Discussion Apple Silicon M series thread

[moinmoin]

Going to SoC is quite radical for PC world.

Intel's Atom processors as well as AMD's Zen processors are SoCs or partially at least SoPs.

 

[thunng8]

Dude you need to calm down. You get way too excited and irritable during these conversations

Nobody answered the question, probably because no one is familiar that software. I just found it odd that the Intel CPU had such low vector performance according to that benchmark, when vector performance has typically been one of their strongest assets.

That's why I wondered whether it was properly optimized for AVX/AVX2. From the performance, I doubt that it has been.

Case in point, in this thread, several users have uploaded their results. Here's a result for an AMD 2700x:

And here's one for a 9900K overclocked to 5ghz:

Now to me that's very weird, given the raw SIMD performance of both of these CPUs. The 9900K should blow the doors off of that 2700x since the 2700x uses 128 bit paths if I'm not mistaken.

When they are measuring vector they dont mean they are measuring the vector units of the CPU. They are measuring how fast their software is at vector illustration on the cpu. Vector illustration might not use the cpu’s vector units at all (I don‘t have insight on what part of the cpu it stresses) but we know it is something that cannot be offloaded to the gpu since their are no vector gpu results

 

[amrnuke]

they know perfectly well everything they're not doing optimally and have a plan to improve this aggressively every year over the next few years.

You think that this doesn't apply to AMD or any other well-run company? AMD know they can't keep clocking higher indefinitely, and they know TSMC and others have warned that clocks may actually need to come down on more advanced nodes, so they're designing for IPC (hard with x86, just as it is with Arm!) and leveraging their position. Apple and Samsung already realized they had an IPC advantage on Arm ISA on top of that, so they've been careful to design wider and slower cores to leverage their position.

Bottom line is that team x86 bet on the wrong horse (getting performance via raw frequency rather than smarts) and every year they are finding fewer and fewer niches where that bet pays off.

That's interesting. I asked a question before, I think it got mixed in the shuffle. You know that things like branch prediction and overall uarch decisions are largely predicated on ISA. For instance, if your ISA has predication, your uarch and branch prediction design (or lack thereof) would be vastly different. So why do you keep implying that somehow the ISA-dependent uarch design can be used to judge whether one team is smarter than the other?

Also, reminder, the majority of Apple's gains in performance on their fast core for the last couple of (eh, more than that) years have come in the majority from increasing clock speeds and leveraging process node improvements. Only single digit raw PPC gains (not attributable to process node) have been realized each cycle, and (shockingly) sometimes even <5%.

It is astonishing the denial Team x86 is in about this.

I think most people just want to see advances in technology. That's the case with me. But I have a lot more skepticism about the ability to just toss a bunch of Firestorm cores on a chip and suddenly it's a great workstation or server processor. That's not how it works. So when I see folks calling the end of x86 in the server market, I think that's kind of premature, even if it may not end up being incorrect.

At least from many people in these types of discussions, what I'm seeing is Arm/Apple evangelists denigrating x86 and the chip designers who work with it as being less intelligent and warning about a shrinking x86 market. I see x86 evangelists expressing significant cynicism about the ability of Apple, Annapurna, or any other manufacturer to actually produce the server and HTPC and workstation performance that Arm evangelists claim will come so easily. Neither side is, or can be, wrong right now, because all of this is completely based on prediction and expectation rather than real hard data.

So rather than rely on data and numbers and information, so many rely on speculation and manufacturer claims and extrapolation, and take sides and act like children, even to the point of ignoring data that runs opposite their belief or preference. There are no sides in this, only the data. All uarch and ISA and CS advancements help all of us in this forum.

 

[guidryp]

Why wouldn't using an older SoC save on BoM? Right now the M1 is using the most expensive node in the world(5nm). Next year, 5nm will be cheaper. The year after that, 3nm will take over and 5nm will be even cheaper. All the while, 256GB SSDs, 8GB RAM should also get cheaper.

