Pierwsze krokiRankingDecode calculatorModeleOrganizationsReportsMatrix runsSprzętBenchmarkiMarketplaceWynajemProDokumentacja API
Język
Actual Computer — Every computer, one endpoint

Speed test matrix

DeepSeek-V4.1-Flash on H200 SXM 141GB ×8

vllm · 0.1.1.dev5+ge77daef89 · MXFP4+FP88 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 SXM
GPU count
8
VRAM
1128 GB (8 × 141 GB)
OS
linux

Engine

Engine
vllm
Version
0.1.1.dev5+ge77daef89
Quantization
MXFP4+FP8
Backend
cuda
Build
cuda-13.2
Commit
e77daef89e18

Run

Served model
deepseek
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

"$HOME/env_h200-deepseek-v41-vllm-e77daef89/bin/vllm" serve deepseek-ai/DeepSeek-V4.1-Flash --served-model-name deepseek --trust-remote-code --tensor-parallel-size 8 --reasoning-parser deepseek_v41 --enable-auto-tool-choice --tool-call-parser deepseek_v41 --max-model-len 1048576 --gpu-memory-utilization 0.88 --host 0.0.0.0 --port 8000 --api-key <redacted> --dtype bfloat16 --tokenizer-mode deepseek_v41 --mm-encoder-tp-mode data --engram-config '{"cpu_offload":false}'

