Primate Labs has unveiled Geekbench 7, the latest iteration of its cross-platform benchmarking tool, now available for macOS, iOS, Android, Windows, and Linux. This update aims to more accurately represent contemporary computing demands by incorporating workloads that reflect real-world applications.
One of the significant changes in Geekbench 7 is the revision of its multi-core testing methodology. Unlike previous versions that treated all workloads as multithreaded, Geekbench 7 now only runs tasks in multithreaded mode if they are multithreaded in actual software. This approach prevents inflated benchmark results and offers a more realistic assessment of how devices handle everyday applications.
The new version introduces several CPU workloads designed to mirror common computing tasks. These include encoding screen-sharing footage with the AV1 codec, compressing audio using the Opus codec, and generating live captions through the Whisper speech-recognition model. Additionally, a Game Physics workload utilizing the Jolt Physics engine has been added, along with expanded Photo Editor tests and updated Photo Library workloads that support JPEG XL and DNG image formats.
In the realm of GPU testing, Geekbench 7 shifts its focus toward machine learning and content creation tasks. New GPU workloads encompass face tracking, live video effects application, image upscaling via machine learning, and background blurring in video conferencing streams. The benchmark also introduces RAW image processing, LUT-based video color grading, path tracing, and fluid simulation. Notably, CUDA support has been added alongside existing Metal, OpenCL, and Vulkan APIs, enabling testing of Nvidia GPUs through CUDA while Apple devices continue to use Metal.
To better reflect the increasing demands of modern computing, Geekbench 7 processes larger and more varied data sets. For instance, the File Compression test now handles a diverse collection of source code, object code, and text documents, while the PDF Viewer workload includes files ranging from park maps to technical documents and academic papers. Developer and image-processing workloads also utilize more assets and additional image formats, aligning with the types of files and projects managed by current devices.
It’s important to note that Geekbench 7 scores are not directly comparable to those from Geekbench 6 due to changes in workloads, data sets, and multi-core methodology. For example, tests conducted on the same 11-inch M4 iPad Pro using iPadOS 27 yielded a single-core score of 3,719 and a multi-core score of 13,635 in Geekbench 6, whereas Geekbench 7 reported scores of 3,197 and 12,959, respectively. This discrepancy underscores the need to establish new baselines for Geekbench 7 results.
Geekbench 7 is available for free for personal use across all supported platforms. For professional and commercial users, Geekbench 7 Pro is offered at a discounted price of $79 until August 6, after which it will be priced at $99. The Pro version includes features such as command-line tools and automated benchmark deployment.
By integrating AI workloads and refining its multi-core testing approach, Geekbench 7 provides a more accurate and relevant measure of device performance in today’s computing landscape. As artificial intelligence and machine learning become increasingly integral to various applications, benchmarks like Geekbench 7 are essential for evaluating how well hardware can handle these evolving workloads.
