Summary created by Smart Answers AI
In summary:
- Macworld reports that Primate Labs has released Geekbench 7, three years after version 6, featuring redesigned multi-core tests and modern workloads like AV1 encoding and machine learning applications.
- The new benchmark better reflects real-world performance demands with GPU tests focused on content creation, face tracking, and image upscaling rather than traditional graphics rendering.
- Early M5 MacBook Air results show 3,608 single-core and 17,470 multi-core CPU scores, though Geekbench 7 scores aren’t comparable to previous versions.
It’s not easy to make a quick, reliable benchmark that works across architectures and operating systems. There’s a reason Geekbench has been sort of the gold standard for comparing overall CPU performance over the years, and to a lesser degree, GPU compute performance.
On Thursday, around three years after the release of Geekbench 6, Primate Labs is releasing Geekbench 7. To better represent the demands of modern applications, it includes new workload tests, a redesigned multi-core benchmark, and a new GPU benchmark. The benchmarks operate on larger data sets as well, again better representing the demands of modern applications.
New media workloads include encoding screen-sharing video with the AV1 codec, compressing music and speech with the Opus audio codec, and decoding audio and video while using the Whisper speech recognition model to generate live captions. Other new test include Game Physics, an updated Photo editing test, and new Photo library tests that processes JPEG XL and DNG formats.
The multi-core tests are especially changed, redesigned to “better reflect how real applications behave,” according to Primate Labs. Instead of forcing a series of exclusively multi-threaded tasks, which produces a distorted view of performance, the Geekbench 7 tests only include tests in the multi-threaded benchmark if it’s highly multi-threaded in real life.
The example Primate Labs gives is the HTML5 Browser test–web browsers and the tasks they run are typically highly dependent on single-thread performance, so it’s not included in the multi-threaded tests.
The GPU benchmarks are still focused on the compute applications for GPUs rather than on real-time 3D graphics rendering, but the types of applications have evolved. The tests now focus on the content creation and machine learning uses of GPUs, since those dominate GPU compute applications. For example: tracking faces in video, upscaling images using machine learning algorithms, or blurring the background of a video chat.
Here at Macworld, we’ll be using Geekbench 7 to benchmark new Apple hardware moving forward, while still including Geekbench 6 for a while to provide comparisons against older hardware.
Scores are not comparable between versions. We quickly ran Geekbench 7 on an M5 MacBook Air, and got a single-core CPU score of 3,608 and a multi-core score of 17,470. That compares to 4,164 and 16,995 on Geekbench 6.6. The GPU benchmark, using the Metal API, scored 72,514 on Geekbench 7 and 76,377 on Geekbench 6.6.
You can download Geekbench 7 for Mac from Primate Labs’ website, and Geekbench 7 for iPhone and iPad from the App Store. At the time of publishing, the App Store version does not yet appear to be available, but should be coming soon.
