What are the technical reasons why the M1 chip is so fast. Does Apple use any exotic tricks? Can Intel and AMD do the same trick in their chips?
Fortune was talking about it almost 10 years ago.
https://macdailynews.com/2011/07/05/how-apple-became-a-monop...
Of course they’ve been accused in the Supreme Court of being a monopsonist with regard to the App Store, but that’s a separate discussion.
(Update: regrettably my economic theory is weak, so while I think it makes sense to apply the term, after more reading I can't argue the case with a reasonable level of conviction.
Certainly there should be some way to describe a buyer who locks up a supply of components to the exclusion of competitors, as Apple has done on multiple occasions dating at least to the original iPod hard drives.)
See the box at the top of article showing difference between monopoly, monopsony, oligopoly, oligopsony.
A monopsony would mean the buyer gets to dictate all the terms, and the seller has to sell at very low price. In this case, TSMC has something very valuable, and Apple is paying a considerable amount (presumably, I haven’t seen real numbers) to the seller to make sure no one else gets it.
In fact, TSMC is a monopoly, the only seller of a much sought after product.
This is different with say Windows architecture. In windows vendors work on generic cpus/gpus/motherboards. Some hardware isn't exactly available or uniformally accessible due to drivers and such. This is kind of the huge benefit Apple has by creating the hardware, OS, drivers, languages and libraries.
Calling it ‘a cheat’ is somewhat weird.
The point is that a general purpose CPU is a bad solution for a personal computer.
The neural engine is about processing the ambiguous world that humans occupy so that the human doesn’t have to translate so much for the computer.
The GPU is about communicating visually with humans.
Point is that if you try to run a different OS, it will likely run slower, perhaps significantly slower than running macOS
If another OS doesn’t take advantage of the specialized processors, it will run slower, but only on specialized loads.
The future is keeping Apple "fresh", "cool", etc... While keeping product costs low to compensate for a falling market share.
Really says something that Apple is willing to compromise on long term hardware performance.
I don’t see any logic that supports this claim.
Apple, on the other hand, has been running rings around Qualcomm and other ARM vendors for a while now, and sharing tech makes it all the more important that they continue to be competitive.
They Apple'd themselves and locked themselves in.
In 2 years, secrets aside, Apple would need to beat all 3 big dogs to be the top. How long can they sustain that?
But why would they? The users don't buy for performance.
And battery life seems to be the sleeper performance metric here. Every review is going to be "you could get 6+ more hours if you just bought the macbook"
There's been a lot of hype about things like "specialized hardware for NSObjects," but AFAIK that's more of a flawed/outdated design on Intel' side than a specialization on Apple's: the "magic" is really just ARM's weaker & more modern memory consistency model, which makes things like reference counting substantially faster.
Customers may end up with having the choice between a very high performance system at competitive prices, vs an open architecture which has more flexibility in what you add, but which will ultimately cost more and have worse performance.
You can see this play out in different markets. Just look at Tesla e.g. They are taking the Apple approach to car making with full vertical integration. They are doing everything from making their own alloys to creating their own machine learning hardware.
I think we are at some kind of paradigm shift and the old guard has been caught off guard.
- 5nm process (1.8x the density of 7nm [1])
- unified memory directly on the SoC
[1] https://hexus.net/tech/news/industry/130175-tsmc-5nm-will-im...
The main reason for M1's performance is just that Apple managed to get an 8-wide CPU to work & be fed. That's it. That's the entirety of the story of the M1. Apple got 8-wide to work, while Intel & AMD are still 4-wide. ARM's fixed length instructions are helping a ton there, but Apple also put in work to feed it.
All the other shit about specialized units & unified memory & vertical integration is all irrelevant and mostly wrong. The existing Intel Macbook Airs are all unified memory with specialized hardware units for specialized tasks, too (Intel QuickSync is 9 years old now - dedicated silicon for the specialized task of video encoding). Apple did absolutely nothing new or interesting on that front. Other than marketing. They are magic at marketing. And also magic at making a super wide CPU front-end, with really good cache latency numbers.
They also managed in a cheap platform (mac mini = $700) get 4266 Mhz memory working with about half the latency of any x86-64 I've seen. The mac mini can manage a random cacheline (assuming a relatively TLB friendly pattern) of around 30-33ns.
Maybe the usual intel core i7/i9 or ryzen 5000 look rather quaint with their 60ns or higher memory latencies.
Where are you getting those numbers? From https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste...
"In terms of memory latency, we’re seeing a (rather expected) reduction compared to the A14, measuring 96ns at 128MB full random test depth, compared to 102ns on the A14."
Which would put M1's DRAM latency at worse than a modern Intel or AMD desktop, which is measuring around 70-80ns: https://www.anandtech.com/show/16214/amd-zen-3-ryzen-deep-di...
On pretty much any current CPU if you randomly access an array significantly larger than cache (12MB in the M1 case) you end up thrashing the TLB which significantly impacts latency. The number of pages that can be quickly access depends on the number of page in the TLB.
To separate out TLB latency from memory latency I allow controlling size of the sliding window for randomizing the array, so that only a few pages are heavily used at any one time, while visiting each cache line exactly once.
That's exactly what the brown "R per RV prange" does. For more info look at the description at: https://www.anandtech.com/show/14072/the-samsung-galaxy-s10p...
My code builds an array, then does a knuth shuffle, but modified so the maximum shuffle distance is 64 pages, so the average shuffle is 32 pages or so. I get a nice clean line at 34ns. With 2 or 4 threads I get a throughput (not true latency) of a cacheline every 21ns. With 8 threads (using the 4 slow and 4 fast cores) I get a somewhat better cacheline per 12.5ns.
Pretty stunning to have latencies that low on a low end $700 mac mini that embarrasses machines that costs 10x that much. Even high end Epyc machines (200 watt TDP) with 8 x 64 bit memory channels have to try hard to get a cacheline every 10ns.
Eh? That's not how memory latency works. The cheaper consumer chips with "non-spec" RAM and without ECC are regularly better here than the enterprise stuff. This isn't something that scales with price.
The M1's memory bandwidth is ~68GB/s, which is of course a tiny fraction of AMD Epyc's ~200GB/s per socket.
Epyc's latency isn't even competitive with AMD's own consumer parts, so I'm really not sure why you're surprised that Epyc's latency is also worse than the M1's?
Even more impressively is that the random cacheline throughput is also excellent. So if all 8 cores have a cache miss the M1 memory system is very good at keeping multiple pending requests in flight to achieve surprisingly good throughput. Granted this isn't pure latency, so I call it throughput. Getting a random cacheline per 12ns is quite good, especially for a cheap low power system. Normally getting more than 2 memory channels on a desktop requires something exotic like an AMD threadripper.
Do I get my Turing tester badge?