Popular benchmarking utility Geekbench has released a new cross-platform tool to evaluate a device's performance in AI-intensive workloads. Geekbench AI measures a device's CPU, GPU and NPU (Neural Processing Unit) to determine how well it can handle machine learning applications.
Primate Labs, the developer of Geekbench, has been working on the software under the name Geekbench ML, which was released as a preview in 2021, but was renamed to AI for obvious reasons. To find out how different hardware responds to different AI-related tasks, it supports various frameworks such as ONNX, CoreML, TensorFlow Lite, and OpenVINO, and evaluates their performance based on both accuracy and speed.
Three scores are provided: full precision, half precision, and quantized. Primate Labs says the score also includes an accuracy measure to assess how close the workload's output is to the truth, or “whether the model performs as expected.”
We'll need to spend more time with devices running AI on them, like the Copilot Plus PC and all the new smartphones, to see how performance on real-world tasks correlates with the Geekbench AI numbers. Checking frame rates and load times is one thing, but now we might be checking the accuracy of predictive text or what kind of results a generative AI-enabled image editor produces.
You can download the tool now and try it out on Windows, macOS, Linux, Android, and iOS.
