21.1 C
New York
Wednesday, May 22, 2024

Buy now

Apple Silicon Face-Off: M1 Ultra and M1 Max take on high-end PC juggernauts

Its high time Digital Foundry took a look at Apple Silicon and today were going to be looking at the higher-end chips in the line-up. Our focus is on the monster that is the M1 Ultra, found within the latest Apple Mac Studio, but were also going to be checking out the MacBook Pros M1 Max. Its the M1 Ultra that truly commands our attention though: this system-on-chip represents the highest end computer processor Apple has designed to date, with the firm claiming it should be as fast as a high-end Windows desktop. Packing 20 CPU cores, a 21 teraflop GPU, and 800GB/s of memory bandwidth, it certainly seems like it could be – but how does it measure up in real-world testing and how well does it game?

The M1 and M2 lines are the culmination of a long journey that has seen the firm transition away from SoCs based on third party design and IP, moving all chip design in-house to the furthest extent realistically possible. Apple designs its own GPUs, its own CPUs, and handles SoC design and integration. This results in tremendous control over processor design – the kind of control you would need to scale a phone processor up for high-end desktops.

Which brings us to the M1 Ultra. Since 2020, Apple has been moving its Mac desktops and notebooks away from Intel CPUs and AMD GPUs and over to its in-house SOCs, taking the same fundamental tech from iPhones and integrating it into computers. Apple started with lower-end and lower-power form factors, but finally came around to high-end desktops with the release of the Ultra a few months ago. The M1 Ultra isnt really its own unique chip, however. Its actually two M1 Max SoCs connected over a high-bandwidth 2.5TB/s interposer. To the operating system and the user it seems like one monolithic chip with 1 CPU and GPU, but in reality this is two chips linked through a first-of-its-kind interconnect with the performance to support a dual-chip GPU and CPU.

Watch on YouTube

Digital Foundrys video analysis of the M1 Ultra and M1 Max processors, stacked up against powerful PC equivalents including the Core i9 12900K and RTX 3090.

The Ultra packs a whopping 20 CPU cores, split between 16 performance cores and 4 efficiency cores in a configuration similar to modern Intel designs. While the clockspeeds may be lower than desktop PC CPUs, instructions-per-clock are higher on the performance cores, leading to similar overall performance-per-core. In its highest-end spec, the 21 teraflop GPU features 64 of Apples in-house graphics cores, with performance similar to an RTX 3090 according to Apple, though well touch on this later. To round things out, the system packs a stunning 800GB/s of memory bandwidth to keep those GPU and CPU cores well-fed.

M1 Ultra is only currently available in the Mac Studio desktop computer, which we tested in its maxed-out configuration, with 128GB of memory and an 8TB SSD. Most interestingly, this computer has a volume of just 3.7 litres, which is truly tiny and only slightly larger than an Xbox Series S. It uses two blower-style fans that pull air through a large copper heatsink to dissipate the roughly 200W that the system pulls at load, which is a small fraction of the energy used by a high-end desktop PC.

So lets move on and actually measure how fast this machine is. Were going to start off with gaming tests before closing with productivity benchmarks and synthetics. Is this machine truly as fast as a high end desktop PC – or possibly even faster? Lets take a look at our gaming benchmarks, calculated via video capture as is the Digital Foundry way. While internal benchmarks are largely accurate these days, our philosophy is that the only frames that matter are the frames that actually make it to the video output of the hardware.

M1 Max (MBP 2021) M1 Ultra (Mac Studio 2022) RTX 3080M 150W (MSI GP66 Laptop) RTX 3090 (Desktop PC)
Shadow of the Tomb Raider 31.0 49.0 29.0 65.0
Metro Exodus 27.9 34.8 30.3 71.6
Total Warhammer 3 14.9 25.4 25.3 47.4
World of Warcraft 18.4 36.2 32.9 81.6
Wildlife Extreme Bench 20215 35498 24247 42451

Its not a particularly large table because, unfortunately, there arent many high-end Mac games that we can actually test, particularly when it comes to big-budget games. But we do have a few titles here – and the results are intriguing. For our gaming tests, weve got a 16 inch Macbook Pro with the fully-enabled M1 Max chip, our maxed-out Mac Studio, an MSI GP66 gaming laptop, with an eleventh-gen i9 and a 150W RTX 3080 mobile processor, and a high-end desktop PC with a Core i9 12900K paired with the mighty RTX 3090.