I could definitely see Apple shaving off $300 in 1 or 2 years and also willingly lower the margin in order to promote their services. If they can lower price of components by $150 in 2 years and then also take $150 lower profit margin, they can release a $700 Macbook.

Because it's already less than $100. You can't shave $150 off.

They also make near NOTHING on Mac services. Macs are just a fraction of iPad/iPhone sales and always will be.

Plus they have no lock in on Macs.

 

[Thala]

I think most people just want to see advances in technology. That's the case with me. But I have a lot more skepticism about the ability to just toss a bunch of Firestorm cores on a chip and suddenly it's a great workstation or server processor. That's not how it works. So when I see folks calling the end of x86 in the server market, I think that's kind of premature, even if it may not end up being incorrect.

If you want to see advances in technology, you should not put your bets on the awful x86 ISA. Whatever you can do with x86 you can do better (with less power and/or higher performance) with ARM - indeed i do not believe that Apple engineers are inherently more capable than Intels or AMDs. The engineering teams working within the solution space given by the instruction set architecture specification.

 

[Eug]

Got the iPhone 12 Pro Max on Friday. Man, this thing flies! I knew it was much faster on paper, esp. considering my 7 Plus is so old now, but I’m a little surprised it still feels so much faster even just for basic OS navigation and light mobile websites. And it’s nice to see my iPhone doesn’t appear to throttle significantly. FWIW, this is what I got the one and only time I’ve run Geekbench on this phone.

View attachment 33868

Just a reminder that the phone runs at 3.0 GHz. So, if we were to extrapolate the A14 Geekbench 5 single core speed of 1604 up to 3.2 GHz, that would give a single core score of 1711, which is right in line with what M1 is scoring.

Given the volumes of A14, I can’t imagine the yields are bad, and they’ve increased the clock speeds on the much larger M1.

BTW, I ran the bench with my case on. It’s not a thick case but it covers the back and all the sides, yet no throttling here. I know it’s a short test but it’s possible to get older iPhones to throttle just with this test if you pre-warm the phone by charging it in a case.

Also, battery life seems to be excellent, although part of that is probably the dark mode I’m using, which would help these self-emissive OLED displays.

It should also be noted that this thing is scoring higher than my 27” iMac i5-7600, and I bought that thing just 3 years ago.

 

[guidryp]

Where will the savings come from? Well, why should a low-end macbook cost much more than an iPad (not iPad generically, the device called iPad that sells for $330)? The minimum macbook has a larger (but lower quality, non-touch) screen, a little more flash, a little more RAM, a keyboard. Total BOM maybe $100?

Is this a joke? Everything on MBA is vastly more expensive than the $330 iPad. It's not a remotely valid comparison.

Here is a more realistic comparison, the 13" iPad Pro and MBA:

Model Price Screen Size, Storage, RAM
MBA $999 13" 256GB 8GB
iPro $999 13" 128GB 6GB

For the same price you get an iPad with less RAM, less Flash Storage, no keyboard, smaller battery.

You already get more for the money with MBA, so margins are already significantly thinner on the MBA, than on a 13" iPad, plus there is no Service lock in on the Mac, to make up for any lack of margins.

 

[amrnuke]

If you want to see advances in technology, you should not put your bets on the awful x86 ISA. Whatever you can do with x86 you can do better (with less power and/or higher performance) with ARM - indeed i do not believe that Apple engineers are inherently more capable than Intels or AMDs. The engineering teams working within the solution space given by the instruction set architecture specification.

I can only laugh that you think I'm putting my bets on any ISA. Why should I do so? So I can brag that I was right? LOL.

As you can tell from my posts in this thread and others, I have congratulated both Apple and AMD for what they've accomplished working with their target market and with the ISAs they utilize. So, let's evaluate your claims:

I am not sure if it's bias or ignorance when you say that "whatever you can do with x86 you can do better [...] with ARM". That is a false statement as things currently stand, because ARM does not hold any single-thread or multi-thread performance lead over x86 (at least in the reviews we have, namely A14 deep dive by Andrei and Graviton2 evaluation, and so on). The 5950X has faster single-threaded performance than the A14 and Rome beats Graviton2.

