시작하기리더보드Decode calculator모델OrganizationsReportsMatrix runs하드웨어벤치마크마켓플레이스렌탈ProAPI 문서
언어
Actual Computer — Every computer, one endpoint

Speed test matrix

DeepSeek-V4-Flash-DSpark on H200 NVL 141GB ×2

sglang · 0.5.16 · 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 NVL
GPU count
2
VRAM
282 GB (2 × 141 GB)
OS
linux

Engine

Engine
sglang
Version
0.5.16
Quantization
MXFP4+FP8
Backend
cuda
Build
cuda-13.2

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-sglang/bin/sglang" serve --model-path deepseek-ai/DeepSeek-V4-Flash-DSpark --served-model-name deepseek --trust-remote-code --tp 2 --reasoning-parser deepseek-v4 --tool-call-parser deepseekv4 --context-length 1048576 --mem-fraction-static 0.808234 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted> --moe-runner-backend marlin --speculative-algorithm DSPARK

Prefill

prompt tok/s by context length
8K context10.2K803 ms TTFT
16K context14.8K1,096 ms TTFT
32K context15K2,153 ms TTFT
64K context14.7K4,382 ms TTFT
128K context13.8K9,342 ms TTFT
256K context12K21,422 ms TTFT
512K context9,54454,018 ms TTFT
1019K context6,762151,709 ms TTFT
1024K context6,686154,185 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K228.4108 ms TTFT628.5133 ms TTFT681.5176 ms TTFT1,108.6249 ms TTFT1,878.4263 ms TTFT2,544.2842 ms TTFT3,714.5726 ms TTFT4,695.41,317 ms TTFT
16K301.6135 ms TTFT492.3160 ms TTFT735.5195 ms TTFT1,175.5185 ms TTFT1,570.9392 ms TTFT2,320.2755 ms TTFT3,589.01,757 ms TTFT4,348.22,533 ms TTFT
32K338.1158 ms TTFT524.7185 ms TTFT691.6216 ms TTFT1,055.8434 ms TTFT1,625.8765 ms TTFT2,264.21,463 ms TTFT3,505.02,873 ms TTFTOver capacity
64K245.1209 ms TTFT530.0245 ms TTFT643.8259 ms TTFT1,017.5771 ms TTFT1,765.81,583 ms TTFT2,306.03,091 ms TTFTOver capacityOver capacity
128K272.0307 ms TTFT383.8510 ms TTFT761.4696 ms TTFT939.01,616 ms TTFT1,672.93,146 ms TTFTOver capacityOver capacityOver capacity
256K235.7531 ms TTFT417.6696 ms TTFT612.51,071 ms TTFT1,057.83,049 ms TTFTOver capacityOver capacityOver capacityOver capacity
512K218.61,026 ms TTFT438.81,310 ms TTFT513.33,154 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacity
1019K185.42,056 ms TTFT353.9Requests queued2,699 ms TTFTOver capacityOver capacityOver capacityOver capacityOver capacityOver capacity

