Primate Labs has launched Geekbench AI 1.0, a new benchmark tool designed to evaluate artificial intelligence performance across various devices and platforms. This release marks the evolution of the previously known Geekbench ML (Machine Learning) into a more comprehensive AI testing suite.
Geekbench is best known for its comprehensive benchmarking of CPUs and GPUs, providing users with clear and comparative performance scores. Chances are you have seen them featured prominently in product reviews and marketing materials of your favorite computing devices. These scores help consumers and professionals alike assess how well their devices will perform in real-world scenarios, from gaming and content creation to software development.
Now, Geekbench AI 1.0 aims to bring that same standardized approach to measuring AI performance across various platforms and devices. The benchmarks evaluate how well a device handles real-world machine learning tasks using CPU, GPU, and NPU (neural processing unit) capabilities. Primate Labs has integrated ten different AI workloads that measure device performance in areas like computer vision and natural language processing, using large datasets that closely mimic real-world use cases.
One of the critical elements of Geekbench AI is its multidimensional approach to evaluating AI performance. The benchmark uses three scores: Single Precision, Half Precision, and Quantized. These scores provide insight into how different hardware components perform at various levels of data precision, from high-performance tasks requiring exact outputs to those that can tolerate less precision. Importantly, Geekbench AI also measures accuracy alongside speed, ensuring that performance isn’t just about how fast a task is completed but how reliably the model achieves correct results.
Geekbench AI supports multiple AI frameworks across 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 assess AI performance across various hardware and software configurations.
A key feature of Geekbench AI is its focus on both speed and accuracy. The benchmark measures how quickly a device can complete AI tasks and how accurately it performs them, providing a more holistic view of AI performance.
In an attempt to make fair comparisons and prevent performance tuning, all workloads in Geekbench AI run for a minimum of one second. This ensures that devices reach their peak performance before results are recorded, offering a more accurate reflection of real-world performance.
You can download Geekbench AI now for Windows, macOS, Linux, Android, and iOS.