[Performance]:

[Serenagirl]

OpenVINO Version

2023.3.0

Operating System

Red Hat Enterprise Linux 8

Device used for inference

CPU

OpenVINO installation

Build from source

Programming Language

Python

Hardware Architecture

ARM (64 bits)

Model used

srresnet

Model quantization

No

Target Platform

aarch64，armv8.2-sve ,openvino

Performance issue description

sued 48s one core,but onnxruntime used 22s, and memory used too high.

Step-by-step reproduction

No response

Issue submission checklist

• I'm reporting a performance issue. It's not a question.
• I checked the problem with the documentation, FAQ, open issues, Stack Overflow, etc., and have not found a solution.
• There is reproducer code and related data files such as images, videos, models, etc.

[rkazants]

Hi @Serenagirl,

Can you please try the latest OpenVINO package 2025? For model conversion, you need to use ovc tool: ovc acl_fp16.onnx.
There are many performance enhancement were done for ARM since 2023.3 release:) That is because anyway we can't help you with this old package.

If the latest package does not help to increase performance, you can share with us your model or run benchmark_app and collect performance profile with operation hotspot stats. Please share with us model or benchmark_app data you received.

Best regards,
Roman

[Serenagirl]

Hi @Serenagirl,

Can you please try the latest OpenVINO package 2025? For model conversion, you need to use ovc tool: ovc acl_fp16.onnx. There are many performance enhancement were done for ARM since 2023.3 release:) That is because anyway we can't help you with this old package.

If the latest package does not help to increase performance, you can share with us your model or run benchmark_app and collect performance profile with operation hotspot stats. Please share with us model or benchmark_app data you received.

Best regards, Roman

Thank you so very much but our environment up to python 3.8,I will try a higher version.My colleagues will upload the onnx model later, please see what else can be optimized?

[hhhcr]

acl_fp16.zip
Our MSRResNet input value is (1, 3, 720, 1280), and upscale rate 1.5x

[Serenagirl]

acl_fp16.zip Our MSRResNet input value is (1, 3, 720, 1280), and upscale rate 1.5x

here is the model,and i tried with openvino 2025, 22s and memory used 1470, the performance seems to be the same as that of ONNX 1.19.0.