Looking at Shadow of the Tomb Raider. This isnt a native Apple Silicon game, as the title was written for x86, so the M1 chips here have to use the Rosetta 2 translation layer to function – but it doesnt really seem like that has much of an impact on performance. The benchmark sequence running at max settings at 4K shows the 3080M and M1 Max are neck-and-neck, while the Ultra falls squarely between the M1 Max and the 3090. The Ultra has solid performance and reasonable scaling from the Max, but isnt quite holding the line against ultra high-end GPUs.

Metro Exodus – the original non-RT version – has a decent Mac port, although again it was written for x86. The Ultra splits the PCs here as well, while the Max does a good job of fending off the 3080M. On the flip side, there seem to be very serious problems with frame-times and stuttering when vsync is disabled for on Macs for some reason, which I noticed across these tests. Total War: Warhammer 3 is another x86 game, but it doesnt seem to hold up quite as well as Metro or Tomb Raider. M1 Ultra is far behind the 3090 here and barely keeps pace with a high-end gaming laptop. Perhaps this can be chalked up to a sub-optimal port, or problems with the Rosetta translation.

Apple Silicon games and benchmarks are hard to find, but 3DMarks Wildlife Extreme is indeed a native application.

But what about native Apple Silicon games? There are remarkably few games for Apple Silicon, and most of them are iOS ports, not conventional PC software. There is one prominent game that we can test across platforms though – World of Warcraft. This is a full-bore Apple Silicon version of Blizzards long-running MMO, but despite running natively, the same pattern emerges with the M1 Ultra yet again falling squarely between the two PC systems, falling well short of the 3090 but still delivering performance in line with a high-end PC GPU. The Max is borderline unplayable while the 3080M hovers around 30fps. All of these systems would be perfectly fine with the game at remotely reasonable settings, of course – we are running the game essentially maxed out at a whopping 8K internal resolution to create a proper stress test.

Theres one cross-platform game graphics benchmark that runs natively on Apple Silicon as well – 3DMark Wildlife Extreme, which renders a set of relatively simple 3D scenes at 4K. Here, the Ultra falls somewhat short of the 3090 but comes in a solid 76 percent faster than the Max. Ultimately, the Ultra seems to sit somewhere below the 3090 in graphics performance, at least as far as we can tell from benchmarking across operating systems. Its still a powerful processor though and seems to slot in at roughly the 3070 or 3080 level depending on workload.

Scaling from the M1 Max is reasonable, but not perfect. Typically, you should expect a 60-70 percent performance improvement over the single-chip option. Perhaps the interposer is causing some minor hiccups here, as using multiple chips for one GPU requires a massive amount of bandwidth.

These results are really just for evaluating raw performance though, as the Mac is not a good gaming platform. Very few games actually end up on Mac and the ports are often low quality. If there is a future for Mac gaming it will probably be defined by “borrowing” games from other platforms, either through wrappers like Wine or through running iOS titles natively, which M1-based Macs are capable of. In the past, Macs could run games by installing Windows through Apples Bootcamp solution, but M1-based chips cant boot natively into any flavour of Windows, not even Windows for ARM.

