เริ่มต้นใช้งานลีดเดอร์บอร์ดDecode calculatorโมเดลOrganizationsReportsMatrix runsฮาร์ดแวร์เบนช์มาร์กตลาดซื้อขายเช่าProเอกสาร API
ภาษา
Actual Computer — Every computer, one endpoint

Speed test matrix

Qwen3.8-Flash-Next-FP8 on RTX PRO 6000 Blackwell Server Edition 96GB ×4

vllm · 0.1.dev1+g50a061f79 · 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
RTX PRO 6000 Blackwell Server Edition
GPU count
4
VRAM
384 GB (4 × 96 GB)
OS
linux

Engine

Engine
vllm
Version
0.1.dev1+g50a061f79
Quantization
FP8
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_rtxpro6k-qwen-flash-next-vllm-pr-54129/bin/vllm" serve Qwen/Qwen3.8-Flash-Next-FP8 --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 0.79 --host 0.0.0.0 --port 8000 --api-key <redacted> --enable-expert-parallel --moe-backend triton

Prefill

prompt tok/s by context length
8K context13.1K627 ms TTFT
16K context14.3K1,133 ms TTFT
32K context14.7K2,185 ms TTFT
64K context14.8K4,329 ms TTFT
128K context14.7K8,708 ms TTFT
251K context14.3K17,488 ms TTFT
256K context13.6K18,850 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K105.71,533 ms TTFT177.71,325 ms TTFT311.2366 ms TTFT526.8384 ms TTFT842.2571 ms TTFT1,332.61,172 ms TTFT1,988.22,573 ms TTFT3,120.63,170 ms TTFT
16K105.51,152 ms TTFT176.8235 ms TTFT309.1426 ms TTFT536.4571 ms TTFT854.3794 ms TTFT1,352.51,316 ms TTFT1,968.02,436 ms TTFT3,093.03,467 ms TTFT
32K105.52,714 ms TTFT177.7186 ms TTFT313.8587 ms TTFT532.2366 ms TTFT843.3566 ms TTFT1,340.8825 ms TTFT1,958.71,489 ms TTFT3,052.02,529 ms TTFT
64K105.14,411 ms TTFT175.6332 ms TTFT312.4592 ms TTFT525.2808 ms TTFT824.91,199 ms TTFT1,305.91,830 ms TTFT1,927.93,051 ms TTFT3,003.25,621 ms TTFT
128K104.28,742 ms TTFT176.9523 ms TTFT308.9637 ms TTFT511.91,261 ms TTFT807.31,987 ms TTFT1,251.83,569 ms TTFT1,842.66,299 ms TTFT2,867.412,731 ms TTFT
251K104.517,585 ms TTFT173.6869 ms TTFT303.51,277 ms TTFT516.12,061 ms TTFT803.83,837 ms TTFT1,245.17,002 ms TTFT1,800.212,649 ms TTFT2,729.624,706 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
Prefill8K1Valid13.1K—627 ms——1——
Prefill16K1Valid14.3K—1,133 ms——1——
Prefill32K1Valid14.7K—2,185 ms——1——
Prefill64K1Valid14.8K—4,329 ms——1——
Prefill128K1Valid14.7K—8,708 ms——1——
Prefill251K1Valid14.3K—17,488 ms——1——
Prefill256K1Valid13.6K—18,850 ms——1——
Sustained decode8K1Valid105.7105.81,533 ms9.5 ms1.003,17230.0 s
Sustained decode8K2Valid177.785.81,325 ms11.7 ms2.005,33030.0 s
Sustained decode8K4Valid311.277.7366 ms12.9 ms4.009,33230.0 s
Sustained decode8K8Valid526.866.0384 ms15.2 ms8.0015,80030.0 s
Sustained decode8K16Valid842.251.6571 ms19.4 ms16.0025,26430.0 s
Sustained decode8K32Valid1,332.642.11,172 ms23.8 ms32.0039,96830.0 s
Sustained decode8K64Valid1,988.231.32,573 ms32.0 ms64.0059,64830.0 s
