Primate Labs, known for its widely-used Geekbench benchmarking tool, has recently introduced Geekbench AI 1.0. This new benchmark tool is designed to assess artificial intelligence performance across a wide range of devices and platforms, marking a significant evolution from the previous Geekbench ML (Machine Learning).
Geekbench has been a go-to tool for evaluating CPU and GPU performance, providing users with valuable insights into how their devices stack up against others in a variety of scenarios. With the launch of Geekbench AI 1.0, the focus has shifted to measuring AI performance, encompassing CPU, GPU, and NPU capabilities for handling real-world machine learning tasks.
The new benchmark tool incorporates ten distinct AI workloads that evaluate device performance in areas like computer vision and natural language processing, using rich datasets that closely resemble real-world applications. One of the key features of Geekbench AI is its multi-dimensional approach to measuring AI performance through three scores: Single Precision, Half Precision, and Quantized, providing a detailed look at how hardware components operate at different levels of precision.
Geekbench AI places a strong emphasis on both speed and accuracy, ensuring that devices not only perform tasks quickly but also achieve accurate results. To facilitate fair comparisons and prevent performance tuning, all workloads run for at least one second, allowing devices to reach their optimal performance levels before results are recorded.
Furthermore, Geekbench AI supports various AI frameworks across different platforms, including OpenVINO and ONNX on Windows, Core ML on Apple devices, and vendor-specific TensorFlow Lite Delegates on Android. This broad support enables developers and users to evaluate AI performance across diverse hardware and software setups.
Overall, with the launch of Geekbench AI 1.0, Primate Labs aims to provide a comprehensive testing suite that offers users a standardized approach to measuring AI performance and empowers them to make informed decisions about their computing devices.