As more Apple's Mac Studio desktops end up in the hands of curious end-users, more details about their performance and features emerge. This time around someone has run PassMark Software's CPU benchmarks on a Mac Studio equipped with Apple's 20-core M1 Ultra processor. The chip demonstrated very good single-thread performance, its general CPU performance was ahead of most desktop CPUs, yet it could not compete against high-end workstation-grade processors.
Single-Thread: On-Par with Alder Lake
Apple's CPU developers have always strived to maximize single-thread performance of their designs as they rightly believed that this leads to lower power consumption of their smartphones and PCs. As a result, Apple's M1 is known for particularly strong single-thread performance and it is not really surprising that an M1 Ultra processor clocked at around 3.20 GHz is on par with Intel's Core i7-12700 that boosts its fastest core all the way to 4.90 GHz in PassMark's single-thread CPU performance benchmark (3,896 vs 3,918 points).
Like other M1s, Apple's M1 Ultra does not increase its clock speed significantly higher than 3.20 GHz. The M1 Ultra's massive system level caches (SLC) largely provides the maximum bandwidth required for single-thread workloads, even M1 Ultra's monstrous memory subsystem (800GB/s aggregated bandwidth) but it doesn't bring much difference with single-thread workloads when compared to M1 Max, M1 Pro or even M1. All of these system-on-chips are great when it comes to single-thread performance.
General CPU Workloads: 2.6 Slower than Threadripper Pro
Packing 16 high-performance Firestorm and four energy-efficient Icestorm cores, Apple's M1 Ultra promises some serious performance in multi-threaded workloads as well. Indeed, it scores 41,306 points in PassMark's CPU Mark that synthetically measures virtually all CPU capabilities, including integer, floating-point math, and things like compression, encryption, and physics simulation.
While 41,306 points in CPU Mark is a very good result that is even a bit higher when compared to Intel's Cores i9-12900KF (40,895) that can process more threads at once and has higher clocks in boost mode (albeit the same 3.20 GHz base), it is not high enough to be comparable to CPUs used in high-end workstations and servers.
AMD's 16-core Ryzen 9 5950X scores 46,212 points in the said test, whereas 64-core monster Ryzen Threadripper Pro 5995WX hits 108,882, beating Apple's top-of-the-range workstation CPU by 2.6 times. AMD's 64-core chip even with a base clock of 2.70 GHz simply has more computational resources than Apple's M1 Ultra, so in workloads that demand every single arithmetic logic unit or floating-point unit, Ryzen Threadripper Pro will always be faster.
Promises Vs. Facts
When Apple announces its latest computer hardware, it usually demonstrates benchmark results that present performance of its PCs in the best light possible. This tactic is hardly surprising, it is widely used in the industry. But in reality, there are other performance benchmarks and real-world use cases that at times contradict promises of unbeatable performance promised in ads and during presentation.
In general, if you need to choose between an M1 Ultra-based desktop or an AMD / Intel-based machine, you should probably check performance of such systems in your workloads rather than rely on benchmark results published over the Internet.