I'll refrain from my commentary other than to say that it would be fairly damning if Arm is a superior ISA AND Apple's engineers are smarter as others claim - and they still can't produce a chip faster than Zen3. Regardless, much of this evaluation is somewhat irrelevant, because the A14 and Firestorm cores are designed for a purpose that is different from (and on a different ISA than) the Zen3 core. So I am not (nor is anyone else) in a position to make claims about the superiority of one ISA or one set of chip engineers over another.

So at the very least, in performance, x86 is still ahead. Efficiency? Arm is way ahead, absolutely. But in performance? No. Might it be in the future? Sure. But again... one cannot judge superiority because the two chip designers (Apple and AMD) are designing for different ISAs, for different markets, with different performance targets and power limits. So one cannot make any claims about ISA superiority.

As of now, you can speculate that ARM might be able to do everything x86 can do, and do it better, but you cannot state it as fact until it's actually done. (FWIW, I'd guess it will be done, and soon. But it hasn't yet. So I'm not going to make any unsubstantiated claims.)

 

[Roland00Address]

So you expect Apple to leave their working high margin strategy behind and allow competing subscription based services to fight for the same TAM?

It is a fact we are seeing the rise of web apps. This is not a change Apple wants but the world is more Dynamic than a single company. It is happening for several reasons one of which is the 30% but there are other reasons why it is happening completely unrelated to the 30%.

We are seeing the rise of web apps for the same reason we previously saw the rise of the web. Except web apps can run code on the device much like Java of old. Literally you just go to a website and then you get an “add to your home screen” and the progressive web app is now on your device and looks like a normal app to the user.

————

Apple does not like this but there is nothing Apple can do from stopping developers from making these apps.

Apple is going to fight tooth and nail for the 30% (as they should for native apps), see 2020 since June with Hey Email / Basecamp “drama” and then later ”drama” with Fortnite and Epic.

My point is people are going to try to move away from traditional native store apps regardless. (Even if native apps will never go away, and they do have several advantages!) Yet Apple has a golden opportunity, due to scale they can make good hardware cheaper than everyone else and keep the margins high. Second there hardware outperforms everyone else. Web apps require more cpu and ram than native apps, they also do not take advantage of any optimizations you can do for that specific hardware, PWA can not do metal, native apps can.

Thus Apple if they want to release a $700 iPad (with more multitasking features) or a $700 MacBook with the M1 in 2 years when the M3 goes 2022 model goes for $1k, they will be able to make money on hardware at these price points.

Furthermore once you sold the hardware it will be easier to sell other Apple services, including taking the service revenue of 30% apps that are native.

—————

The rise of web apps is not something Apple likes, it is beyond their control. Businesses will go to web apps and the web even if consumers prefer native apps.

Yet Apple has the advantage where they can sell $700 devices and it is not low margin to sell at $700 even for PC OEMs. And at $700 price points Apple will have the supply chain to make more money on the device sold for $700 than any competitor.

 

[Eug]

As of about 10 years ago, BOM of $999 MacBook Air was estimated at $718. Some spitballing by hardware analysts recently have suggested that the newest MacBook Air lowers cost by $150-200.

I think $200 sounds a bit too optimistic, but if that $150 number is in the ballpark of truth, then I don't see why Apple would consider selling a MacBook Air below $899 (aside from edu pricing).

However, at $899, that would make Apple's offering one of the more inexpensive ones with that kind of build, design, and performance, so I'm not sure that Apple needs to do even that. (Windows ultrabooks that are nice with good specs from major companies tend to be quite pricey.)

Back to the general claim that ARM can do anything x86 does, and do it better -- we already established it's not as fast in single-threaded operations. But perhaps the fastest 4-core, 6-core, 8-core, 12-core, 16-core, 32-core, or 64-core chips use ARM? As of what's been proven via independent reviews, no.

