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

Speed test matrix

Qwen3.8-Flash-Next on H200 SXM 141GB ×4

vllm · 0.1.dev1+g50a061f79 · 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 SXM
GPU count
4
VRAM
564 GB (4 × 141 GB)
OS
linux

Engine

Engine
vllm
Version
0.1.dev1+g50a061f79
Quantization
BF16
Backend
cuda
Build
cuda-13.2
Commit
50a061f792f3

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_MMAP=1 "$HOME/env_h200-qwen-flash-next-vllm-pr-54129/bin/vllm" serve Qwen/Qwen3.8-Flash-Next --served-model-name qwen --tensor-parallel-size 4 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.84 --host 0.0.0.0 --port 8000 --api-key <redacted> --enable-expert-parallel

Prefill

prompt tok/s by context length
8K context25.2K325 ms TTFT
16K context29.5K547 ms TTFT
32K context30.2K1,065 ms TTFT
64K context30.9K2,075 ms TTFT
128K context31K4,136 ms TTFT
251K context29.6K8,481 ms TTFT
256K context28.4K9,021 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K137.1208 ms TTFT234.9151 ms TTFT382.4295 ms TTFT603.6310 ms TTFT929.7815 ms TTFT1,544.2674 ms TTFT2,557.51,212 ms TTFT4,301.21,959 ms TTFT
16K138.1365 ms TTFT239.3200 ms TTFT391.8373 ms TTFT626.61,221 ms TTFT953.41,065 ms TTFT1,587.91,155 ms TTFT2,660.31,641 ms TTFT4,373.42,593 ms TTFT
32K137.8662 ms TTFT239.8244 ms TTFT396.1636 ms TTFT622.8710 ms TTFT956.9790 ms TTFT1,581.91,211 ms TTFT2,688.62,008 ms TTFT4,333.33,695 ms TTFT
64K136.51,300 ms TTFT241.7517 ms TTFT393.2866 ms TTFT623.01,251 ms TTFT975.11,707 ms TTFT1,542.12,645 ms TTFT2,621.84,427 ms TTFT4,274.08,062 ms TTFT
128K133.82,337 ms TTFT240.7718 ms TTFT389.3918 ms TTFT619.71,492 ms TTFT953.52,678 ms TTFT1,540.84,724 ms TTFT2,563.79,129 ms TTFT4,117.318,522 ms TTFT
251K131.74,745 ms TTFT239.91,378 ms TTFT390.72,026 ms TTFT607.09,916 ms TTFT907.55,231 ms TTFT1,468.99,579 ms TTFT2,468.318,372 ms TTFT3,864.036,752 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
Prefill8K1Valid25.2K—325 ms——1——
Prefill16K1Valid29.5K—547 ms——1——
Prefill32K1Valid30.2K—1,065 ms——1——
Prefill64K1Valid30.9K—2,075 ms——1——
Prefill128K1Valid31K—4,136 ms——1——
Prefill251K1Valid29.6K—8,481 ms——1——
Prefill256K1Valid28.4K—9,021 ms——1——
Sustained decode8K1Valid137.1138.6208 ms7.2 ms1.014,11230.0 s
Sustained decode8K2Valid234.9117.5151 ms8.5 ms2.007,04630.0 s
Sustained decode8K4Valid382.496.0295 ms10.4 ms4.0011,46830.0 s
Sustained decode8K8Valid603.675.6310 ms13.2 ms8.0018,10430.0 s
Sustained decode8K16Valid929.757.8815 ms17.3 ms16.0027,88830.0 s
Sustained decode8K32Valid1,544.248.2674 ms20.7 ms32.0046,30430.0 s
Sustained decode8K64Valid2,557.539.41,212 ms25.4 ms64.0076,73630.0 s
Sustained decode8K128Valid4,301.233.11,959 ms30.2 ms128.00129,02430.0 s
