Speed test matrix

Qwen3.5-397B-A17B-GPTQ-Int4 on RTX PRO 6000 Blackwell Server Edition 96GB ×4

vllm · 0.30.1rc1.dev452+g73a583112 · INT46 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Oct 13, 2026, 17:27 UTC

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: 0.30.1rc1.dev452+g73a583112
Quantization: INT4
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2
Commit: 73a5831127a9

Run

Served model: qwen
Revision: main
Benchmark tool: llm-inference-bench v0.7.9
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_rtxpro6k-qwen-vllm-73a5831127a9/bin/vllm" serve Qwen/Qwen3.5-397B-A17B-GPTQ-Int4 --served-model-name qwen --trust-remote-code --tensor-parallel-size 4 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.91 --host 0.0.0.0 --port 8000 --api-key <redacted> --attention-backend FLASH_ATTN --linear-backend marlin --moe-backend triton --quantization moe_wna16

Prefill

prompt tok/s by context length

8K context5,1651,586 ms TTFT

16K context5,1803,131 ms TTFT

32K context5,4525,923 ms TTFT

64K context5,51711,676 ms TTFT

128K context5,15824,951 ms TTFT

251K context4,63954,360 ms TTFT

256K context4,59855,939 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	64.4336 ms TTFT	89.5503 ms TTFT	125.5876 ms TTFT	183.61,335 ms TTFT	273.31,736 ms TTFT	405.13,070 ms TTFT	631.95,789 ms TTFT	939.09,913 ms TTFT
16K	64.1305 ms TTFT	89.3475 ms TTFT	124.4682 ms TTFT	175.9917 ms TTFT	262.41,421 ms TTFT	406.91,662 ms TTFT	616.23,381 ms TTFT	Over capacity
32K	63.4372 ms TTFT	87.9545 ms TTFT	124.1871 ms TTFT	177.11,301 ms TTFT	266.11,651 ms TTFT	365.83,195 ms TTFT	Over capacity	Over capacity
64K	62.0180 ms TTFT	85.8235 ms TTFT	121.2322 ms TTFT	182.3537 ms TTFT	245.6979 ms TTFT	Over capacity	Over capacity	Over capacity
128K	59.4633 ms TTFT	82.2865 ms TTFT	116.71,307 ms TTFT	163.72,360 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	55.21,007 ms TTFT	76.51,273 ms TTFT	107.21,786 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over 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.

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	5,165.0	—	1,586 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,180.0	—	3,131 ms	—	—	1	—	—
Prefill	32K	1	Valid	5,452.0	—	5,923 ms	—	—	1	—	—
Prefill	64K	1	Valid	5,517.0	—	11,676 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,158.0	—	24,951 ms	—	—	1	—	—
Prefill	251K	1	Valid	4,639.0	—	54,360 ms	—	—	1	—	—
Prefill	256K	1	Valid	4,598.0	—	55,939 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	64.4	64.4	336 ms	15.5 ms	1.0	0	1,932	30.0 s
Sustained decode	8K	2	Valid	89.5	44.6	503 ms	22.4 ms	2.0	0	2,686	30.0 s
Sustained decode	8K	4	Valid	125.5	32.1	876 ms	31.2 ms	4.0	0	3,764	30.0 s
Sustained decode	8K	8	Valid	183.6	24.3	1,335 ms	41.1 ms	8.0	0	5,504	30.0 s
Sustained decode	8K	16	Valid	273.3	18.0	1,736 ms	55.6 ms	16.0	0	8,192	30.0 s
Sustained decode	8K	32	Valid	405.1	13.1	3,070 ms	76.1 ms	32.0	0	12,128	29.9 s
Sustained decode	8K	64	Valid	631.9	9.7	5,789 ms	103.3 ms	64.0	0	18,880	29.9 s
Sustained decode	8K	128	Valid	939.0	6.3	9,913 ms	159.0 ms	128.0	0	28,032	29.9 s
Sustained decode	16K	1	Valid	64.1	64.1	305 ms	15.6 ms	1.0	0	1,922	30.0 s
Sustained decode	16K	2	Valid	89.3	44.5	475 ms	22.5 ms	2.0	0	2,678	30.0 s
Sustained decode	16K	4	Valid	124.4	32.1	682 ms	31.1 ms	4.0	0	3,728	30.0 s
Sustained decode	16K	8	Valid	175.9	23.5	917 ms	42.6 ms	8.0	0	5,272	30.0 s
Sustained decode	16K	16	Valid	262.4	18.2	1,421 ms	55.1 ms	16.0	0	7,856	29.9 s
Sustained decode	16K	32	Valid	406.9	13.4	1,662 ms	74.5 ms	32.0	0	12,192	30.0 s
Sustained decode	16K	64	Valid	616.2	9.8	3,381 ms	102.0 ms	64.0	0	18,432	29.9 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	63.4	63.4	372 ms	15.8 ms	1.0	0	1,902	30.0 s
Sustained decode	32K	2	Valid	87.9	43.8	545 ms	22.8 ms	2.0	0	2,636	30.0 s
Sustained decode	32K	4	Valid	124.1	32.0	871 ms	31.2 ms	4.0	0	3,720	30.0 s
Sustained decode	32K	8	Valid	177.1	23.8	1,301 ms	42.0 ms	8.0	0	5,312	30.0 s
Sustained decode	32K	16	Valid	266.1	17.7	1,651 ms	56.5 ms	16.0	0	7,968	29.9 s
Sustained decode	32K	32	Valid	365.8	12.3	3,195 ms	81.4 ms	32.0	0	10,944	29.9 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	62.0	62.0	180 ms	16.1 ms	1.0	0	1,860	30.0 s
Sustained decode	64K	2	Valid	85.8	42.9	235 ms	23.3 ms	2.0	0	2,574	30.0 s
Sustained decode	64K	4	Valid	121.2	30.3	322 ms	33.0 ms	4.0	0	3,632	30.0 s
Sustained decode	64K	8	Valid	182.3	22.8	537 ms	43.9 ms	8.0	0	5,464	30.0 s
Sustained decode	64K	16	Valid	245.6	15.5	979 ms	64.6 ms	16.0	0	7,360	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	59.4	59.4	633 ms	16.8 ms	1.0	0	1,780	30.0 s
Sustained decode	128K	2	Valid	82.2	40.9	865 ms	24.4 ms	2.0	0	2,466	30.0 s
Sustained decode	128K	4	Valid	116.7	29.2	1,307 ms	34.2 ms	4.0	0	3,500	30.0 s
Sustained decode	128K	8	Valid	163.7	20.9	2,360 ms	47.8 ms	8.0	0	4,904	30.0 s
Sustained decode	128K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	1	Valid	55.2	55.3	1,007 ms	18.1 ms	1.0	0	1,655	30.0 s
Sustained decode	251K	2	Valid	76.5	38.0	1,273 ms	26.3 ms	2.0	0	2,294	30.0 s
Sustained decode	251K	4	Valid	107.2	26.4	1,786 ms	37.9 ms	4.0	0	3,212	30.0 s
Sustained decode	251K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	128	Over capacity	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.