The new Geekbench 7 benchmark raises the bar for CPU testing

Primate Labs released Geekbench 7 for macOS, iOS, Android, Windows, and Linux, redesigning its cross-platform benchmark to better reflect how current software uses CPUs and GPUs.

The update changes Geekbench’s multi-core methodology and adds workloads for artificial intelligence, media processing, gaming, and content creation. Primate Labs says the revisions account for increasingly demanding computing tasks and the larger data sets handled by modern devices.

Geekbench 7 remains free for personal use.

Geekbench 7 scores aren’t compatible with Geekbench 6

Geekbench scores can look like a universal measurement of processor performance, but the numbers only have meaning within the version of the benchmark that produced them. A score from Geekbench 7 can be compared with another Geekbench 7 result, but not directly with a result from Geekbench 6.

Because Geekbench 7 changes its workloads, data sets, and multi-core methodology, its results shouldn’t be treated as directly comparable with Geekbench 6 scores.

We ran both versions on the same 11-inch M4 iPad Pro using iPadOS 27. Geekbench 6 reported a 3,719 single-core score and a 13,635 multi-core score, while Geekbench 7 returned 3,197 and 12,959, respectively.

Those results make the same iPad appear about 14% slower in single-core performance and 5% slower in multi-core performance. The hardware didn’t lose performance between tests. Geekbench 7 changed what it measures and how it calculates its scores, demonstrating why results from the two versions can’t be placed on the same performance chart.

Comparing results between Geekbench 6 and Geekbench 7

A device’s Geekbench 6 result should therefore remain in charts and comparisons built around Geekbench 6. New Geekbench 7 scores will need their own baseline as results accumulate across different processors and devices.

Geekbench 7 changes how it measures multi-core performance

The difference between Geekbench 6 and Geekbench 7 results reflects more than a recalibrated scoring scale. Geekbench 7 also changes which workloads contribute to its multi-core score.

Geekbench 7 only runs a workload in multithreaded mode when the task it represents is also multithreaded in real software. Primate Labs says treating every workload as multithreaded can inflate benchmark results without accurately showing how a device handles everyday apps.

Not every real-world task benefits from being divided across all available processor cores. The HTML5 Browser test, for example, isn’t included in the multithreaded suite because Primate Labs says browsers are generally single-threaded or lightly threaded.

The revised approach should make Geekbench 7’s multi-core score more representative of how software uses a processor. It also makes the methodology behind the score more important when comparing devices instead of relying on the number alone.

A new video workload encodes screen-sharing footage with AV1, modeling the technology behind screen-sharing features in videoconferencing apps. Another compresses music and spoken audio with the Opus codec, reflecting work performed by voice-recording and podcast apps.

Geekbench 7 also decodes video and audio while generating live captions with the Whisper speech-recognition model. The workload represents video playback with automatic subtitles enabled rather than testing speech recognition by itself.

Recognizable tasks give Geekbench scores a clearer relationship to common software features.

A new Game Physics workload uses the Jolt Physics engine found in popular video games. Primate Labs has also expanded its Photo Editor test and updated the Photo Library workload to import and process JPEG XL and DNG images.

The recognizable tasks give Geekbench scores a clearer relationship to common software features, although no benchmark can reproduce the performance of every app or workflow.

GPU testing adds AI and content-creation workloads

Geekbench 7 shifts its GPU benchmark toward machine learning, video, and content-creation tasks. New GPU workloads track faces and apply live video effects, upscale images with machine learning, and blur backgrounds in videoconferencing streams.

Each workload reflects a feature commonly used in social media, photo-editing, or communication apps. The benchmark also adds RAW image processing, LUT-based video color grading, path tracing, and fluid simulation.

CUDA joins Metal, OpenCL, and Vulkan as a supported GPU API. The addition allows Geekbench 7 to test Nvidia GPUs through CUDA, while Apple devices continue to use Metal.

Larger data sets make Geekbench 7 more demanding

Primate Labs says computing workloads and data sets have grown more demanding since it released Geekbench 6, prompting the company to increase the amount and variety of data processed by Geekbench 7.

The File Compression test now handles a larger and more varied collection of source code, object code, and text documents. Its PDF Viewer workload includes files ranging from park maps to technical documents and academic papers.

Developer and image-processing workloads also use more assets and additional image formats. The changes are intended to better reflect the files and projects handled by current phones, tablets, and computers.

Geekbench 7 pricing and availability

Geekbench 7 is available now for macOS, iOS, Android, Windows, and Linux. The benchmark is free for personal use across supported platforms.

Primate Labs is also offering 20% off Geekbench 7 Pro through August 6. The paid version adds features intended for professional and commercial use, including command-line tools and automated benchmark deployment.