Geekbench 7 launches with CUDA support and a major multi-core overhaul

Primate Labs has officially released Geekbench 7, the latest major version of its popular cross-platform benchmarking software. The update introduces a reworked CPU benchmark, new media and gaming workloads, a revamped GPU benchmark with CUDA support, and some fundamental changes to how Geekbench scores should be interpreted.

One of the biggest changes concerns multi-core performance. Geekbench 7 only runs a workload in multi-threaded mode if the application it represents actually behaves that way. The HTML5 Browser workload, for example, is excluded from the multi-threaded suite because browsers tend to be single-threaded or lightly threaded. Primate Labs says the change should produce multi-core scores that better represent real-world performance instead of rewarding processors for throwing more cores and threads at every workload.

Geekbench 7 also establishes a new performance baseline. The AMD Ryzen 7 7700 serves as the reference processor and is normalised to 2,500 points, replacing the Intel Core i7-12700 used for Geekbench 6's 2,500-point baseline. This means Geekbench 6 and Geekbench 7 results shouldn't be directly compared. An apparent increase or decrease between versions doesn't necessarily correspond to an equivalent change in actual hardware performance.

Preliminary Geekbench 7 results supplied by Primate Labs give us an early look at the new 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 manages around 2,734 and 24,439 points, respectively, while Apple's 14-core M4 Pro lands at approximately 3,360 and 23,188 points.

Qualcomm's 18-core Snapdragon X2 Elite isn't far behind, averaging around 3,225 points in single-core and 22,735 points in multi-core. Apple's M5, meanwhile, posts an impressive single-core result of approximately 3,753 points and a multi-core score of 18,722. Geekbench 7 also adds AV1 encoding for screen sharing, Opus audio compression and live caption generation using OpenAI's Whisper speech recognition model. There's also a new Game Physics test based on the Jolt Physics engine, alongside expanded Photo Editor and Photo Library workloads.

The GPU benchmark gets an equally substantial overhaul, with new tests covering ML image upscaling, face tracking, video background blurring, RAW image processing, colour grading, path tracing and fluid simulation. More importantly for Nvidia users, CUDA joins OpenCL, Vulkan and Metal as a supported compute API. Primate Labs' preliminary numbers illustrate why that addition matters. A GeForce RTX 4070 averages around 236,094 points using CUDA, versus 163,749 with OpenCL and 170,245 with Vulkan. An RTX 4090 produces an even larger variance, scoring approximately 444,438 points under CUDA versus 252,318 with OpenCL. 

Lastly, Primate Labs has increased dataset sizes across workloads such as File Compression and PDF Viewer to better reflect modern usage. Geekbench 7 is available now for Android, iOS, Windows, macOS and Linux and remains free for personal use. Geekbench 7 Pro is discounted by 20% through August 6 as part of the launch promotion.