Aan de slagLeaderboardDecode calculatorModellenOrganizationsReportsMatrix runsHardwareBenchmarksMarktplaatsVerhuurProAPI Docs
Taal
Actual Computer — Every computer, one endpoint

Speed test matrix

GLM-5.3-Flash-NVFP4 on H200 SXM 141GB ×4

vllm · 0.28.1rc1.dev359+g98ed0856f · NVFP4+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
4
VRAM
564 GB (4 × 141 GB)
OS
linux

Engine

Engine
vllm
Version
0.28.1rc1.dev359+g98ed0856f
Quantization
NVFP4+FP8
Backend
cuda
Build
cuda-13.2
Commit
98ed0856f31f

Run

Served model
glm
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_USE_V2_MODEL_RUNNER=1 VLLM_ENGINE_READY_TIMEOUT_S=3600 "$HOME/env_h200-glm53flash-vllm-pr-53906/bin/vllm" serve RedHatAI/GLM-5.3-Flash-NVFP4 --served-model-name glm --tensor-parallel-size 4 --reasoning-parser glm45 --enable-auto-tool-choice --tool-call-parser glm47 --max-model-len 1048576 --gpu-memory-utilization 0.88 --host 0.0.0.0 --port 8000 --api-key <redacted> --no-enable-flashinfer-autotune