Requests queued: the server could not run all requests at once, so some waited in a queue. 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
Prefill8K1Valid10.2K—803 ms——1——
Prefill16K1Valid14.8K—1,096 ms——1——
Prefill32K1Valid15K—2,153 ms——1——
Prefill64K1Valid14.7K—4,382 ms——1——
Prefill128K1Valid13.8K—9,342 ms——1——
Prefill256K1Valid12K—21,422 ms——1——
Prefill512K1Valid9,544.0—54,018 ms——1——
Prefill1019K1Valid6,762.0—151,709 ms——1——
Prefill1024K1Valid6,686.0—154,185 ms——1——
Sustained decode8K1Valid228.4230.1108 ms4.4 ms1.016,85130.0 s
Sustained decode8K2Valid628.5340.5133 ms2.9 ms2.0418,85230.0 s
Sustained decode8K4Valid681.5166.2176 ms6.0 ms4.0320,43930.0 s
Sustained decode8K8Valid1,108.6120.2249 ms8.3 ms7.9533,24230.0 s
Sustained decode8K16Valid1,878.4101.6263 ms9.8 ms15.9956,33130.0 s
Sustained decode8K32Valid2,544.278.9842 ms12.7 ms32.0076,24830.0 s
Sustained decode8K64Valid3,714.550.4726 ms19.8 ms64.00111,33430.0 s
Sustained decode8K128Valid4,695.433.01,317 ms30.3 ms128.00140,77030.0 s
Sustained decode16K1Valid301.6256.0135 ms3.9 ms1.029,04830.0 s
Sustained decode16K2Valid492.3216.3160 ms4.6 ms2.0314,76730.0 s
Sustained decode16K4Valid735.5175.0195 ms5.7 ms4.0422,06330.0 s
Sustained decode16K8Valid1,175.5134.5185 ms7.4 ms7.9835,26230.0 s
Sustained decode16K16Valid1,570.989.0392 ms11.2 ms16.0347,07430.0 s
Sustained decode16K32Valid2,320.262.2755 ms16.1 ms32.0069,62630.0 s
Sustained decode16K64Valid3,589.049.91,757 ms20.1 ms64.00107,67130.0 s
Sustained decode16K128Valid4,348.230.52,533 ms32.8 ms128.00130,23030.0 s
Sustained decode32K1Valid338.1386.6158 ms2.6 ms0.9210,14130.0 s
Sustained decode32K2Valid524.7221.2185 ms4.5 ms2.0415,72330.0 s
Sustained decode32K4Valid691.6157.2216 ms6.4 ms4.0420,73930.0 s
Sustained decode32K8Valid1,055.8116.8434 ms8.6 ms7.9631,67130.0 s
Sustained decode32K16Valid1,625.887.2765 ms11.5 ms15.9548,76030.0 s
Sustained decode32K32Valid2,264.268.41,463 ms14.6 ms32.0067,89930.0 s
Sustained decode32K64Valid3,505.050.12,873 ms20.0 ms64.00105,16230.0 s
Sustained decode32K128Over capacity————————
Sustained decode64K1Valid245.1244.9209 ms4.1 ms1.027,35430.0 s
Sustained decode64K2Valid530.0301.8245 ms3.3 ms2.0315,89130.0 s
Sustained decode64K4Valid643.8155.9259 ms6.4 ms3.9519,30430.0 s
Sustained decode64K8Valid1,017.5122.0771 ms8.2 ms7.9530,51830.0 s
Sustained decode64K16Valid1,765.899.21,583 ms10.1 ms16.0652,95730.0 s
Sustained decode64K32Valid2,306.068.63,091 ms14.6 ms32.0069,15430.0 s
Sustained decode64K64Over capacity————————
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid272.0232.1307 ms4.3 ms1.028,16030.0 s
Sustained decode128K2Valid383.8190.9510 ms5.3 ms2.0211,50930.0 s
Sustained decode128K4Valid761.4226.0696 ms4.4 ms3.9422,83430.0 s
Sustained decode128K8Valid939.0111.01,616 ms9.0 ms8.0228,16730.0 s
Sustained decode128K16Valid1,672.988.53,146 ms11.3 ms15.9750,12830.0 s
Sustained decode128K32Over capacity————————
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode256K1Valid235.7238.3531 ms4.2 ms1.017,06830.0 s
Sustained decode256K2Valid417.6189.5696 ms5.3 ms2.0212,52630.0 s
Sustained decode256K4Valid612.5158.41,071 ms6.3 ms3.9418,36630.0 s
Sustained decode256K8Valid1,057.8130.93,049 ms7.6 ms7.9531,73430.0 s
Sustained decode256K16Over capacity————————
Sustained decode256K32Over capacity————————
Sustained decode256K64Over capacity————————
Sustained decode256K128Over capacity————————
Sustained decode512K1Valid218.6224.51,026 ms4.5 ms1.016,55630.0 s
Sustained decode512K2Valid438.8239.81,310 ms4.3 ms1.9213,15730.0 s
Sustained decode512K4Valid513.3137.93,154 ms7.3 ms3.9315,39130.0 s
Sustained decode512K8Over capacity————————
Sustained decode512K16Over capacity————————
Sustained decode512K32Over capacity————————
Sustained decode512K64Over capacity————————
Sustained decode512K128Over capacity————————
Sustained decode1019K1Valid185.4201.52,056 ms5.0 ms1.015,56130.0 s
Sustained decode1019K2Requests queued353.9183.32,699 ms5.5 ms1.9210,61530.0 s
Sustained decode1019K4Over capacity————————
Sustained decode1019K8Over capacity————————
Sustained decode1019K16Over capacity————————
Sustained decode1019K32Over capacity————————
Sustained decode1019K64Over capacity————————
Sustained decode1019K128Over capacity————————

Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.