Primate Labs revamps its machine learning benchmark to introduce Geekbench AI 1.0

Primate Labs has launched Geekbench AI 1.0, the latest iteration of its Geekbench ML, Primate Labs' benchmark for machine learning, deep learning, and AI-centric workloads. Although renaming an existing benchmark may seem superficial, there is some backstory. Considering that most companies and industries that make the heaviest use of these workloads increasingly use the term "AI" to refer to and market them, the choice to rename the benchmark is meant to signal that Geekbench AI is relevant to everyone from engineers to performance enthusiasts.

Some features of the Geekbench AI 1.0 include the decision to list three overall scores (full precision, half precision, and quantized) to reflect the complexities associated with the complexity of AI hardware and software ecosystems, capturing real-world AI performance; and provide an accuracy measure for each test to reflect how the size of a data type may increase the speed at the expense of sacrificing accuracy. Additionally, Geekbench AI 1.0 supports more frameworks and leverages more extensive datasets to reflect the tools available to developers and real-world inputs in AI use cases. Finally, each test runs on the device for at least a full second to more accurately reflect maximum performance levels while capturing the "bursty nature of real-world use cases."

The benchmark is available for Windows, macOS, Linux, Android, and iOS, allowing cross-platform comparisons. Since Geekbench AI is integrated into the Geekbench Browser, results can be shared and compared through the platform, featuring a dedicated Geekbench AI Benchmark Chart comparing the top-performing devices and a Latest Geekbench AI Results page compiling the most up-to-date results available.