Prefill

prompt tok/s by context length
8K context18.7K437 ms TTFT
16K context20.5K790 ms TTFT
32K context20.6K1,568 ms TTFT
64K context20.2K3,193 ms TTFT
128K context19K6,759 ms TTFT
256K context16.7K15,340 ms TTFT
512K context13.4K38,366 ms TTFT
1019K context7,827130,577 ms TTFT
1024K context7,755132,438 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K130.950 ms TTFT249.858 ms TTFT464.995 ms TTFT840.0129 ms TTFT1,435.2222 ms TTFT2,272.0333 ms TTFT3,373.1597 ms TTFT5,277.21,148 ms TTFT
16K130.664 ms TTFT248.988 ms TTFT463.8116 ms TTFT840.7188 ms TTFT1,433.4308 ms TTFT2,255.8571 ms TTFT3,361.2990 ms TTFT5,234.21,872 ms TTFT
32K130.2107 ms TTFT248.1127 ms TTFT462.2201 ms TTFT836.3327 ms TTFT1,429.2547 ms TTFT2,239.9995 ms TTFT3,331.41,794 ms TTFT5,166.03,403 ms TTFT
64K129.7184 ms TTFT246.9223 ms TTFT460.1363 ms TTFT835.0594 ms TTFT1,417.31,044 ms TTFT2,223.41,974 ms TTFT3,283.63,844 ms TTFT5,044.07,493 ms TTFT
128K128.4354 ms TTFT244.1449 ms TTFT458.4749 ms TTFT828.11,263 ms TTFT1,394.42,281 ms TTFT2,173.44,316 ms TTFT3,176.88,326 ms TTFTOver capacity
256K126.0732 ms TTFT243.61,102 ms TTFT452.51,573 ms TTFT811.42,711 ms TTFT1,350.05,014 ms TTFT2,063.79,529 ms TTFTOver capacityOver capacity
512K126.91,640 ms TTFT237.02,340 ms TTFT441.83,715 ms TTFT775.66,367 ms TTFT1,238.411,163 ms TTFTOver capacityOver capacityOver capacity
1019K123.33,652 ms TTFT225.75,413 ms TTFT398.57,068 ms TTFT665.311,788 ms TTFTOver 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
Prefill8K1Valid18.7K—437 ms——1——
Prefill16K1Valid20.5K—790 ms——1——
Prefill32K1Valid20.6K—1,568 ms——1——
Prefill64K1Valid20.2K—3,193 ms——1——
Prefill128K1Valid19K—6,759 ms——1——
Prefill256K1Valid16.7K—15,340 ms——1——
Prefill512K1Valid13.4K—38,366 ms——1——
Prefill1019K1Valid7,827.0—130,577 ms——1——
Prefill1024K1Valid7,755.0—132,438 ms——1——
Sustained decode8K1Valid130.9131.250 ms7.6 ms1.013,92630.0 s
Sustained decode8K2Valid249.8125.158 ms8.0 ms1.927,49530.0 s
Sustained decode8K4Valid464.9116.395 ms8.6 ms4.0013,94830.0 s
Sustained decode8K8Valid840.0105.2129 ms9.5 ms8.0025,20030.0 s
Sustained decode8K16Valid1,435.289.9222 ms11.1 ms16.0043,05630.0 s
Sustained decode8K32Valid2,272.071.2333 ms14.0 ms32.0068,16030.0 s
Sustained decode8K64Valid3,373.152.8597 ms18.9 ms64.00101,18630.0 s
Sustained decode8K128Valid5,277.241.11,148 ms24.3 ms128.00158,33630.0 s
Sustained decode16K1Valid130.6130.964 ms7.6 ms1.013,91730.0 s
Sustained decode16K2Valid248.9124.888 ms8.0 ms2.027,46730.0 s
Sustained decode16K4Valid463.8116.0116 ms8.6 ms4.0013,91230.0 s
Sustained decode16K8Valid840.7105.2188 ms9.5 ms8.0025,21630.0 s
Sustained decode16K16Valid1,433.489.8308 ms11.1 ms16.0042,99230.0 s
Sustained decode16K32Valid2,255.870.6571 ms14.2 ms32.0067,64830.0 s
Sustained decode16K64Valid3,361.252.5990 ms19.0 ms64.00100,79730.0 s
Sustained decode16K128Valid5,234.241.01,872 ms24.4 ms128.00157,05530.0 s
Sustained decode32K1Valid130.2131.1107 ms7.6 ms1.013,90630.0 s
Sustained decode32K2Valid248.1124.6127 ms8.0 ms2.027,44230.0 s
Sustained decode32K4Valid462.2115.6201 ms8.7 ms4.0013,86430.0 s
Sustained decode32K8Valid836.3104.6327 ms9.6 ms8.0025,09030.0 s
Sustained decode32K16Valid1,429.289.3547 ms11.2 ms16.0042,86530.0 s
Sustained decode32K32Valid2,239.970.1995 ms14.3 ms32.0067,20030.0 s
Sustained decode32K64Valid3,331.452.11,794 ms19.2 ms64.0099,90030.0 s
Sustained decode32K128Valid5,166.040.23,403 ms24.9 ms128.00155,00930.0 s
Sustained decode64K1Valid129.7131.3184 ms7.6 ms1.013,89030.0 s
Sustained decode64K2Valid246.9125.0223 ms8.0 ms2.027,40430.0 s
Sustained decode64K4Valid460.1115.1363 ms8.7 ms4.0013,80030.0 s
Sustained decode64K8Valid835.0104.4594 ms9.6 ms8.0025,04930.0 s
Sustained decode64K16Valid1,417.388.51,044 ms11.3 ms16.0042,51330.0 s
Sustained decode64K32Valid2,223.469.51,974 ms14.4 ms32.0066,68730.0 s
Sustained decode64K64Valid3,283.651.43,844 ms19.5 ms64.0098,49530.0 s
Sustained decode64K128Valid5,044.039.47,493 ms25.4 ms128.00151,30130.0 s
Sustained decode128K1Valid128.4132.1354 ms7.6 ms1.013,85330.0 s
Sustained decode128K2Valid244.1126.2449 ms7.9 ms1.927,32330.0 s
Sustained decode128K4Valid458.4114.7749 ms8.7 ms4.0013,74930.0 s
Sustained decode128K8Valid828.1103.41,263 ms9.7 ms8.0024,84030.0 s
Sustained decode128K16Valid1,394.487.02,281 ms11.5 ms16.0041,83430.0 s
Sustained decode128K32Valid2,173.467.84,316 ms14.7 ms32.0065,18530.0 s
Sustained decode128K64Valid3,176.849.98,326 ms20.1 ms64.0095,29530.0 s
Sustained decode128K128Over capacity————————
Sustained decode256K1Valid126.0132.9732 ms7.5 ms1.013,78030.0 s
Sustained decode256K2Valid243.6122.71,102 ms8.1 ms2.017,30930.0 s
Sustained decode256K4Valid452.5113.11,573 ms8.8 ms4.0013,57330.0 s
Sustained decode256K8Valid811.4101.22,711 ms9.9 ms8.0024,33830.0 s
Sustained decode256K16Valid1,350.084.15,014 ms11.9 ms16.0040,49630.0 s
Sustained decode256K32Valid2,063.764.59,529 ms15.5 ms32.0061,88830.0 s
Sustained decode256K64Over capacity————————
Sustained decode256K128Over capacity————————
Sustained decode512K1Valid126.9127.51,640 ms7.8 ms1.013,69129.1 s
Sustained decode512K2Valid237.0120.62,340 ms8.3 ms2.017,11030.0 s
Sustained decode512K4Valid441.8110.53,715 ms9.0 ms4.0013,25530.0 s
Sustained decode512K8Valid775.697.36,367 ms10.3 ms8.0123,26830.0 s
Sustained decode512K16Valid1,238.477.811,163 ms12.9 ms15.9237,14730.0 s
Sustained decode512K32Over capacity————————
Sustained decode512K64Over capacity————————
Sustained decode512K128Over capacity————————
Sustained decode1019K1Valid123.3124.43,652 ms8.0 ms1.013,68729.9 s
Sustained decode1019K2Valid225.7116.35,413 ms8.6 ms2.016,77230.0 s
Sustained decode1019K4Valid398.5104.17,068 ms9.6 ms3.9211,95630.0 s
Sustained decode1019K8Valid665.387.511,788 ms11.4 ms7.7319,95530.0 s
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.