Speed test matrix

Seed-OSS-36B-Instruct-FP8-Static on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP8 static7 contexts (0–320K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 19:03 UTC

Hardware

Type: Discrete GPU
GPU: CMP 170HX
VRAM: 64 GB
CPU: 2x Intel Xeon E5-2630 v3
RAM: 64 GB
OS: Ubuntu 26.04.1

Engine

Engine: vllm
Version: 0.32.0rc1.dev249+g18305948a
Quantization: FP8 static
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: ghcr.io/morrowmake/vllm-cmp170hx@sha256:5b45c4d6feafefe2eba962375b0a9a0b64fd22b9e05abb4d1966f14a32a2db79

Run

Served model: bench
Revision: main
Benchmark tool: llm-inference-bench v0.8.0
Decode test: Fixed duration
Prefill mode: Measured during decode warm-up
Time per measurement: 30 s
Max output: 8,192 tokens

Launch command

# Untuned out-of-the-box baseline. Spec decode: none. KV cache: int4_per_token_head
# 1x NVIDIA CMP 170HX 64GB (cmpunlocker), 180 W limit, PCIe Gen2 x4 | 2x Intel Xeon E5-2630 v3 | 64 GB DDR4 | Ubuntu 26.04.1, NVIDIA driver 610.57.04 | vLLM: Morrowmake vllm-cmp170hx image
vllm serve AngelSlim/Seed-OSS-36B-Instruct-FP8-Static --max-model-len 335872 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code  --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context573.014,321 ms TTFT

16K context344.047,236 ms TTFT

32K context195.0165,986 ms TTFT

64K context153.0423,147 ms TTFT

Sustained decode

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

Context	1	2	4	8	16	32	64	128
0	26.942,318 ms TTFT	52.728,595 ms TTFT	101.6306 ms TTFT	191.4365 ms TTFT	344.8418 ms TTFT	559.6611 ms TTFT	Over capacity	Over capacity
16K	19.7289 ms TTFT	33.0440 ms TTFT	53.5539 ms TTFT	78.9726 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	15.7Requests queued476 ms TTFT	24.1Requests queued650 ms TTFT	36.6924 ms TTFT	49.71,343 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	3.4Requests queued21,454 ms TTFT	15.3Requests queued1,311 ms TTFT	22.21,822 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
256K	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
320K	Over capacity	Over capacity	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	573.0	—	14,321 ms	—	—	1	—	—
Prefill	16K	1	Valid	344.0	—	47,236 ms	—	—	1	—	—
Prefill	32K	1	Valid	195.0	—	165,986 ms	—	—	1	—	—
Prefill	64K	1	Valid	153.0	—	423,147 ms	—	—	1	—	—
Sustained decode	0	1	Valid	26.9	26.9	42,318 ms	37.1 ms	1.0	0	807	30.0 s
Sustained decode	0	2	Valid	52.7	26.4	28,595 ms	37.9 ms	2.0	0	1,580	30.0 s
Sustained decode	0	4	Valid	101.6	25.5	306 ms	39.2 ms	4.0	0	3,044	30.0 s
Sustained decode	0	8	Valid	191.4	24.1	365 ms	41.5 ms	8.0	0	5,736	30.0 s
Sustained decode	0	16	Valid	344.8	21.9	418 ms	45.8 ms	16.0	0	10,336	30.0 s
Sustained decode	0	32	Valid	559.6	17.7	611 ms	56.5 ms	32.0	0	16,768	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	19.7	19.8	289 ms	50.6 ms	1.0	0	592	30.0 s
Sustained decode	16K	2	Valid	33.0	16.5	440 ms	60.6 ms	2.0	0	988	30.0 s
Sustained decode	16K	4	Valid	53.5	13.4	539 ms	74.5 ms	4.0	0	1,604	30.0 s
Sustained decode	16K	8	Valid	78.9	9.9	726 ms	101.4 ms	8.0	0	2,360	29.9 s
Sustained decode	16K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Requests queued	15.7	15.7	476 ms	63.6 ms	1.0	0	470	29.9 s
Sustained decode	32K	2	Requests queued	24.1	12.1	650 ms	82.7 ms	1.9	0	723	30.0 s
Sustained decode	32K	4	Valid	36.6	9.2	924 ms	109.3 ms	4.0	0	1,096	29.9 s
Sustained decode	32K	8	Valid	49.7	6.2	1,343 ms	161.4 ms	8.0	0	1,488	29.9 s
Sustained decode	32K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Requests queued	3.4	11.3	21,454 ms	88.7 ms	0.7	0	103	30.0 s
Sustained decode	64K	2	Requests queued	15.3	7.8	1,311 ms	127.9 ms	1.9	0	457	29.9 s
Sustained decode	64K	4	Valid	22.2	5.6	1,822 ms	179.5 ms	4.0	0	664	29.9 s
Sustained decode	64K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Over capacity	—	—	—	—	0.8	0	0	30.0 s
Sustained decode	128K	2	Over capacity	—	—	—	—	0.8	0	0	30.0 s
Sustained decode	128K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	8	Over capacity	—	—	—	—	—	—	—	—
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	256K	1	Over capacity	—	—	—	—	0.1	0	0	30.0 s
Sustained decode	256K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	256K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	256K	8	Over capacity	—	—	—	—	—	—	—	—
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	320K	1	Over capacity	—	—	—	—	1.0	0	0	30.0 s
Sustained decode	320K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	320K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	320K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	320K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	320K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	320K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	320K	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.