![images 1.fill .size 2000x1125.v1724926773 images 1.fill .size 2000x1125.v1724926773](https://utterbuzz.com/wp-content/uploads/2024/08/images-1.fill_.size_2000x1125.v1724926773-640x360.jpg)
Artificial intelligence, and generative AI in particular, are now such prominent technologies, they’ll often get mentioned before anything else at product launches—from the Pixel 9 series to the Microsoft Copilot+ PC range. These devices are getting more AI features, and more components dedicated to AI tasks. If you’re wondering just how well your computer can handle the most common AI tasks of the moment, turn to the newly launched Geekbench AI 1.0, from the benchmarking experts at Geekbench.
For those of you new to benchmarking, it essentially stress-tests your hardware with specially designed tasks, to help you get an idea of how powerful your system is—and how it compares to others. You can benchmark many different components of your computer, but for this situation, we’re specifically interested in testing for AI performance.
What AI benchmarks are actually testing
![Geekbench AI](https://lifehacker.com/imagery/articles/01J6ETBHVA9GPFG7Y3XRJKY98R/images-1.fill.size_2000x1125.v1724926773.jpg)
Credit: Lifehacker
Run an AI benchmark such as Geekbench AI, and it will look at how well your computer can perform tasks that are normally handled by AI: detecting objects and faces in pictures, generating a new image based on the style of an existing one, or translating between languages. You might use web apps for these jobs, but these benchmarks tell you how well your computer can handle them locally.
These different tests utilize different types of AI, including machine learning (training models on vast amounts of data) and deep learning (a more complex version of machine learning). They also measure different aspects of AI performance: how quickly AI tasks can be processed, for example, as well as how accurate they are.
Modern-day systems now often have dedicated Neural Processing Units (NPUs), processors built for processing AI tasks. These tasks often require different types of calculations and workflows than more general computing processes, so having hardware components made specifically for them can speed everything up (in the same way a GPU or Graphics Processing Unit is designed specifically to handle visuals).
Geekbench AI produces three scores at the end: Full Precision, Half Precision, and Quantized. Full Precision reflects AI tasks where high levels of accuracy are required and most demands are placed on the system, whereas the other two compromise a little on accuracy versus speed (often required if efficiency is important).
“Just as CPU-bound workloads vary in how they can take advantage of multiple cores or threads for performance scaling (necessitating both single-core and multi-core metrics in most related benchmarks), AI workloads cover a range of precision levels, depending on the task needed, the hardware available, and the frameworks in between,” explains Geekbench’s John Poole in a blog post.
How to run Geekbench AI on your computer
![Geekbench AI](https://lifehacker.com/imagery/articles/01J6ETBHVA9GPFG7Y3XRJKY98R/images-2.fill.size_2000x1125.v1724926773.jpg)
Credit: Lifehacker
Head to the Geekbench AI download page to find the packages for Windows, macOS, and Linux (as well as links to the Google Play Store and the Apple App Store if you want to run benchmarks on your mobile devices). All of these programs are straightforward to use, even if you have no experience running benchmarking software.
The main Geekbench AI interface on the desktop shows some information about your system first of all, including the operating system you’re running and the processor you have installed. You then get a description of what the benchmark is going to cover. (Note that it’ll take several minutes to complete.)
You then get drop-down menus to configure the benchmark. First is the AI Framework, which is essentially the AI toolset you want to use for the test: The ones you’ll see will depend on your system, and may include ONNX (the open-source Open Neural Network Exchange) and Core ML (the AI framework used on Apple hardware).
Then there’s the AI Backend, which means the system component you want to test the AI capabilities of. This will depend on what’s inside your computer, but you will see CPU, and perhaps GPU and NPU (or Neural Engine) too. On some systems, you’ll also see an AI Device option, which lets you switch between available processors.
When you’re happy with the way the benchmark is configured, click Run AI Benchmark to set it in motion. Your system’s results will pop up in a web browser, and you can check out other results posted from other devices through the same portal. Note that scores are calibrated against a baseline of 1,500, which reflects the performance of an Intel Core i7-10700 processor. Higher scores indicate more capable AI handling.