Speed test matrix

GLM-5.3-Flash on H200 SXM 141GB ×4

vllm · FP88 contexts (8K–1019K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 256K, 512K, 1019K, 1024K

by @keennaylmx dev importMeasured Sep 11, 2026, 02:03 UTC

Hardware

Type: Discrete GPU
GPU: H200 SXM
GPU count: 4
VRAM: 564 GB (4 × 141 GB)
OS: linux

Engine

Engine: vllm
Quantization: FP8
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2
Commit: 374e367e2b87

Run

Served model: glm-5.3-flash-dflash2
Revision: main
Benchmark tool: llm-inference-bench v0.6.2
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-dflash2-vllm-pr-55423/bin/vllm" serve zai-org/GLM-5.3-Flash --served-model-name glm-5.3-flash-dflash2 --trust-remote-code --tensor-parallel-size 4 --reasoning-parser glm45 --enable-auto-tool-choice --tool-call-parser glm47 --max-model-len 1048576 --gpu-memory-utilization 0.883543 --host 0.0.0.0 --port 8000 --api-key <redacted> --dtype bfloat16 --kv-cache-dtype bfloat16 --no-enable-flashinfer-autotune --speculative-config '{"method":"dflash","model":"incoai/GLM-5.3-Flash-DFlash2","num_speculative_tokens":7,"disable_eagle_block_drop":true}'

Prefill

prompt tok/s by context length

8K context18.9K433 ms TTFT

16K context21.2K764 ms TTFT

32K context22.8K1,417 ms TTFT

64K context4,76413,540 ms TTFT

128K context16.9K7,647 ms TTFT

256K context22.2K11,579 ms TTFT

512K context20.1K25,572 ms TTFT

1019K context16.3K62,712 ms TTFT

1024K context16.3K63,320 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests

Context	1	2	4	8	16	32	64	128
8K	175.8109 ms TTFT	345.6126 ms TTFT	511.7176 ms TTFT	915.611,091 ms TTFT	1,412.4514 ms TTFT	2,270.411,511 ms TTFT	2,793.5Requests queued11,897 ms TTFT	1,583.7Requests queued1,236 ms TTFT
16K	169.2127 ms TTFT	307.5161 ms TTFT	540.7198 ms TTFT	883.9304 ms TTFT	1,438.7523 ms TTFT	2,247.3758 ms TTFT	1,505.1Requests queued1,193 ms TTFT	Over capacity
32K	168.7185 ms TTFT	319.2194 ms TTFT	543.0237 ms TTFT	869.8421 ms TTFT	1,394.2695 ms TTFT	2,265.51,086 ms TTFT	2,291.4Requests queued2,033 ms TTFT	Over capacity
64K	173.4366 ms TTFT	337.3394 ms TTFT	476.6473 ms TTFT	923.1969 ms TTFT	1,419.31,388 ms TTFT	2,231.32,053 ms TTFT	Over capacity	Over capacity
128K	172.8506 ms TTFT	329.8523 ms TTFT	492.8549 ms TTFT	903.11,166 ms TTFT	1,383.42,040 ms TTFT	Over capacity	Over capacity	Over capacity
256K	158.1847 ms TTFT	302.0977 ms TTFT	470.31,072 ms TTFT	892.71,973 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
512K	144.91,556 ms TTFT	309.21,624 ms TTFT	449.32,656 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
1019K	138.33,458 ms TTFT	265.4Requests queued4,621 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over 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.

Phase	Context	Requests at once	Status	Total tok/s	Per request tok/s	TTFT	Time between tokens	Average active requests	Completed requests	Output tokens	Measured for
Prefill	8K	1	Valid	18.9K	—	433 ms	—	—	1	—	—
Prefill	16K	1	Valid	21.2K	—	764 ms	—	—	1	—	—
Prefill	32K	1	Valid	22.8K	—	1,417 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,764.0	—	13,540 ms	—	—	1	—	—
Prefill	128K	1	Valid	16.9K	—	7,647 ms	—	—	1	—	—
Prefill	256K	1	Valid	22.2K	—	11,579 ms	—	—	1	—	—
Prefill	512K	1	Valid	20.1K	—	25,572 ms	—	—	1	—	—
Prefill	1019K	1	Valid	16.3K	—	62,712 ms	—	—	1	—	—
Prefill	1024K	1	Valid	16.3K	—	63,320 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	175.8	177.8	109 ms	5.6 ms	1.0	1	5,272	30.0 s
Sustained decode	8K	2	Valid	345.6	166.2	126 ms	6.0 ms	2.0	2	10,365	30.0 s
Sustained decode	8K	4	Valid	511.7	134.5	176 ms	7.4 ms	4.0	4	15,351	30.0 s
Sustained decode	8K	8	Valid	915.6	114.9	11,091 ms	8.7 ms	8.0	2	27,451	30.0 s
Sustained decode	8K	16	Valid	1,412.4	85.0	514 ms	11.8 ms	16.0	0	42,382	30.0 s
Sustained decode	8K	32	Valid	2,270.4	71.3	11,511 ms	14.0 ms	32.0	0	68,078	30.0 s
Sustained decode	8K	64	Requests queued	2,793.5	43.5	11,897 ms	23.0 ms	56.4	0	83,773	30.0 s
Sustained decode	8K	128	Requests queued	1,583.7	44.4	1,236 ms	22.5 ms	54.5	9	47,337	29.9 s
Sustained decode	16K	1	Valid	169.2	173.4	127 ms	5.8 ms	1.0	1	5,077	30.0 s
Sustained decode	16K	2	Valid	307.5	163.0	161 ms	6.2 ms	2.0	2	9,224	30.0 s
Sustained decode	16K	4	Valid	540.7	140.9	198 ms	7.1 ms	4.0	4	16,215	30.0 s
Sustained decode	16K	8	Valid	883.9	113.0	304 ms	8.9 ms	8.0	1	26,507	30.0 s
Sustained decode	16K	16	Valid	1,438.7	91.3	523 ms	11.0 ms	16.0	0	43,163	30.0 s
Sustained decode	16K	32	Valid	2,247.3	71.9	758 ms	13.9 ms	32.0	0	67,413	30.0 s
Sustained decode	16K	64	Requests queued	1,505.1	44.0	1,193 ms	22.7 ms	54.3	8	45,046	29.9 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	168.7	172.4	185 ms	5.8 ms	1.0	1	5,061	30.0 s
Sustained decode	32K	2	Valid	319.2	163.7	194 ms	6.1 ms	2.0	2	9,591	30.0 s
Sustained decode	32K	4	Valid	543.0	141.3	237 ms	7.1 ms	4.0	4	16,288	30.0 s
Sustained decode	32K	8	Valid	869.8	114.3	421 ms	8.8 ms	8.0	0	26,086	30.0 s
Sustained decode	32K	16	Valid	1,394.2	88.3	695 ms	11.3 ms	16.0	0	41,817	30.0 s
Sustained decode	32K	32	Valid	2,265.5	72.3	1,086 ms	13.8 ms	32.0	0	67,896	30.0 s
Sustained decode	32K	64	Requests queued	2,291.4	51.1	2,033 ms	19.6 ms	53.4	47	68,691	30.0 s
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	173.4	191.1	366 ms	5.4 ms	1.0	1	5,200	30.0 s
Sustained decode	64K	2	Valid	337.3	173.6	394 ms	5.8 ms	2.0	2	10,118	30.0 s
Sustained decode	64K	4	Valid	476.6	130.4	473 ms	7.7 ms	4.0	2	14,296	30.0 s
Sustained decode	64K	8	Valid	923.1	116.2	969 ms	8.6 ms	8.0	1	27,690	30.0 s
Sustained decode	64K	16	Valid	1,419.3	89.0	1,388 ms	11.2 ms	16.0	0	42,577	30.0 s
Sustained decode	64K	32	Valid	2,231.3	69.1	2,053 ms	14.5 ms	32.0	0	66,902	30.0 s
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	172.8	181.6	506 ms	5.5 ms	1.0	1	5,186	30.0 s
Sustained decode	128K	2	Valid	329.8	170.9	523 ms	5.9 ms	2.0	2	9,904	30.0 s
Sustained decode	128K	4	Valid	492.8	130.9	549 ms	7.6 ms	4.0	4	14,782	30.0 s
Sustained decode	128K	8	Valid	903.1	115.0	1,166 ms	8.7 ms	8.0	1	27,095	30.0 s
Sustained decode	128K	16	Valid	1,383.4	85.3	2,040 ms	11.7 ms	16.0	0	41,505	30.0 s
Sustained decode	128K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	256K	1	Valid	158.1	163.3	847 ms	6.1 ms	1.0	1	4,743	30.0 s
Sustained decode	256K	2	Valid	302.0	161.8	977 ms	6.2 ms	1.9	2	9,062	30.0 s
Sustained decode	256K	4	Valid	470.3	125.8	1,072 ms	8.0 ms	4.0	2	14,103	30.0 s
Sustained decode	256K	8	Valid	892.7	113.6	1,973 ms	8.8 ms	8.0	3	26,766	30.0 s
Sustained decode	256K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	256K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	256K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	256K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	512K	1	Valid	144.9	170.5	1,556 ms	5.9 ms	1.0	1	4,348	30.0 s
Sustained decode	512K	2	Valid	309.2	171.7	1,624 ms	5.8 ms	1.9	2	9,274	30.0 s
Sustained decode	512K	4	Valid	449.3	119.6	2,656 ms	8.4 ms	3.9	2	13,472	30.0 s
Sustained decode	512K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	512K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	512K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	512K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	512K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	1019K	1	Valid	138.3	177.3	3,458 ms	5.7 ms	0.9	1	4,150	30.0 s
Sustained decode	1019K	2	Requests queued	265.4	183.2	4,621 ms	5.5 ms	1.8	2	7,960	30.0 s
Sustained decode	1019K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	1019K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	1019K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	1019K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	1019K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	1019K	128	Over 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.