'Geekbench AI' has been released, which can measure the AI performance of PCs and smartphones, so I tried using it
Primate Labs, the developer of the standard benchmark app '
Geekbench AI 1.0 | Geekbench Blog
https://www.geekbench.com/blog/2024/08/geekbench-ai/
To install Geekbench AI, first click the link below to visit the download page.
Download Geekbench AI
https://www.geekbench.com/ai/download/
Geekbench AI is compatible with macOS, Windows, Linux, Android, and iOS. Since I wanted to measure the performance of a Windows PC this time, I clicked 'Download for Windows.'
Click 'Download for Windows' to start downloading the installer. The file size was 583MB.
Once the installer has finished downloading, double-click it to run it.
Click “Next”.
When the terms of use are displayed, read them carefully and click 'I Agree'.
When the installation folder selection screen appears, click 'Next' without setting anything.
When the Start Menu Registration Settings screen appears, click 'Install' without making any changes.
The installation is now complete. Click 'Finish' and Geekbench AI will start automatically.
The first time you start Geekbench AI, a license authentication screen will be displayed. Geekbench AI can be used for free, but all measurement results of free users will be uploaded to the Geekbench score aggregation site. For this reason, if you do not want to upload your scores, you will need to purchase a license for the paid version 'Geekbench AI Pro'. Since I want to use it for free this time, I clicked 'Later'.
This is the basic screen of Geekbench AI. It shows system information and benchmark settings.
The benchmark settings screen allows you to select the AI framework, AI backend, and processing device available on the running system. On the PC used this time, the AI frame could be selected from 'ONNX' and 'OpenVINO'. The PC's RAM capacity is 16GB, the CPU is '
The available AI backends vary depending on the AI framework.
Additionally, the processing devices available will change depending on the AI framework and backend you choose.
So, select 'ONNX' as the AI framework, 'CPU' as the AI backend, and 'AMD Ryzen 5 7600X' as the processing device, then click 'Run AI Benchmark'.
Then the benchmark began.
Once the benchmark is complete, the results will be automatically uploaded to the score aggregation site and your browser will launch and open the score page.
At the top of the score page, three types of scores are displayed: 'Single Precision Score', 'Half Precision Score', and 'Quantized Score'. The scores this time were 3182 points for single precision, 891 points for half precision, and 4325 points for quantized score.
Scroll down to see your detailed scores.
Next, we will measure the AI processing performance using the GPU. We set the AI framework to 'ONNX', the AI backend to 'DirectML', and the processing device to 'NVIDIA GeForce RTX 3090' to run the benchmark.
The single precision score was 21,928 points, the half precision score was 31,938 points, and the quantization score was 13,525 points.
In addition, the processing device was changed to 'AMD Radeon(TM) Graphics' and the benchmark was run.
The single precision score was 1711 points, the half precision score was 1948 points, and the quantization score was 1301 points. Since the 'AMD Radeon(TM) Graphics' is a GPU built into the CPU, it has a significantly lower score than the 'NVIDIA GeForce RTX 3090'.
The price of Geekbench AI Pro is $99 (about 14,700 yen) and can be purchased from the following link.
Geekbench AI - Primate Labs Store
https://www.primatelabs.com/store/ai/
◆ Forum is currently open
A forum related to this article has been set up on the official GIGAZINE Discord server . Anyone can post freely, so please feel free to comment! If you do not have a Discord account, please refer to the account creation procedure article to create an account!
• Discord | 'Show us your Geekbench AI results!' | GIGAZINE
https://discord.com/channels/1037961069903216680/1273929450442194975
Related Posts:
