PC Users in Denial About Apple Silicon Performance
Leading up to Apple’s first ARM computer release and even today, a lot of people, especially PC users have been skeptical.
For a while I have been reading comments and discussions going between Mac and PC fans on Apple’s transition from x86 to ARM based CPUs (Apple Silicon). There has been a loud and persistent insistence that an ARM based chip cannot possibly compete against AMD and Intel workhorses.
People who have spent a small fortune building a large gaming rig with a dozen fans, liquid cooling and overclocking to the max, don’t want to hear that some CPU used in slim boutique cell phones can somehow compete with their monsters.
But the “unfortunate” reality for them, is that Apple has rapidly gained on Intel and now surpassed it. This isn’t just Apple bragging and exaggerating. Today we have countless benchmarks by independent testers verifying the same overall picture. AnandTech is one example, Arstechnica another. Their Cinbench 5 tests against the other CPUs used for laptops shows the M1 beating the competition both in singlethreaded and multithreaded performance tests.
The highest performing CPU today that money can buy is the AMD Ryzen Threadripper 3990X Processor. However this is primarily due to an insane number of cores (64). The AMD Ryzen 9 5950X CPU is regarded as having the worlds fastest CPU cores. It has 16 CPU cores with max 32 hardware threads. You also get variants with fewer cores. But common to all of these is that the thermal design power (TDP), which is the max heath generated running real applications is 105W. That is what the cooling system have to be able to deal with. The Apple Silicon M1, has a TDP of a mere 10W, which is 10x lower.
Despite this Arstechnica Cinbench R23 tests show that the M1 gets dangerously close to the 5950X in single thread performance.
And when the Ryzen and M1 are both running 4 cores they get 6092 and 5622 score respectively. That is impressive for a chip designed for the low end of Apple’s computers. Keep in mind that the AMD Ryzen 9 5000 series is made for high performance desktops with massive cooling systems. Of course these run a lot more cores. But e.g. the Ryzen 5 5600X has 6 cores and has TDP of 65W. Yet in this Cinbench R23 performance comparison done by CPU Monkey there M1 gets 7,508 while the 5600X gets 11,268. That is of course higher but performance per core is pretty much identical:
AMD: 11268 / 6 = 1878.0
Apple: 7508 / 4 = 1877.0
So the AMD has 50% higher performance by having 50% more cores. Yet the TDP 650% higher. And let us not forget that the 10W on the M1 is not just for the CPU. The 5600X is just a CPU. The M1 in contrast contains:
- 16-core Neural Engine. Which make machine learning tasks such as image and text recognition, various video and photo editing tasks up to 15x faster by Apple’s claims. Previous claims about Neural Engine by Apple has been independently verified.
- 8-core GPU at 2.6TFLOPS which makes it faster than any other integrated GPU by a large margin.
- Fabric, which we don’t know what is for yet.
- Secure Enclave and specialized AES encryption hardware. This allows encryption to be done without wasting CPU cycles.
In short these 10W delivers a lot more functionality than the 65W AMD Ryzen delivers. This also underscores how Apple’s tactic differs from the competition. Apple prefers to offload a lot of specialized tasks with high performance demands to custom silicon. That is easy for them to do because they have full vertical integration. E.g. you don’t use the Neural Engine directly, but through Apple frameworks such as Core ML.
What About the Real World?
A frequent argument by PC diehards in response to impressive benchmarks was that “well… these are just synthetic benchmarks and not real applications.” That excuse does not carry much weight anymore either. We have started seeing a string of tests of real world applications and they are equally impressive.
I don’t know how many times in the past I have read claims such as:
ARM based Macs will be fine for light workloads such as browsing, writing email and watching Netflix.
— PC die-hards in denial
But the performance results are ticking in from all corners of the world. They tell a story which makes it ever harder to persist in these sorts of claims with a straight face.
The last desperate attempt at saving face has been to insist that the M1 is useless anyway because most apps will have to run through Rosetta 2 and that will give terrible performance. One PC die-hard told me you could easily expect a 5x performance drop from this. Yet the benchmarks we now got suggests this is a desperate dystopian fantasy among Apple haters:
In fact the M1 is so fast that the Rosetta 2 performance hit is small enough that many users of upgrading from Intel Macs will still experience increased performance despite running apps which have not yet been ported.
