Apple's M1: the fastest and bestest ever silicon = revolution? Nah, there's far more interesting stuff happening in tech that matters to everyone
- Reference: 1607333409
- News link: https://www.theregister.co.uk/2020/12/07/apple_apple_apple_wip/
- Source link:
Nah. What Apple has done is very impressive, but for every piece of spot-on engineering, there’s an aspect of the company’s business model that limits its ultimate importance. It may seem like a bold step to change the basic architecture of all MacOS devices, but it also reveals a timidity that may in the end render it curiously impotent.
Let’s celebrate what Apple got right. The first two devices with the M1 at their heart are breathtakingly fast, superbly power-frugal and surprisingly good value. The Mac Mini has no peer at its price point, and it puts in a performance equivalent to Intel’s fastest chips - although not AMD’s, explaining why Apple has not substantiated its marketing claim that the M1 is "the world’s fastest CPU". Still, that’s some screamer.
The system software is also impressive. You won’t notice that your x86 applications are going through a translation layer - but then, you don’t anyway, as no processor has run x86 code natively for decades, there’s always a much more efficient core ALU chewing through microinstructions after the x86 code has been stripped down. What you can do in software you can do in hardware and vice-versa - that’s a basic rule of computation - and which you choose is just a matter of efficiency and performance.
How Apple's M1 uses high-bandwidth memory to run like the clappers [3]READ MORE
Don't stray over the garden wall
This is where Apple’s true focus lies. The M1 isn’t a processor, it’s a system-on-chip (SoC) of exceptional efficiency. Apple has combined sixteen compute cores of various sorts with a ton of cache and a sea of RAM, together with most of the ancillary circuitry that makes up a modern computer.
It is a modern Apple computer on a chip, very fast, very power efficient and very profitable - but it can’t do anything else. Future iterations of the M-series chips will be beefier and even more capable, but they too will be Apple computers. MacOS and apps built in Apple’s development ecosystem will be finely tuned to make the most [4]efficient use of the memory architecture , something that can make a huge difference in performance between systems that look on paper to be pretty comparable.
Because Apple controls its entire hardware and software stack, it can make the whole thing work extremely well, and make the company a lot of money at it. But that’s it. Apple won’t sell the chip to anyone else, partially because Apple would rather sell Tim Cook’s granny than sell anything other than finished computers, but mostly because a general purpose part would have compromises that takes the shine off.
You can’t add RAM to your Mac Mini M1, because putting an external memory bus on the SoC would drag it back to the same limitations of physics that any other CPU faces. Much better to have all those tightly integrated very wide on-chip buses, and slam the door on the big nasty slow world outside.
And, shocking to relate, there’s a much, much bigger world of computing than just the bits with a piece of fruit on the front. How many CPU cycles of your daily IT workload happen on your local CPUs, and how many in the distant caverns of Amazon, Microsoft and Google? How many of those will be affected by Apple’s industry-shaking switch?
The real problems - the interesting problems - in computing are never solved by an SoC. The real problems - the interesting problems - are moving data, not doing sums in a CPU. The very first electronic programmable computer, the Bletchley Park Colossus, had 1930s-era valves at its heart, but even they were limited not by their intrinsic speed but by how quickly data could be streamed in from paper tape. The most modern clusters of GPU-powered supercomputing live and die by the interconnection fabric and the logistics of getting huge data sets out of store, into compute, and out again.
Here is where revolutions happen. Intel has no magic right to life here either, of course, but it’s a market it knows well that might as well be on Pluto for all Apple cares. The company wants to make shiny phones, shiny laptops and shiny desktops, and it wants to make as much money from shiny phones, shiny laptops and shiny desktops as it can without breaching the walls of its own magic kingdom.
How big that kingdom can be is up for discussion, but by and large most people who want Apple phones, laptops and desktops have them. For my own reasons, I choose to eschew MacOS and iOS as my primary platforms, so it doesn’t really matter how superfast they run; most of my friends are Apple fans and will doubtless migrate to the new world in time, as they would no matter what.
The M1 chip is Apple daring to make ever more money by being more Apple-y than ever. It shows with absolute clarity that Apple has no ambitions beyond that, no plans to change its world, no new ideas or new ambitions. Extreme efficiency is it. Which is fine, it’s clearly an excellent way to be very rich, and if that’s all you want from life then congratulations.