Sustained decode8K128Valid3,120.624.43,170 ms41.0 ms128.0093,56830.0 s
Sustained decode16K1Valid105.5105.51,152 ms9.5 ms1.003,16530.0 s
Sustained decode16K2Valid176.888.3235 ms11.3 ms2.005,30430.0 s
Sustained decode16K4Valid309.177.6426 ms12.9 ms4.009,27230.0 s
Sustained decode16K8Valid536.467.1571 ms14.9 ms8.0016,08830.0 s
Sustained decode16K16Valid854.353.1794 ms18.8 ms16.0025,61630.0 s
Sustained decode16K32Valid1,352.541.61,316 ms24.1 ms32.0040,57630.0 s
Sustained decode16K64Valid1,968.031.02,436 ms32.3 ms64.0059,00830.0 s
Sustained decode16K128Valid3,093.023.83,467 ms42.0 ms128.0092,80030.0 s
Sustained decode32K1Valid105.5105.52,714 ms9.5 ms1.003,16430.0 s
Sustained decode32K2Valid177.788.9186 ms11.2 ms2.005,32830.0 s
Sustained decode32K4Valid313.878.3587 ms12.8 ms4.009,41230.0 s
Sustained decode32K8Valid532.265.5366 ms15.3 ms8.0015,96030.0 s
Sustained decode32K16Valid843.352.8566 ms18.9 ms16.0025,29630.0 s
Sustained decode32K32Valid1,340.841.7825 ms24.0 ms32.0040,22430.0 s
Sustained decode32K64Valid1,958.730.21,489 ms33.1 ms64.0058,75230.0 s
Sustained decode32K128Valid3,052.023.32,529 ms43.0 ms128.0091,52030.0 s
Sustained decode64K1Valid105.1105.24,411 ms9.5 ms1.003,15330.0 s
Sustained decode64K2Valid175.687.9332 ms11.4 ms2.005,26830.0 s
Sustained decode64K4Valid312.477.9592 ms12.8 ms4.009,36830.0 s
Sustained decode64K8Valid525.265.2808 ms15.3 ms8.0015,75230.0 s
Sustained decode64K16Valid824.951.21,199 ms19.5 ms16.0024,73630.0 s
Sustained decode64K32Valid1,305.940.11,830 ms24.9 ms32.0039,16830.0 s
Sustained decode64K64Valid1,927.928.93,051 ms34.6 ms64.0057,79230.0 s
Sustained decode64K128Valid3,003.221.65,621 ms46.3 ms128.0090,11230.0 s
Sustained decode128K1Valid104.2104.58,742 ms9.6 ms1.003,12530.0 s
Sustained decode128K2Valid176.988.1523 ms11.3 ms2.005,30630.0 s
Sustained decode128K4Valid308.976.7637 ms13.0 ms4.009,26430.0 s
Sustained decode128K8Valid511.961.81,261 ms16.2 ms8.0015,35230.0 s
Sustained decode128K16Valid807.347.81,987 ms20.9 ms16.0024,20830.0 s
Sustained decode128K32Valid1,251.836.93,569 ms27.1 ms32.0037,53630.0 s
Sustained decode128K64Valid1,842.627.66,299 ms36.2 ms64.0055,23230.0 s
Sustained decode128K128Valid2,867.421.312,731 ms46.9 ms128.0086,01430.0 s
Sustained decode251K1Valid104.5104.717,585 ms9.6 ms1.003,13630.0 s
Sustained decode251K2Valid173.687.0869 ms11.5 ms2.005,20630.0 s
Sustained decode251K4Valid303.575.71,277 ms13.2 ms4.009,10430.0 s
Sustained decode251K8Valid516.163.92,061 ms15.6 ms8.0015,48030.0 s
Sustained decode251K16Valid803.849.33,837 ms20.3 ms16.0024,11230.0 s
Sustained decode251K32Valid1,245.138.07,002 ms26.3 ms32.0037,34430.0 s
Sustained decode251K64Valid1,800.227.712,649 ms36.1 ms64.0053,95230.0 s
Sustained decode251K128Valid2,729.620.824,706 ms48.0 ms128.0081,79230.0 s