Blender (CPU Samples Per Min) M1 Max (MBP 2021) M1 Ultra (Mac Studio 2022) Core i9 10850K (Desktop PC) Core i9 12900K (Desktop PC)
Monster 99.4 195.9 88.7 178.1
Junkshop 53.8 107.33 50.7 101.1
Classroom 43.3 84.4 37.8 82.3
Geekbench CPU M1 Max (MBP 2021) M1 Ultra (Mac Studio 2022) Core i9 10850K (Desktop PC) Core i9 12900K (Desktop PC)
Multi-Core 12577 23580 9599 17446
Single-Core 1774 1784 1285 1820
Cinebench CPU M1 Max (MBP 2021) M1 Ultra (Mac Studio 2022) Core i9 10850K (Desktop PC) Core i9 12900K (Desktop PC)
Multi-Core 12259 23908 11171 25160
Single-Core 1528 1531 1095 1858
Handbrake 4K60 Encode M1 Max (MBP 2021) M1 Ultra (Mac Studio 2022) Core i9 10850K (Desktop PC) Core i9 12900K (Desktop PC)
Time (mins:secs) 7:10 4:08 5:43 2:44

As youve likely realised from the table above, I also spent some time benchmarking the CPU in the M1 Ultra. I tested Blender, Geekbench, Cinebench and Handbrake – and the Ultras results are compelling. Weve swapped the GP66 for my desktop computer here, which packs a Core i9 10850K. Think of this as Core i9 10900K with a barely perceptible clock-speed reduction. Across these tests, the 12900K and M1 Ultra prove very comparable. The two chips are essentially a match with respect to multicore performance, though the ultra-high frequencies the 12900K is capable of can give it the edge in some single-threaded tests. The 10850K and M1 Max are closely matched as well.

The scaling from M1 Max to M1 Ultra is close-to-linear across these runs, unlike our graphics benchmarks. On average, M1 Ultra is 88 percent faster, with some results approaching 100%. Linking up two clusters of cores across an inter-chip medium is something weve seen in the PC space for years now and very good scaling is to be expected here.

Finally, I thought Id throw in some real-world benchmarks from a couple of programs I frequently use – Final Cut Pro and Topaz Video Enhance AI. Were looking at the two M1 computers here, as well as a 16 inch 2019 MacBook Pro with an eight-core Intel CPU and an AMD RDNA 1-based GPU. The results are very curious in Final Cut. While both M1 machines trounce the Intel-based MacBook, export times are virtually identical across the M1s. So whats going on?

With typical Final Cut workloads on M1 chips, export performance seems to be dictated by the hardware video encoders. The M1 Ultra has the same video hardware encoders as the Max, so theres no meaningful performance difference when encoding a ProRes or h.264 video without many effects. To actually see a difference in export times, youd need to really stress the GPU with lots of effects and Motion templates. Even then it would be hard to see a large difference. That isnt to say that there arent big moment-to-moment performance differences, though – Final Cut generates video thumbnails in real-time on the CPU cores, which occurs nearly instantly on an M1 Ultra and is significantly slower on M1 Max. In general, the timeline is more responsive and the editing process is more fluid – but that wont be reflected in simple export tests.

M1 Max (MBP 2021) M1 Ultra (Mac Studio 2022) Core i9 9980HK, Radeon Pro 5500M (MBP 2019)
Final Cut h.264 Export (mins:secs) 1:05 1:03 1:33
Final Cut ProRes Export (mins:secs) 0:23 0:23 1:25
Topaz Video Enhance (1080p to 4K Upscale, Artemis High) 6:01 4:12 13:33

Topaz AI is much more straightforward. Were strictly GPU-bound here and the M1 Ultra shows a solid performance improvement – completing the test 43% faster – though not particularly impressive given the doubling of GPU hardware. Both machines crush the 2019 MacBook Pro, as expected.

So, the M1 Ultra packs similar performance to the highest-end PC chips, trading blows across a variety of metrics. CPU performance is up there with the best Intel has to offer, while the GPU sits one or two rungs beneath the PC performance leaders at the moment. The key metric with M1 Ultra isnt raw performance, however, though it is largely competitive with PCs on that front. Its power consumption. The Ultra manages to pull even with fast consumer desktops while consuming one quarter to one third of the power consumption. The Mac Studio itself only pulls about 200W when fully loaded, and usually draws much less.