It’s not much of a future, though. Extreme efficiency in an ecosystem invariably leads to extinction when that ecosystem changes. Enjoy the beautiful plumage of those benchmarks, but watch the sky for asteroids coming out of the clouds. ®
Get our [5]Tech Resources
[1] https://www.theregister.com/2020/11/10/apple_silicon_debut/
[2] https://www.theregister.com/2020/11/12/geekbench_stats_show_apple_silicon/
[3] https://www.theregister.com/2020/11/19/apple_m1_high_bandwidth_memory_performance/
[4] https://www.theregister.com/2020/11/19/apple_m1_high_bandwidth_memory_performance/
[5] https://whitepapers.theregister.com/
I fear that too much shiny is taking a toll on some people's attention span.
He just that article, look at this:
"The real problems - the interesting problems - in computing are never solved by an SoC. The real problems - the interesting problems - are moving data, not doing sums in a CPU."
See? I know, it's Shocking, isn't it.
Re: I fear that too much shiny is taking a toll on some people's attention span.
One of the reasons Apple silicon is so fast is that that the RAM embedded in the SOC is shared between the CPU and GPU, and this removes the need to move/copy data between the two.
You forgot 'Apple is doomed'.
You got paid for this drivel!
Wow.
I'm no fan out, but bloody hell that just smacked of "I know better than the richest company in the world"
Re: Wow.
The Register has a history of mocking the first few generations of Apple products in a new line. This included the iPod and iPhone. So it is unsurprising that this tend continues with Apple silicon.
Remember that Apple once touted racks of Minis as a replacement for XServe
But I don't think any BOFH would consider a device with with fixed storage, fixed RAM, and no gigabit Ethernet as a server device.
Also people who use an eGPU with a 2018 Mini are also a bit hosed when it comes to an upgrade path at the moment.
That was an article?
You may as well have added 'just do your research!!!' to the assertion that there's 'far more interesting stuff happening in tech that matters to everyone'
Butt Hurt?
"...and if that’s all you want from life then congratulations." Did you buy MacPro wheels? Pretty bitter...
Why do you write about Apple if you don't use it? Seems a bit biased?
Re: Butt Hurt?
Why do you write about Apple if you don't use it? Seems a bit biased?
Even Apple things can be compared with not-Apple things.
I guess the bias creeps in when the Apple things are found wanting, right?
Re: Butt Hurt?
So if he used / owned Apple products it would be unbiased?
Re: Butt Hurt?
At least he could be speaking from experience then, and not parroting what other non apple users are saying? I get the walled garden push back. I resist as much of it as I can, and faced with it being a problem, I don't use that service or feature of the walled garden. I use Apple gear, but don't drink the kool-aid...
Why does it hurt you so much, you're not an apple shareholder are you?
Every time I see someone making a big emotional investment in an industrial corporation that only exists to turn a profit I wonder why. Why did it all have to go so wrong?
Tim Cook's Granny
At the time of writing, she's not listed on ebay
[1]https://www.ebay.com/sch/i.html?_nkw=Tim+Cook+granny
Right now that search brings up 3 Hammer Horror cards - at #4, #5 and #6, and much cheaper than the top 3 results which are Tim Cook signed memorabilia
[1] https://www.ebay.com/sch/i.html?_nkw=Tim+Cook+granny
Well...OK
I take your point. It's Apple doing what Apple does and doing it very well. I'm not convinced that writing off architectural changes to produce something that is very fast and very efficient is a complete goer though. Energy usage by a single consumer device may be inconsequential compared to energy use by a screamingly fast 4U beast but (guessing TBH, I've not checked the numbers) there are a lot more consumer devices than there are 4U beasts. Less energy used at the consumer end is less energy used overall and that's mostly a good thing.
The other bit that doesn't seem quite fair when writing this off is that it's a different approach that has been shown to give good results. That by itself is likely to cause some thinking outside of Cupertino about similar architectural changes; if only to get up the noses of Fanbois bleating on about their phone going 15 weeks between charges (not the greatest of reason for doing anything mind). Even if it is just phones and laptops at the minute there's little historical precedent within IT for anyone respecting artificial market segment boundaries when it comes to implementing good ideas. Aforementioned 4U beast only with 1/10th of the power consumption for the same performance would be a big saving for anyone's energy bill.
