Primate Labs has launched Geekbench 7, a major revision of its widely used cross-platform benchmarking tool. The release delivers a redesigned CPU test suite, new workloads for media and gaming, a significantly overhauled GPU benchmark with CUDA support, and an important shift in how results should be understood.
Rethinking multi-core measurement
One of the most consequential changes lies in how the software handles multi-threaded performance. Geekbench 7 now activates a workload in multi-threaded mode only if the real-world application it models actually scales across cores in that fashion. The HTML5 Browser workload, for instance, is excluded from the multi-core suite because modern browsers remain predominantly single-threaded or lightly threaded. According to Primate Labs, this approach yields multi-core scores that more faithfully track genuine user experience, rather than simply rewarding chips that throw extra cores and threads at every task.
A fresh baseline and early leaderboard
The benchmark also resets its scoring foundation. The AMD Ryzen 7 7700 now serves as the reference processor, normalized to 2,500 points. This replaces the Intel Core i7-12700, which anchored Geekbench 6 at the same 2,500-point mark. As a result, scores from the two generations should not be compared directly; a numerical increase or decrease does not automatically signify an equivalent hardware gain.
Preliminary figures released by Primate Labs sketch an initial CPU hierarchy. AMD’s 16-core Ryzen AI Max+ 395 averages roughly 2,742 points in single-core and 27,412 points in multi-core. Intel’s 24-core Core i9-13900KF posts approximately 2,734 and 24,439 points, while Apple’s 14-core M4 Pro reaches around 3,360 single-core and 23,188 multi-core. Qualcomm’s 18-core Snapdragon X2 Elite sits close behind at about 3,225 and 22,735 points respectively. Apple’s M5 records a notable single-core result of roughly 3,753 points alongside a multi-core score of 18,722.
The update further expands its workload library with AV1 encoding for screen sharing, Opus audio compression, and live caption generation using OpenAI’s Whisper speech recognition model. A new Game Physics test built on the Jolt Physics engine joins enhanced Photo Editor and Photo Library scenarios.
GPU overhaul and CUDA arrival
The GPU benchmark has received an equally comprehensive rework, introducing tests for ML image upscaling, face tracking, video background blur, RAW image processing, colour grading, path tracing, and fluid simulation. For Nvidia users, the headline addition is full CUDA support alongside existing OpenCL, Vulkan, and Metal APIs. Primate Labs’ early data underlines the significance of this move: a GeForce RTX 4070 averages roughly 236,094 points under CUDA, compared to 163,749 with OpenCL and 170,245 with Vulkan. An RTX 4090 shows an even wider spread, scoring approximately 444,438 points via CUDA versus 252,318 on OpenCL.
Dataset sizes across workloads including File Compression and PDF Viewer have also been increased to better reflect contemporary usage patterns. Geekbench 7 is available now for Android, iOS, Windows, macOS, and Linux, remaining free for personal use. Through August 6, Geekbench 7 Pro is offered at a 20 percent discount as part of the launch promotion.