Primate Labs Releases Geekbench 7 With AI Workloads
Primate Labs released Geekbench 7 on July 23 for macOS, iOS, Android, Windows, and Linux, adding AI, media, gaming, and content-creation workloads. Tom's Hardware reports that the redesigned multi-core test now runs workloads in multi-threaded mode only when the modeled application does so. Geekbench 7 scores use a new 2,500-point Ryzen 7 7700 reference baseline and cannot be compared directly with Geekbench 6 results.
Primate Labs released Geekbench 7 on July 23 for macOS, iOS, Android, Windows, and Linux, revising the cross-platform CPU and GPU benchmark with workloads for AI, media processing, gaming, and content creation. The release changes both the workload mix and the scoring baseline, meaning Geekbench 7 results are not directly comparable with Geekbench 6 scores.
Tom's Hardware reports that the multi-core benchmark no longer forces every workload across all available CPU cores. Instead, a workload runs in multi-threaded mode only when the real-world application being modeled would use multiple threads. The publication notes that Geekbench removed the HTML5 browser test from the multi-core suite because browsers are generally single- or lightly threaded.
Updated CPU and media tests
VideoCardz reports that Geekbench 7 adds dedicated Photo Editor, Game Physics, Video Encoder, Audio Encoder, and Video Decoder workloads. The Game Physics test uses the Jolt Physics engine, while the media tests use AV1 video, Opus audio, and the Whisper speech-recognition model.
The workload refresh also consolidates or removes several Geekbench 6 tests. According to VideoCardz, Object Remover, Horizon Detection, and Photo Filter are covered by the broader Photo Editor workload, while separate Object Detection and Background Blur tests have been removed. Photo Library now imports images, generates thumbnails, and detects objects with the MobileNetV1 SSD model.
Other updates include Python 3.13 for Text Processing, ASTC and BC7 textures alongside Draco geometry in Asset Compression, and LZ4, zlib, and Zstandard in File Compression, VideoCardz reports.
Only eight workloads contribute to Geekbench 7's multi-core score: File Compression, Photo Library, Clang, Text Processing, Asset Compression, HDR, Photo Editor, and Ray Tracer. VideoCardz reports that Navigation, HTML5 Browser, PDF Viewer, Structure from Motion, Game Physics, and the media workloads appear only in the single-core portion.
GPU focus expands to machine learning
MacRumors reports that the revised GPU benchmark includes machine-learning and content-creation tests such as face tracking, real-time video filters, ML image upscaling, and video-conferencing background blur. It also includes RAW image processing, LUT-based video color grading, path tracing, and fluid simulation.
Tom's Hardware reports that Geekbench 7 adds CUDA support, broadening the available GPU testing paths for Nvidia hardware. The revised suite focuses on machine-learning, image and video processing, and graphics workloads.
For practitioners evaluating laptops, workstations, or edge devices, the change makes version discipline especially important. A benchmark score is useful only relative to the workload definition and reference system behind it.
New reference system resets score scale
Geekbench 7 uses a 2,500-point reference score based on a Lenovo Legion system with an AMD Ryzen 7 7700, according to VideoCardz. Geekbench 6 used a Dell Precision 3460 with an Intel Core i7-12700 as its reference system. A score of 5,000 represents twice the measured performance of the Geekbench 7 reference system, not a target score for every Ryzen 7 7700 machine.
AppleInsider tested both versions on an 11-inch M4 iPad Pro running iPadOS 27. It reported Geekbench 6 scores of 3,719 single-core and 13,635 multi-core, compared with 3,197 and 12,959 in Geekbench 7. Those results illustrate why cross-version comparisons can misstate hardware changes: altered workloads, datasets, threading rules, and the baseline all affect the resulting numbers.
Geekbench 7 remains free for personal use, while Tom's Hardware reports that Geekbench 7 Pro is discounted 20% through August 6.
Key Points
- 1Geekbench 7 changes multi-core execution rules, making its CPU score more workload-aware but discontinuous with Geekbench 6 historical results.
- 2New tests incorporate Whisper, MobileNetV1 SSD, AV1, and GPU image processing, bringing common AI-adjacent client workloads into the suite.
- 3Practitioners comparing devices should treat Geekbench 7 as a separate measurement series because its workload mix and Ryzen-based calibration changed.
Scoring Rationale
Geekbench is widely used for cross-platform CPU and GPU comparisons, and version 7 changes how multi-core performance and AI-adjacent workloads are measured. The release is particularly relevant to practitioners evaluating client hardware, but it is a benchmarking-tool update rather than a new model, platform, or infrastructure deployment.
Sources
Primary source and supporting public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