Yes, I know that it's more about shunting data around as quickly as possible rather than chewing through it at the moment. That's likely to change; storage, bandwidth and compute tend to jockey around playing pass the bottleneck so it'd be surprising it at some point in th next couple of decades people weren't complaining about not having enough fast enough cores to chew through the bits they're delivering. It'd be nice to have the compute side of things going screamingly fast when questum entangled transfer interfaces (that's made up BTW) are drowning them under a tsunami of data.
Rosie
Re: Well...OK
I agree with all you said.
But the 4U example, if we go with AMD and their 64 core CPUs (7702P), they currently draw max ~4.4W per core (from the TDP) compared to Apples cores (firestorm) of ~3.45W.
But AMD is currently using 7nm (14nm for IO), Apple is built on 5nm, so AMD switching to 5nm should drop that power consumption.
So on the high end, high core count Apple haven't really done much better, but they may be able to reduce their power draw on the higher counts like AMD have done. But so will AMD going to 5nm and probably with their next design. Which will be before Apple even consider such a CPU.
Natural selection
" It’s not much of a future, though. Extreme efficiency in an ecosystem invariably leads to extinction when that ecosystem changes. "
Having studied ecology, I would say "extreme specialisation" rather than "extreme efficiency" and I might be a bit cautious about "invariably" . Nevertheless Rupert Goodwins makes an important general point. The fundamental advantage of the personal microcomputer that led it to dominance was versatility. The original IBM machine had practically no peripherals on board - you could select the ones you wanted (from an ultimately massive range of alternatives) and plug them in to create a system to your own specification. Now, that's much less possible, and certainly not possible at all with Apple kit.
Shortsighted analysis
If the only laptop/desktop products Apple were to make with its new silicon were the three already released, then the author of the article might have a point. But it seems likely that Apple will release desktop/workstation machines with even faster silicon and more RAM. Some people will appreciate faster webpage loading or faster games, but the big win for Apple is likely to come from niche application areas in which an Apple silicon-based computer is as fast as other computers costing, say, five times as much. Machine learning (ML) might be such a niche area: https://www.zdnet.com/article/the-new-m1-macs-make-cutting-edge-machine-learning-workstations/
Although ML is niche, it is a large and quickly growing niche. Perhaps Apple will even produce blades containing multiple SOCs for use in data centres. If Apple can compete successfully in the ML/data-centre niches, expect its stock price to rise significantly. And also expect other CPU manufacturers to start designing CPU+GPU+RAM SOCs to compete against Apple silicon. So, yes, I would say Apple Silicon has a good chance of being revolutionary.
For what it's worth, I don't own any Apple products (I prefer Linux and use a dumb phone).
Re: Shortsighted analysis
For ML, there are far faster ASICs out there than what Apple Silicon can do.
The games console model
Isn't what Apple are doing essentially the same as Sony, Microsoft and Nintendo do with their games consoles? A lot of people like this approach – a machine you can just turn on and it does pretty much what you expect and you will probably want to upgrade to a new model every 3-5 years. Come to think of it, that sounds like most consumer products from cookers through to large format TVs. Its almost as though Apple want to appeal to consumers...
Except the Cloud is going in the same direction
Amazon is not-so-slowly but surely moving towards it own Graviton2 SOCs for AWS, even if they are not yet as tightly integrated, and I would very surprised if Apple didn’t have a data center variant of the M1 (or perhaps the M2 that will inevitably follow for higher-end MacBook Pros, iMacs and Mac Pros) for its own extensive data center operations (running Linux, BTW).
...leads to extinction when that ecosystem changes.
This is the company that switched from Motorola 680x0 to PowerPC to Intel to Apple's own take on ARM pretty painlessly*. They seem quite good at dealing with major changes to me. Pretty ridiculous article if you ask me.
* I'm not counting the 6502 - switching from that also involved the user switching to a quite different OS/Interface.
Headline-whoring
"there's far more interesting stuff happening in tech that matters to everyone"
OK...
Like what?
Maybe actually write an article about that?