Geekbench 7 Launches With Overhauled CPU and AI Tests

What's new in Geekbench 7

Primate Labs has released Geekbench 7, the newest version of its widely used cross-platform benchmarking tool, and it represents the most substantial rework of the software in years. The update replaces older synthetic workloads with tests meant to mirror how people actually use their phones, tablets, and computers day to day, spanning everyday productivity tasks, gaming, and increasingly common AI-driven workflows 124.

The overhaul touches nearly every part of the benchmark. CPU tests have been redesigned around more realistic workloads, multi-core testing has been restructured, and new media-focused tasks have been added to reflect modern content creation and consumption habits 1. GPU testing also gets an upgrade, with expanded support that now includes CUDA for Nvidia graphics cards, letting Nvidia hardware be measured using its native compute framework rather than relying solely on cross-platform APIs 1. Dataset sizes used in testing have also been increased, which developers say allows the benchmark to better stress modern hardware that has grown considerably more powerful since the last major revision 1.

A central theme across the update is artificial intelligence. Geekbench 7 introduces dedicated AI benchmarks, acknowledging that machine learning tasks — from image processing to on-device inference — have become a standard part of how consumer devices are used, not a niche feature 12. Both 9to5Mac and Macworld frame this as part of a broader effort to make the tool's scores reflect real-world relevance rather than abstract computational throughput 24.

Why the update matters

Geekbench has long served as one of the most widely cited cross-platform benchmarks, used by reviewers, manufacturers, and enthusiasts to compare performance across Windows, macOS, Android, and iOS devices. As chipmakers increasingly design silicon around AI acceleration and media processing rather than raw clock speed alone, a benchmark still built around older assumptions risks giving misleading or incomplete comparisons. By rebuilding its CPU, GPU, and multi-core tests around contemporary use cases, Primate Labs is attempting to keep the tool aligned with how devices are actually marketed and used in 2024 and beyond 14.

The addition of CUDA support is particularly notable for GPU comparisons, since Nvidia's compute ecosystem has historically been tested through more generic cross-platform APIs in Geekbench, potentially understating performance available to Nvidia hardware specifically 1. Expanding to native CUDA testing gives a more accurate picture of Nvidia GPU capability, which matters for an industry where GPU compute is increasingly tied to AI workloads rather than just graphics rendering.

Where the reporting agrees

All three outlets covering the Geekbench update — Yahoo Tech, 9to5Mac, and Macworld — describe the same core set of changes: refreshed CPU and GPU workloads, a redesigned multi-core test, new AI-focused benchmarks, and a general shift toward measuring real-world usage patterns like gaming and everyday tasks rather than purely synthetic computation 124. There is no meaningful disagreement among these three on what changed or why. Each frames the release as the tool's biggest revision in some time, driven by the need to keep pace with how modern hardware is actually used, especially as AI processing becomes a standard workload rather than a specialty feature 124.

Where it doesn't

The Geekbench coverage itself shows no factual conflicts — the outlets diverge mainly in emphasis rather than substance. Yahoo Tech and Macworld give more attention to the CUDA and dataset-size changes, while 9to5Mac frames the release more around the everyday-use and gaming angle 124. None of that amounts to a real dispute; it reflects different editorial priorities covering the same product announcement.

The other two items included alongside this story are unrelated to the Geekbench release and use "benchmark" in an entirely different sense. One concerns DC Studios failing to hit a box-office or critical milestone that Marvel Studios sustained for over a decade 3, and another concerns the 10-year Treasury yield potentially testing the psychologically significant 5% threshold 5. Neither has any bearing on software benchmarking and should not be read as corroborating or conflicting with the Geekbench reporting — they simply share the word "benchmark" in a different context.

The takeaway

On the actual subject at hand, the evidence is consistent: Geekbench 7 is a ground-up refresh of how the tool measures CPU, GPU, and AI performance, built to match how people use devices now rather than a decade ago. The convergence across Yahoo Tech, 9to5Mac, and Macworld on the same list of changes leaves little room for doubt about what shipped, even if each outlet chose to highlight slightly different pieces of it.