Sustained decode16K1Valid138.1139.8365 ms7.2 ms1.014,14430.0 s
Sustained decode16K2Valid239.3120.5200 ms8.3 ms2.027,17830.0 s
Sustained decode16K4Valid391.898.2373 ms10.2 ms4.0011,75230.0 s
Sustained decode16K8Valid626.678.31,221 ms12.8 ms8.0018,79230.0 s
Sustained decode16K16Valid953.460.21,065 ms16.6 ms16.0028,59230.0 s
Sustained decode16K32Valid1,587.949.81,155 ms20.1 ms32.0047,61630.0 s
Sustained decode16K64Valid2,660.341.51,641 ms24.1 ms64.0079,80830.0 s
Sustained decode16K128Valid4,373.433.62,593 ms29.8 ms128.00131,20030.0 s
Sustained decode32K1Valid137.8140.1662 ms7.1 ms1.014,13530.0 s
Sustained decode32K2Valid239.8120.6244 ms8.3 ms2.027,18129.9 s
Sustained decode32K4Valid396.198.1636 ms10.2 ms4.0011,88230.0 s
Sustained decode32K8Valid622.877.6710 ms12.9 ms8.0018,68030.0 s
Sustained decode32K16Valid956.959.0790 ms17.0 ms16.0028,70430.0 s
Sustained decode32K32Valid1,581.947.91,211 ms20.9 ms32.0047,45630.0 s
Sustained decode32K64Valid2,688.640.22,008 ms24.9 ms64.0080,64030.0 s
Sustained decode32K128Valid4,333.332.13,695 ms31.1 ms128.00130,04830.0 s
Sustained decode64K1Valid136.5139.31,300 ms7.2 ms1.014,09330.0 s
Sustained decode64K2Valid241.7120.6517 ms8.3 ms2.027,22429.9 s
Sustained decode64K4Valid393.297.9866 ms10.2 ms4.0011,79630.0 s
Sustained decode64K8Valid623.077.51,251 ms12.9 ms8.0018,68830.0 s
Sustained decode64K16Valid975.160.11,707 ms16.6 ms16.0029,24830.0 s
Sustained decode64K32Valid1,542.146.62,645 ms21.5 ms32.0046,24030.0 s
Sustained decode64K64Valid2,621.838.04,427 ms26.3 ms64.0078,65630.0 s
Sustained decode64K128Valid4,274.028.68,062 ms34.9 ms128.00128,25630.0 s
Sustained decode128K1Valid133.8137.92,337 ms7.3 ms1.014,01430.0 s
Sustained decode128K2Valid240.7120.6718 ms8.3 ms2.017,22130.0 s
Sustained decode128K4Valid389.396.7918 ms10.3 ms4.0011,68030.0 s
Sustained decode128K8Valid619.776.21,492 ms13.1 ms8.0018,58430.0 s
Sustained decode128K16Valid953.557.72,678 ms17.3 ms16.0028,59230.0 s
Sustained decode128K32Valid1,540.845.34,724 ms22.1 ms32.0046,20830.0 s
Sustained decode128K64Valid2,563.736.19,129 ms27.7 ms64.0076,86430.0 s
Sustained decode128K128Valid4,117.327.318,522 ms36.7 ms128.00123,52030.0 s
Sustained decode251K1Valid131.7139.14,745 ms7.2 ms1.013,95130.0 s
Sustained decode251K2Valid239.9119.91,378 ms8.3 ms2.007,19630.0 s
Sustained decode251K4Valid390.797.12,026 ms10.3 ms4.0011,72030.0 s
Sustained decode251K8Valid607.077.69,916 ms12.9 ms8.0018,21030.0 s
Sustained decode251K16Valid907.555.95,231 ms17.9 ms16.0027,21630.0 s
Sustained decode251K32Valid1,468.944.49,579 ms22.5 ms32.0044,06430.0 s
Sustained decode251K64Valid2,468.336.418,372 ms27.5 ms64.0074,04830.0 s
Sustained decode251K128Valid3,864.028.236,752 ms35.5 ms128.00115,83930.0 s