On Tuesday, Primate Labs revealed Geekbench 6, the latest version of its popular benchmark tool. Used to gauge the CPU and GPU performance of Macs, iPhones, iPads, and non-Apple computing devices, Geekbench 6’s suite of tests has updates that better reflect the workloads that today’s devices can handle.
“A lot has changed in the tech world in the past three years. Smartphone cameras take bigger and better pictures. Artificial intelligence, especially machine learning, has become ubiquitous in general and mobile applications,” wrote John Poole in a blog post. “The number of cores in computers and mobile devices continues to rise. And how we interact with our computers and mobile devices has changed dramatically – who would have guessed that video conferencing would suddenly surge in 2020?”
To better reflect modern workflows, Geekbench 6 implements new tasks, such as background blur in videoconferencing, filtering and adjusting images for social media, automatically removing unwanted objects from photos, detecting and tagging objects in photos using machine learning models, and using scripting languages to analyze, process, and convert text. Data sets have also been updated with larger files and high-resolution images.
Geekbench explains that the new CPU benchmark scores “are used to evaluate and optimize CPU and memory performance using workloads that include data compression, image processing, machine learning, and compiling code.”
The new tests are similar in nature to Geekbench 5, but perform newer tasks with new data sets and libraries to better represent modern applications. You cannot directly compare Geekbench 5 and Geekbench 6 scores. Multi-core performance, in particular, uses new methods of sharing work between cores that should better represent the performance of CPUs that have “high performance” and “high efficiency” cores. In an interview with Ars Technica, Poole said, “We really want this to be a tool that people can use to figure out questions they have about performance.”
Geekbench 6 features new testing criteria to better measure real-world performance of your Apple devices.
Primate Labs
Macworld uses Geekbench in our iPhone, iPad, and Mac review, and we’ll be using Geekbench 6. However, previous results recorded with Geekbench 5 can’t be compared to Geekbench 6, since the two use different test suites. In the short term, we may include Geekbench 5 results to provide a historical context for older devices we are unable to test with Geekbench 6. it’s hard to compare the scores without doing extensive testing, but a test of the 14-inch MacBook Pro with an 8-core CPU and 14-core GPU returned scores of 2138 (single-core) and 8475 (multi-core) versus an average of 1750 (single) and 9550 (multi) using Geekbench 5. We also tested the 14-inch MacBook Pro with a 10-core M1 Pro and 16-core GPU and got Geekbench 6 scores of 2399 (single) and 12289 (multi), versus Geekench 5 scores of 1778 (single) and 12544 (multi).
Geekbench 6 is free for personal use and can be downloaded from the Geekbench site (Mac) or the App Store (iOS). The Pro version allows users to run the app from a portable storage device or network drive, save results offline, automate testing, and provides a license for commercial use. Geekbench 6 Pro is $99, but Primate Labs offers 20 percent discount ($79) until February 28.
What do the Geekbench scores mean?
What do these Geekbench scores mean? What is it measuring and why should you care?
Geekbench runs through a set of specific repeatable tests that stress either your CPU or your graphics processor (GPU) in specific ways and turns that into a numeric score. The tests and the data they use are the same on every platform, but some platforms may have hardware that can speed up certain parts of it (like file compression). A higher is better, but it doesn’t represent a concrete value like “time elapsed” or “pixels per second.” It’s an entirely synthetic benchmark and score.
CPU tests
The CPU test includes things like file compression (compress and decompress a 75 MB archive with 9,841 files in a variety of methods), generating directions between a sequence of locations, rendering 8 popular web pages, rendering PDFs, running developer scripts, and machine-learning tasks like object detection and background blur.
The CPU test has two result scores: Single-core, which measures the performance of running the tasks on a single CPU core, and multi-core, which uses all CPU cores. Both are important–many applications have their performance limited by a single main thread, so single-core CPU performance will determine how fast they run. But for applications that are well multi-threaded or for running multiple applications at once, multi-core is a good measure of the total maximum CPU performance.
GPU Compute tests
You might think that measuring graphics processor (GPU) performance is akin to knowing what sort of frame rate you can expect in 3D games. When it comes to Geekbench, that is not necessarily the case. The Geekbench tests do not measure 3D graphics rendering performance. Instead, it measures the performance of “GPU Compute” tasks–using the GPU to handle computational tasks like edge detection on images, applying a gaussian blur to a 24-megapixel photo, a particle physics simulation, or detecting faces in photos.
It’s quite possible for one product to run 3D games faster than the other while scoring lower on the Geekbench GPU Compute tests. While GPU Compute performance and 3D graphics performance are often related in some ways, it’s fair to say that if you want to measure 3D graphics performance you should run a 3D graphics benchmark, not a GPU Compute benchmark like Geekbench.
When you run the Geekbench 6 GPU Compute test, you get to choose which API (application programming interface–the way developers talk to the hardware) that you want it to use. On a Mac, you can choose between OpenCL and Metal. On iPhones and iPads, the only supported API is Metal. Your choice does not change the tests at all, only the way that the application talks to the hardware. On the Mac, Metal is probably the most important score and should probably be compared to Vulkan performance on Windows PCs and Android.
If you’re technically inclined, you can read PDF white papers that describe the tests in detail. There’s a white paper for the CPU tests and the GPU Compute tests.
