Primate Labs Published Geekbench AI First version of benchmarking. Although the name is new, the software is Geekbench ML rebranded.
Geekbench ML, which includes computer vision and natural language processing tests, has been in preview for some time, as learned from the official Primate Labs blog. The developers realized that the term ML (machine learning) didn't quite fit the promotional tool, theIn its place is the more marketing-friendly Geekbench AI (Artificial Intelligence) test.
The Geekbench AI 1.0 benchmarks support cross-platform, but since AI algorithms can perform very differently on different platforms, users will also see different scores in different tests. As a result, the AI benchmark leaderboard will be divided into three groups:CPUs, GPUs and NPUs(Neural Processing Unit).
NPUs have been the focus of attention for all processor manufacturers as their goal is to ensure that modern processors achieve high inference speeds with low power consumption, and NPUs are also a requirement for Microsoft Copilot+PC certification.
Primate Labs explains thatAI benchmarking is much more complex than GPU testingThe benchmarks need to measure more than just raw computing power. Over time, developers add new frameworks and API support, which must also be reflected in benchmarks.
Ultimately, the Geekbench AI score will be divided into three sub-scores:Full-precision scores, half-precision scores, and quantized scores.
Meanwhile, users can browse the official Geekbench AI rankings to see the fastest AI hardware.
The Geekbench AI benchmarks are now available as a free trial. Geekbench AI 1.0 is available through the official Windows, macOS, and Linux download pages, as well as the Google Play Store and Apple App Store.