1,516 ms TTFT
64K context21.4K3,007 ms TTFT
128K context20.2K6,384 ms TTFT
251K context18.1K13,908 ms TTFT
256K context17.9K14,384 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K111.9456 ms TTFT209.8165 ms TTFT390.2270 ms TTFT705.9339 ms TTFT944.5611 ms TTFT1,554.4961 ms TTFT2,767.31,504 ms TTFT4,103.22,363 ms TTFT
16K111.7476 ms TTFT209.3234 ms TTFT388.8354 ms TTFT707.5481 ms TTFT925.4834 ms TTFT1,550.01,221 ms TTFT2,646.41,816 ms TTFT4,197.83,285 ms TTFT
32K111.2482 ms TTFT211.0544 ms TTFT388.3321 ms TTFT710.5551 ms TTFT922.1993 ms TTFT1,554.71,608 ms TTFT2,595.22,443 ms TTFT4,077.24,565 ms TTFT
64K112.6502 ms TTFT208.4561 ms TTFT397.4332 ms TTFT705.1467 ms TTFT902.0800 ms TTFT1,546.41,392 ms TTFT2,918.12,794 ms TTFT4,229.45,317 ms TTFT
128K110.4608 ms TTFT206.9692 ms TTFT380.8641 ms TTFT697.81,116 ms TTFT938.21,857 ms TTFT1,524.13,427 ms TTFT2,677.56,794 ms TTFT3,948.212,362 ms TTFT
251K108.91,084 ms TTFT207.61,208 ms TTFT386.61,541 ms TTFT694.42,443 ms TTFT882.53,837 ms TTFT1,444.37,170 ms TTFT2,455.414,780 ms TTFT3,762.131,628 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
Prefill8K1Valid19.6K—420 ms——1——
Prefill16K1Valid21K—773 ms——1——
Prefill32K1Valid21.3K—1,516 ms——1——
Prefill64K1Valid21.4K—3,007 ms——1——
Prefill128K1Valid20.2K—6,384 ms——1——
Prefill251K1Valid18.1K—13,908 ms——1——
Prefill256K1Valid17.9K—14,384 ms——1——
Sustained decode8K1Valid111.9112.0456 ms8.9 ms1.003,35930.0 s
Sustained decode8K2Valid209.8104.8165 ms9.5 ms2.006,29430.0 s
Sustained decode8K4Valid390.297.6270 ms10.2 ms4.0011,70430.0 s
Sustained decode8K8Valid705.988.2339 ms11.3 ms8.0021,18530.0 s
Sustained decode8K16Valid944.558.9611 ms17.0 ms16.0028,33630.0 s
Sustained decode8K32Valid1,554.449.0961 ms20.4 ms32.0046,62430.0 s
Sustained decode8K64Valid2,767.343.81,504 ms22.9 ms64.0083,00830.0 s
Sustained decode8K128Valid4,103.231.72,363 ms31.6 ms128.00123,13630.0 s
Sustained decode16K1Valid111.7111.7476 ms8.9 ms1.003,35030.0 s
Sustained decode16K2Valid209.3104.7234 ms9.6 ms2.006,28030.0 s
Sustained decode16K4Valid388.897.5354 ms10.3 ms4.0011,66530.0 s
Sustained decode16K8Valid707.588.4481 ms11.3 ms8.0021,22430.0 s
Sustained decode16K16Valid925.458.1834 ms17.2 ms16.0027,76030.0 s
Sustained decode16K32Valid1,550.048.61,221 ms20.6 ms32.0046,49630.0 s
Sustained decode16K64Valid2,646.441.21,816 ms24.3 ms64.0079,36030.0 s
Sustained decode16K128Valid4,197.831.83,285 ms31.5 ms128.00125,95230.0 s
Sustained decode32K1Valid111.2111.3482 ms9.0 ms1.003,33530.0 s
Sustained decode32K2Valid211.0105.6544 ms9.5 ms2.006,33030.0 s
Sustained decode32K4Valid388.397.1321 ms10.3 ms4.0011,64730.0 s
Sustained decode32K8Valid710.588.5551 ms11.3 ms8.0021,31230.0 s
Sustained decode32K16Valid922.157.4993 ms17.4 ms16.0027,66430.0 s
Sustained decode32K32Valid1,554.748.41,608 ms20.7 ms32.0046,62430.0 s
Sustained decode32K64Valid2,595.239.12,443 ms25.6 ms64.0077,82430.0 s
Sustained decode32K128Valid4,077.229.24,565 ms34.2 ms128.00122,36730.0 s
Sustained decode64K1Valid112.6113.1502 ms8.8 ms1.003,37830.0 s
Sustained decode64K2Valid208.4104.3561 ms9.6 ms2.006,25230.0 s
Sustained decode64K4Valid397.499.2332 ms10.1 ms4.0011,92030.0 s
Sustained decode64K8Valid705.187.9467 ms11.4 ms8.0021,15230.0 s
Sustained decode64K16Valid902.056.1800 ms17.8 ms16.0027,05630.0 s
Sustained decode64K32Valid1,546.448.11,392 ms20.8 ms32.0046,36830.0 s
Sustained decode64K64Valid2,918.143.82,794 ms22.8 ms64.0087,55230.0 s
Sustained decode64K128Valid4,229.430.55,317 ms32.8 ms128.00126,84830.0 s
Sustained decode128K1Valid110.4111.5608 ms9.0 ms1.003,31530.0 s
Sustained decode128K2Valid206.9104.5692 ms9.6 ms2.006,20630.0 s
Sustained decode128K4Valid380.895.2641 ms10.5 ms4.0011,42530.0 s
Sustained decode128K8Valid697.886.71,116 ms11.5 ms8.0020,92830.0 s
Sustained decode128K16Valid938.257.81,857 ms17.3 ms16.0028,14430.0 s
Sustained decode128K32Valid1,524.146.93,427 ms21.3 ms32.0045,72830.0 s
Sustained decode128K64Valid2,677.540.46,794 ms24.8 ms64.0080,32030.0 s
Sustained decode128K128Valid3,948.229.212,362 ms34.2 ms128.00118,40030.0 s
Sustained decode251K1Valid108.9109.31,084 ms9.2 ms1.003,26730.0 s
Sustained decode251K2Valid207.6103.81,208 ms9.6 ms2.006,22630.0 s
Sustained decode251K4Valid386.695.91,541 ms10.4 ms4.0011,59630.0 s
Sustained decode251K8Valid694.485.52,443 ms11.7 ms8.0020,83230.0 s
Sustained decode251K16Valid882.553.43,837 ms18.7 ms16.0026,46430.0 s
Sustained decode251K32Valid1,444.341.47,170 ms24.2 ms32.0043,32830.0 s
Sustained decode251K64Valid2,455.430.914,780 ms32.3 ms64.0073,66530.0 s
Sustained decode251K128Valid3,762.121.031,628 ms47.6 ms128.00112,89630.0 s