Prefill

prompt tok/s by context length
8K context17.8K462 ms TTFT
16K context19.1K849 ms TTFT
32K context19.8K1,637 ms TTFT
64K context20.1K3,213 ms TTFT
128K context19.4K6,631 ms TTFT
256K context18.6K13,860 ms TTFT
512K context17K30,246 ms TTFT
1019K context14.2K72,222 ms TTFT
1024K context14.1K73,039 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests
Context1248163264128
8K129.676 ms TTFT233.8109 ms TTFT418.6148 ms TTFT706.6377 ms TTFT1,213.5610 ms TTFT1,885.8565 ms TTFT3,062.61,047 ms TTFT4,635.31,338 ms TTFT
16K129.292 ms TTFT231.9147 ms TTFT413.4168 ms TTFT700.7239 ms TTFT1,206.7368 ms TTFT1,882.5571 ms TTFT3,064.9963 ms TTFT4,622.81,698 ms TTFT
32K130.5122 ms TTFT234.8220 ms TTFT421.1231 ms TTFT707.2311 ms TTFT1,208.9572 ms TTFT1,869.5947 ms TTFT3,031.91,686 ms TTFT4,596.83,235 ms TTFT
64K129.8164 ms TTFT233.3265 ms TTFT418.0339 ms TTFT698.5512 ms TTFT1,201.2901 ms TTFT1,851.81,682 ms TTFT3,005.23,235 ms TTFTOver capacity
128K127.3449 ms TTFT228.7610 ms TTFT400.9839 ms TTFT656.31,319 ms TTFT1,169.12,360 ms TTFT1,774.04,443 ms TTFTOver capacityOver capacity
256K125.1816 ms TTFT225.91,122 ms TTFT393.11,647 ms TTFT635.32,730 ms TTFT1,092.14,823 ms TTFTOver capacityOver capacityOver capacity
512K127.81,615 ms TTFT225.52,336 ms TTFT391.73,623 ms TTFT625.06,108 ms TTFTOver capacityOver capacityOver capacityOver capacity
1019K126.93,942 ms TTFT222.65,672 ms TTFT384.38,412 ms TTFTOver capacityOver 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
Prefill8K1Valid17.8K—462 ms——1——
Prefill16K1Valid19.1K—849 ms——1——
Prefill32K1Valid19.8K—1,637 ms——1——
Prefill64K1Valid20.1K—3,213 ms——1——
Prefill128K1Valid19.4K—6,631 ms——1——
Prefill256K1Valid18.6K—13,860 ms——1——
Prefill512K1Valid17K—30,246 ms——1——
Prefill1019K1Valid14.2K—72,222 ms——1——
Prefill1024K1Valid14.1K—73,039 ms——1——
Sustained decode8K1Valid129.6130.076 ms7.7 ms1.013,88730.0 s
Sustained decode8K2Valid233.8116.8109 ms8.6 ms2.007,01330.0 s
Sustained decode8K4Valid418.6104.8148 ms9.5 ms4.0012,55530.0 s
Sustained decode8K8Valid706.688.4377 ms11.3 ms8.0021,19230.0 s
Sustained decode8K16Valid1,213.576.1610 ms13.1 ms16.0036,40130.0 s
Sustained decode8K32Valid1,885.857.9565 ms17.3 ms32.0056,57330.0 s
Sustained decode8K64Valid3,062.647.81,047 ms20.9 ms64.0091,83530.0 s
Sustained decode8K128Valid4,635.335.91,338 ms27.9 ms128.00139,13230.0 s
Sustained decode16K1Valid129.2129.892 ms7.7 ms1.013,87530.0 s
Sustained decode16K2Valid231.9115.9147 ms8.6 ms2.006,95530.0 s
Sustained decode16K4Valid413.4103.4168 ms9.7 ms4.0012,39930.0 s
Sustained decode16K8Valid700.787.6239 ms11.4 ms8.0021,01530.0 s
Sustained decode16K16Valid1,206.775.3368 ms13.3 ms16.0036,19230.0 s
Sustained decode16K32Valid1,882.558.6571 ms17.1 ms32.0056,47630.0 s
Sustained decode16K64Valid3,064.947.4963 ms21.1 ms64.0091,90130.0 s
Sustained decode16K128Valid4,622.835.41,698 ms28.2 ms128.00138,75130.0 s
Sustained decode32K1Valid130.5130.6122 ms7.7 ms1.013,91430.0 s
Sustained decode32K2Valid234.8117.3220 ms8.5 ms2.007,04230.0 s
Sustained decode32K4Valid421.1104.9231 ms9.5 ms4.0012,63230.0 s
Sustained decode32K8Valid707.288.1311 ms11.4 ms8.0021,20830.0 s
Sustained decode32K16Valid1,208.975.1572 ms13.3 ms16.0036,25530.0 s
Sustained decode32K32Valid1,869.557.7947 ms17.3 ms32.0056,06330.0 s
Sustained decode32K64Valid3,031.946.41,686 ms21.6 ms64.0090,94330.0 s
Sustained decode32K128Valid4,596.834.53,235 ms28.9 ms128.00137,85530.0 s
Sustained decode64K1Valid129.8131.0164 ms7.6 ms1.013,89430.0 s
Sustained decode64K2Valid233.3116.8265 ms8.6 ms2.007,00030.0 s
Sustained decode64K4Valid418.0104.4339 ms9.6 ms4.0012,53630.0 s
Sustained decode64K8Valid698.587.3512 ms11.5 ms8.0020,95430.0 s
Sustained decode64K16Valid1,201.275.0901 ms13.3 ms16.0036,03330.0 s
Sustained decode64K32Valid1,851.857.81,682 ms17.3 ms32.0055,55130.0 s
Sustained decode64K64Valid3,005.247.23,235 ms21.2 ms64.0090,10930.0 s
Sustained decode64K128Over capacity————————
Sustained decode128K1Valid127.3131.9449 ms7.6 ms1.013,81830.0 s
Sustained decode128K2Valid228.7114.3610 ms8.8 ms2.006,86030.0 s
Sustained decode128K4Valid400.999.8839 ms10.0 ms4.0012,02730.0 s
Sustained decode128K8Valid656.381.01,319 ms12.3 ms8.0019,68830.0 s
Sustained decode128K16Valid1,169.170.82,360 ms14.1 ms16.0035,07330.0 s
Sustained decode128K32Valid1,774.052.54,443 ms19.0 ms32.0053,21530.0 s
Sustained decode128K64Over capacity————————
Sustained decode128K128Over capacity————————
Sustained decode256K1Valid125.1133.1816 ms7.5 ms1.013,75230.0 s
Sustained decode256K2Valid225.9113.11,122 ms8.8 ms2.006,77630.0 s
Sustained decode256K4Valid393.197.91,647 ms10.2 ms4.0011,79130.0 s
Sustained decode256K8Valid635.378.52,730 ms12.7 ms8.0019,05530.0 s
Sustained decode256K16Valid1,092.166.34,823 ms15.1 ms16.0032,75230.0 s
Sustained decode256K32Over capacity————————
Sustained decode256K64Over capacity————————
Sustained decode256K128Over capacity————————
Sustained decode512K1Valid127.8128.51,615 ms7.8 ms1.013,68328.8 s
Sustained decode512K2Valid225.5113.52,336 ms8.8 ms2.006,76630.0 s
Sustained decode512K4Valid391.798.43,623 ms10.2 ms4.0011,75230.0 s
Sustained decode512K8Valid625.078.56,108 ms12.7 ms8.0018,74430.0 s
Sustained decode512K16Over capacity————————
Sustained decode512K32Over capacity————————
Sustained decode512K64Over capacity————————
Sustained decode512K128Over capacity————————
Sustained decode1019K1Valid126.9128.43,942 ms7.8 ms0.913,58628.3 s
Sustained decode1019K2Valid222.6112.05,672 ms8.9 ms2.016,67830.0 s
Sustained decode1019K4Valid384.397.18,412 ms10.3 ms4.0111,52830.0 s
Sustained decode1019K8Over capacity————————
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.