The PowerPC to Intel Transition Experience
I went through the PowerPC to Intel transition and I got to say that went surprisingly smooth. But it is important to keep in mind what your expectations are. As a software developer I know how hard this kind of things is. If you expect absolutely everything to work flawlessly from day one, then your experience will be different. I have seen people describe the PowerPC to Intel transition as hell, but that tends to be people with a rather naive view of what a transition entails. Day one can be rough if you use a wide variety of non-Apple software, but the point is that things improved rapidly.
A lot has happened since those days. Software for the Mac is to a larger extent developed in Xcode, we got more software distributed as LLVM bitcode. Software simply is not as tied to an architecture as strongly as when the PowerPC to Intel transition happened.
Intel to Apple Silicon Transition
It would be very surprising if the Intel to ARM transition was less smooth than PowerPC to Intel. After all Apple has years of experience with ARM through iPad and iPhone which runs very similar software. iOS is really just a flavor of macOS. The libraries, tools, programming languages and everything else is very similar. Going from developing Mac software to iOS software was not a big transition. At isapplesiliconready.com we can keep up to date on what software has currently been ported to Apple Silicon.
Many will be happy to know that important applications for professionals such as Adobe Photoshop and MS Office has already been ported. Other professional tools such as Affinity Designer, Photo and Publisher have also been ported.
Countless software companies already have both an iOS and a macOS version of their software. They naturally use a lot of code sharing. With SwiftUI they are getting close to merging the iOS and macOS world.
It will be interesting to watch in the next weeks or months as benchmarks for the new Macs pop up. There will be a lot of surprised people. But don’t worry people will still dismiss Apple. They will complain about 16 GB RAM being too little for professionals and that the number for ports are too few (which annoys me too). Sure we would all like the option of more RAM and ports. But that does not change the exceptional advancement Apple has been making with Apple Silicon over the years.
How Long Can AMD Stay on Top?
Some people will cling to the fact that AMD still has a performance edge, but for how long? AMD had a much smaller gap to close to surpass Intel in performance. Apple had a much faster progression. The ARM world is simply moving a lot faster and x86 is living on borrowed time.
There is nothing inherently superior about the x86 architecture which suggests it should stay ahead performance-wise for all future. Quite the contrary ARM has an ISA which is better suited to make a high performance architecture. Eventually this will be a numbers game. Both Intel and AMD will be facing an army of ARM makers, they cannot beat all of them in the long run. And typically the winner will be whoever can throw most cash at the problem.
Apple has a clear edge here as they don’t need to spend money developing lots of architectures. They can focus on making chips for their limited selection of products. Apple has tons of money they can throw at a very limited set of of chip designs.
The other markets AMD and Intel compete against ARM on will be handled by other companies such as Amazon making the Graviton 2 ARM chips or Ampere’s 80-core ARM-based 64-bit server processor.
EDIT: If you are more interested in this, I have a new article (unpolished) out discussing technical details of why the M1 is so fast as well as comparison to AMD. I may revise my view here as M1 seems to be far ahead of AMD as well.
Will I be Getting an ARM Based Mac?
Despite my praise and temptation to get an ARM based Mac I may not actually get one for the simple reason that I am still really happy with my 5–6 year old Mac Pro.
My key issue with my Mac Pro is actually not performance, but rather the number of USB ports and non-standard SSD disk which is not that easy to replace, without a special kit.
I see this as the key problem with the new ARM based Macs: Too few ports. It is not clear to me whether the harddrive can be changed. But these kinds of things seems to be what we have simply have had to put up with when it comes to Apple. Personally I still dream of a return to the old beatiful G4 towers.
These computers really appealed to professionals and hobbyists alike, as they were really easy to open and add things. I would still have loved the ability to add extra hard drives, specialized cards, internal DVD burners etc. But sadly Apple has never showed much love towards this segment of users.
But I still love the Mac experience, macOS and Mac software too much to switch to another system. I will just have to suck it up and have cute little Mac box with a bunch of cables sticking out of it.