Speed test matrix

Step-3.7-Flash-148B on CMP 170HX 64GB ×2

vllm · 0.32.0rc1.dev249+g18305948a · NVFP46 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 12:30 UTC

Hardware

Type: Discrete GPU
GPU: CMP 170HX
GPU count: 2
VRAM: 64 GB (2 × 32 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: NVFP4
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: MTP, 1 tokens lookahead. Layout: PP2 (stock vLLM flag)
# 2x 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 0xSero/Step-3.7-Flash-148B --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --pipeline-parallel-size 2 --language-model-only --hf-overrides {"text_config":{"tie_word_embeddings":false}} --speculative-config {"method":"mtp","num_speculative_tokens":1}

Prefill

prompt tok/s by context length

8K context4,5351,786 ms TTFT

16K context5,0923,148 ms TTFT

32K context4,9596,435 ms TTFT

64K context4,44414,329 ms TTFT

128K context3,58335,504 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	58.7108 ms TTFT	121.3108 ms TTFT	204.1156 ms TTFT	340.7239 ms TTFT	504.9330 ms TTFT	781.0447 ms TTFT	Over capacity	Over capacity
16K	59.0186 ms TTFT	103.5210 ms TTFT	196.0333 ms TTFT	328.2520 ms TTFT	505.7717 ms TTFT	Over capacity	Over capacity	Over capacity
32K	61.6296 ms TTFT	102.9351 ms TTFT	192.7551 ms TTFT	316.4910 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	58.7445 ms TTFT	103.9606 ms TTFT	191.1880 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	59.0870 ms TTFT	105.41,008 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	58.11,630 ms TTFT	Over capacity	Over capacity	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	4,535.0	—	1,786 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,092.0	—	3,148 ms	—	—	1	—	—
Prefill	32K	1	Valid	4,959.0	—	6,435 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,444.0	—	14,329 ms	—	—	1	—	—
Prefill	128K	1	Valid	3,583.0	—	35,504 ms	—	—	1	—	—
Sustained decode	0	1	Valid	58.7	58.6	108 ms	17.1 ms	1.0	0	1,759	30.0 s
Sustained decode	0	2	Valid	121.3	60.4	108 ms	16.6 ms	2.0	0	3,638	30.0 s
Sustained decode	0	4	Valid	204.1	50.7	156 ms	19.7 ms	4.0	0	6,120	30.0 s
Sustained decode	0	8	Valid	340.7	42.5	239 ms	23.5 ms	8.0	0	10,219	30.0 s
Sustained decode	0	16	Valid	504.9	32.2	330 ms	31.1 ms	16.0	0	15,130	30.0 s
Sustained decode	0	32	Valid	781.0	25.0	447 ms	40.0 ms	32.0	0	23,425	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	59.0	59.1	186 ms	16.9 ms	1.0	0	1,771	30.0 s
Sustained decode	16K	2	Valid	103.5	52.0	210 ms	19.2 ms	2.0	0	3,104	30.0 s
Sustained decode	16K	4	Valid	196.0	49.7	333 ms	20.1 ms	4.0	0	5,879	30.0 s
Sustained decode	16K	8	Valid	328.2	41.3	520 ms	24.2 ms	8.0	0	9,844	30.0 s
Sustained decode	16K	16	Valid	505.7	31.9	717 ms	31.3 ms	16.0	0	15,166	30.0 s
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	61.6	61.8	296 ms	16.2 ms	1.0	0	1,848	30.0 s
Sustained decode	32K	2	Valid	102.9	51.6	351 ms	19.4 ms	2.0	0	3,086	30.0 s
Sustained decode	32K	4	Valid	192.7	48.3	551 ms	20.7 ms	4.0	0	5,779	30.0 s
Sustained decode	32K	8	Valid	316.4	39.9	910 ms	25.1 ms	8.0	0	9,490	30.0 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	Valid	58.7	59.2	445 ms	16.9 ms	1.0	0	1,762	30.0 s
Sustained decode	64K	2	Valid	103.9	51.9	606 ms	19.3 ms	2.0	0	3,115	30.0 s
Sustained decode	64K	4	Valid	191.1	48.1	880 ms	20.8 ms	4.0	0	5,732	30.0 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	Valid	59.0	59.3	870 ms	16.9 ms	1.0	0	1,769	30.0 s
Sustained decode	128K	2	Valid	105.4	53.3	1,008 ms	18.8 ms	2.0	0	3,159	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	248K	1	Valid	58.1	58.8	1,630 ms	17.0 ms	1.0	0	1,743	30.0 s
Sustained decode	248K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	128	Over capacity	—	—	—	—	—	—	—	—

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