Friday, Oct 29, 2021, 14:06 Mac: Hardware

MacBook Pro 2021: Testing The M1 Max Performance In Real-World Benchmarks

The MacBook Pro with M1 Pro or even the M1 Max has impressed the entire tech world – similarly to how the release of Apple's first in-house Apple Silicon "M1 chips" one year ago managed to. Although the M1 Max can even leave many desktop PCs in the dust when it comes to performance, it's still an incredibly efficient chip. Graphics cards with similar performance require almost twice as much energy for the same results achievable with an Apple "system on a chip".

Now that we've finally gotten our hands on a MacBook Pro with M1 Max (the 32 GB of Unified Memory, 32-core GPU configuration), we're going to run some of our own benchmark tests using our very own Logoist vector graphics software designed exclusively for the Mac. There are already some benchmark results available from platforms using other tools such as Geekbench. However, we've decided to run our performance tests using our real-world Logoist benchmark platform.

Benchmark Testing Method
The parent company of, Synium Software has developed the vector graphics software by the name of Logoist. Since Logoist 4, the app runs natively on Apple Silicon and we completely switched to Apple's "Metal" graphics framework for GPU-based rendering. Logoist is well-suited for benchmarks, as both the CPU and GPU simultaneously bear load during different tasks and rendering entire vector graphics requires multiple processor cores.



Test 1 – Rendering Thumbnails
We kept the CPU occupied with triangulation tasks for our first test and had it complete several geometric calculations. We also placed the GPU under load by drawing out the shapes and applying shader code simultaneously. Our testing methodology is pretty simple: In a slightly adapted version of Logoist, we measure how long it takes for the Mac to render all 150 provided example documents ten times in a row. The way Logoist accomplishes its tasks efficiently reflects how quickly the tested Mac would handle most graphically intensive applications in non-benchmark scenarios.

Test 2 – Image Enhancement via CoreML
Logoist comes with a CoreML model used to increase the resolution of images, a process known as "Super Resolution". When possible, the CoreML model is executed on the GPU - however, in the case of Intel's GMA, this process takes place on the CPU as the GMA doesn't possess the required abilities. Apple's M1 has a clear advantage here thanks to the Neural Engine. In this test, a Full HD image is increased by a factor of two in resolution using a "Super Resolution" machine learning model.

Test 3 – Metal Compute, Metal Shaders
Modern graphics cards can now do much more than paint a simple triangle. GPUs may now be used to perform complex calculations. In this case, tasks divided into hundreds to thousands of tiny, isolated "sub-problems" are well-suited for benchmarking. In this test, we ran a Gaussian blur via Metal performance shaders, a custom HDR glow filter, and a Sobel edge detection filter 100x on the GPU using Metal Compute on a 4K image.

Test 4 – Drawing With Metal
To compare the performance while drawing triangles, we let Logoist draw 50,000 circles 100x using Metal. The circles are triangulated (meaning: converted to triangles), with about 100 individual triangles per circle. We also measured how long the Mac took to finish this benchmark cycle. The CPU is hardly used in this test, all data for drawing the circles is present in a memory area accessible for the GPU. As far as this task is concerned, Apple chips have a massive advantage; since many of the circles to be drawn overlap, the M1 is able to clip and discard some of the triangles because of its "tile-based rendering"
() – and didn't need to draw them.

Conclusion And Results
The M1 Max of the MacBook Pro 16" performs exceptionally well. It is at the top in all tests, which was to be expected anyway. However, it is simply astonishing how much the chip outperforms the MacBook Pro 2019 with Core i9. Once again, it can be said that Apple rather understated the performance at the „Unleashed" event. Since the benchmark tests were completed so quickly, we could not even notice any heat or fan noise.

The iMac Pro could only keep up in one aspect, namely the shader performance. However, the iMac Pro was tremendously inferior in "Metal Draw", because the architecture advantages of the M1 Max fully come into play here (tile-based rendering).

More articles you might enjoy to read: