We've seen major industry improvements this year just due to fabs - nVidia Ampere, AMDs new Zen3 and RDNA, et. al. This article ignores all of that by just doing comparison to Intel which is clearly struggling to compete.
It also assumes ARM is the only way to go which is far from clear at this point.
Edit: I didn't know before, but AMD also already has 20% of the notebook market. https://seekingalpha.com/news/3615331-amd-to-take-20-share-o...
So that makes it even more weird to say "expand Apple’s share of the PC market — at Intel’s expense" and not mention AMD.
As for MKL, yes it is impressive for some applications like real-time audio processing, but AFAIK with TensorFlow MKL is only used if you don't use a GPU, so pretty much never.
The M1 also tops the latest Ryazan CPU which came out a week ago, especially considering TDP, so what would it add to the report?
Both, Apple and AMD use TSMC as manufacture and next gen Ryzen is planned to be a 5nm chip too. When you compare the performance gains of the past generations of Ryzen you would definitely bet that AMD could for sure have a very potent CPU next year.
The M1 is very incredible for sure, but AMD is very much on track with their plannings and delivered 3 years in row.
Only in single-core performance and performance/watt, and the advantage the M1 has in both of these metrics is pretty much exactly the difference expected between a 7nm and a 5nm process (they’re both being produced by the same fab, after all).
It’s likely we’ll see Apple and AMD compete on pretty much equal terms in the CPU market in the future.
I mean, the Ryzen besting the M1 in multi-core performance isn't really surprising when you consider that it's a 16-core/32-thread, 105W desktop CPU vs an 8-core/8-thread (four of which are low power cores), ~25W fanless laptop CPU (looking at the 5950X from the Anandtech review).
Maybe, maybe not. There's no reliable way to estimate exactly how much improvement AMD would get from moving their design to the same fabrication process that the M1 is using.
1: Anandtech attempted to estimate the CPU power draw and got to around 10W at single-threaded load and a peak of 27W under compute-intensive load: https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste... - that might mean you'd need "only" 10 M1s to reach the TDP of a current Threadripper
Well, neither did Apple - and yet all of the article is about their latest released M1 chip which hasn't even stood the test of time.
Meanwhile we know what Ryzens can do - and per-core performance per Watt is rather close still (IIRC 6W vs. 7-8 power draw at load?).
I see the M1 as strong in single core benchmarks, probably because it's an evolution of the iPad's processor, that doesn't do a lot of multitasking, and kinda meh in multicore.
I disconsider geekbench tests because of the disclaimer: "Geekbench 5 is a cross plattform benchmark that heavily uses the systems memory. A fast memory will push the result a lot. The single-core test only uses one CPU core, the amount of cores or hyperthreading ability doesn't count".
I remember that when the Meltdown and Spectre Linux benchmarks came in, my boss immediately ordered 50x AMD EPYC servers the same day. Later that week, we tried to buy more but they were already sold out at our usual supplier.
We effectively replaced most of our Xeons with EPYCs within the month, because the Intel performance became unbearable and the AMD CPUs were cheap enough that we didn't need approval.
That said, I still keep some Xeon servers around, because Intel's software and their compiler are so f-ing perfect that MKL+ICC is unavoidable for some parallelized C++ services.
In terms of processing power per dollar, I'm not sure that's true.
The latest release just gave Intel nowhere else to hide with their "it's best for gaming" benchmarks, which seemed somewhat artificial anyway, seeing as how they were generally using massively powerful GPUs at very low resolutions.
(edit - The Zen2 and even Zen+ lines gave people looking for lots of cores a better path than intel, even where single threaded performance was a few percent off. The 3900X launched around the same price as the 10900K with two more cores/4 more threads)
I see this a lot as if it's just fact or commonly accepted but this is a very specific thing. Certain benchmarks would show Intel holding onto a narrow lead in single core performance (IPC) which is only the top metric in a few scenarios where parallelism isn't of consequence.
And certainly Intel leads in market share by a significant margin, so in that respect AMD hasn't beat Intel yet.
But for a large number of use cases, AMD Zen-based x86 chips been a superior option to Intel at least since Zen 2 if not earlier for many uses (of course - heavy parallelism, improved efficiency, better value, easier to power and cool, etc.)
Sadly there is no news of competition. Nothing from Intel/AMD or even Qualcomm. Microsoft's ARM attempts to take on Chromebook seem so feeble now.
Remember that AMD is fabricating their CPUs at the same fab than Apple does - they're not crippled by obsolete manufacturing process like Intel.
This is why these articles are such puff pieces - they're focusing on Intel, which has fallen behind by quite a lot and completely ignoring the major changes that also happened on x86-64 front.
In Cinebench ST, it's 3.4W per core only...
With CPU _and_ GPU fully saturated: https://twitter.com/andreif7/status/1328820823877095425?s=20 32.6W