So, why is the M1 Ultra so much more efficient than comparable PC designs? Firstly, Apple has a considerable process node advantage over its competitors. By leveraging TSMCs 5nm process, Apple is one or two silicon fabrication nodes ahead of its nearest rivals at the moment, which means higher density and lower power consumption for Apples chips. Apple generally gets access to TSMCs newest processes before its PC competitors and has been producing chips at 5nm for over two years at this point.

To see this content please enable targeting cookies.

Secondly, Apple is simply throwing way more silicon at the problem. The M1 Ultra uses a whopping 114 billion transistors across two chips; in contrast, the GA102 GPU in the RTX 3090 packs just 28 billion transistors. With so much more logic, Apple can run its chips at lower clocks and lower voltages and still achieve similar performance. The extremely high density of TSMC 5nm helps a lot here. Lastly, Apples CPU and GPU architectures play a significant role here. These are designs that are primarily designed for iPhones and other low-power applications. There are likely many mechanisms inside the chip to keep energy consumption in check, including very effective power gating.

Given the immense potential of the Apple solution, theres one final question thats worth addressing: would a move to ARM be practical for the broader PC market as well? After all, Apple achieved an enormous performance improvement when they moved to ARM, so could this be a good solution for PC vendors too?

Generally the answer is no, at least not at the moment. There are two major problems here. The things that make Apples designs effective arent specific to the ARM instruction set license they use. These are mostly factors weve discussed already – its unique high-performance architectures and process node advantage being the most important. Critically, no-one else is currently offering an ARM CPU core design capable of going toe-to-toe with AMD and Intel. The second problem is the lack of an effective translation layer for x86 code. MacOS has Rosetta 2, which is a relatively efficient and broadly compatible solution for running x86 code seamlessly on ARM-based Macs. Windows 11 for ARM has a software emulator for x86 programs, but performance is degraded and compatibility is lacking.

Apples die-shots of its M1 silicon line-up may or may not be accurate, but essentially, the GPU is M1 Max is twice the size of M1 Pro, which is in turn twice the size of M1s. The M1 Ultra effective stacks up two M1 Max chips, meaning a potential doubling in both CPU and GPU resources.

The M1 Ultra is an extremely impressive processor. It delivers CPU and GPU performance in line with high-end PCs, packs a first-of-its-kind silicon interposer, consumes very little power, and fits into a truly tiny chassis. Theres simply nothing else like it. For users already in the Mac ecosystem, this is a great buy if you have demanding workflows. While the Mac Studio is expensive, it is less costly than Apples old Pro-branded desktops – the Mac Pro and iMac Pro – which packed expensive Xeon processors and ECC RAM. Final Cut, Photoshop, Apple Motion, Handbrake – pretty much everything I use on a daily basis runs very nicely on this machine.

For PC users, however, I dont think this particular Apple system should be particularly tempting. While CPU performance is in line with the best from Intel and AMD, GPU performance is somewhat less compelling. Plus, new CPUs and GPUs are incoming in the next few months that should cement the performance advantage of top-end PC systems. That said, the M1 Ultra is a one-of-a-kind solution. You wont find this kind of raw performance in a computer this small anywhere else.

Gaming on Mac has historically been quite problematic and that remains the case right now – native ports are thin on the ground and when older titles such as No Mans Sky and Resident Evil Village are mooted for conversion, its much more of a big deal than it really should be. Perhaps its the expense of Apple hardware, perhaps its the size of the addressable audience or maybe gaming isnt a primary use-case for these machines, but theres still the sense that outside of the mobile space (where it is dominant), gaming isnt where it should be – Steam Deck has shown that compatibility layers can work and ultimately, perhaps thats the route forward. Still, M1 Max and especially M1 Ultra are certainly very capable hardware and itll be fascinating to see how gaming evolves on the Apple platform going forward.

Related Articles

Latest Articles