So at the very least, in sheer performance, x86 is still ahead. Efficiency? Arm is way ahead, absolutely. But in sheer performance? No. Might it be in the future? Sure. But again... one cannot judge superiority because the two chip designers (Apple and AMD) are designing on different ISAs, for different markets, with different performance targets and power limits. So one cannot make any claims about ISA superiority.

So as of now, you can speculate that ARM can do everything x86 can do, and do it better, but you cannot state it as fact until it's actually done. (FWIW, I'd guess it will be done, and soon. But it hasn't yet. So I'm not going to make any unsubstantiated claims.)

Hmmm... You seem to be arguing very emphatically that Apple hasn't proven itself in designs with lots of cores. Strictly speaking you are right, but really, we will just have to wait a few months to see what the truth will be. It's easy to be #1 when the competition hasn't even released their chips yet.

No, this is a scenario where speculation is the only possibility, but given the performance of their lower power entry level parts, I don't think it's wrong to think Apple may do quite well in their desktop parts.

Now what about the very top end of the multi-core designs? Well, you have to ask yourself, does Apple even care? Apple is unlikely to start becoming a server CPU supplier any time soon, so it's probably not high on their priority list. So, yeah, maybe x86 will retain the crown in these high end server parts, but to scream this out as an x86 win is just to completely miss the point.

Anyhow, I'm just impressed that my 2017 iMac scores lower in Geekbench 5 than my frickin' iphone just 3 years later. Geekbench is by no means the final judge, but still... Damn!

 

[guidryp]

I am not sure if it's bias or ignorance when you say that "whatever you can do with x86 you can do better [...] with ARM". That is a false statement as things currently stand, because ARM does not hold any single-thread or multi-thread performance lead over x86 (at least in the reviews we have, namely A14 deep dive by Andrei and Graviton2 evaluation, and so on). The 5950X has faster single-threaded performance than the A14 and Rome beats Graviton2.

I'll refrain from my commentary other than to say that it would be fairly damning if Arm is a superior ISA AND Apple's engineers are smarter as others claim - and they still can't produce a chip faster than Zen3.

That isn't a victory to harp about. $800, Desktop Monster CPU, 5950x scrapes a small single threaded win running ~5GHz against a phone running ~3GHz. Closer to embarrassment than something to brag about.

Certainly not what I would call that evidence that "they still can't produce a chip faster than Zen3". It' looks more like Apple could do that in their sleep if that was what they were aiming for. But they weren't.

We haven't seen M1 disected yet. But looking the Geekbench improvements, it probably already exceeds Zen3 ST without even really trying.

 

[iwulff]

That isn't a victory to harp about. $800, Desktop Monster CPU, 5950x scrapes a small single threaded win running ~5GHz against a phone running ~3GHz. Closer to embarrassment than something to brag about.

Certainly not what I would call that evidence that "they still can't produce a chip faster than Zen3". It' looks more like Apple could do that in their sleep if that was what they were aiming for. But they weren't.

We haven't seen M1 disected yet. But looking the Geekbench improvements, it probably already exceeds Zen3 ST without even really trying.

With all the hype surrounding M1 due to some geekbench scores and vague marketing and Apple fanboy hype, it's bound to become disappointing. Yes it's a great product, but probably less so than it now seems. We will see soon I guess.

 

[Carfax83]

When they are measuring vector they dont mean they are measuring the vector units of the CPU. They are measuring how fast their software is at vector illustration on the cpu. Vector illustration might not use the cpu’s vector units at all (I don‘t have insight on what part of the cpu it stresses) but we know it is something that cannot be offloaded to the gpu since their are no vector gpu results

Well this definitely goes a long way to explain those results!

 

[amrnuke]

Hmmm... You seem to be arguing very emphatically that Apple hasn't proven itself in designs with lots of cores. Strictly speaking you are right, but really, we will just have to wait a few months to see what the truth will be. It's easy to be #1 when the competition hasn't even released their chips yet.

Not arguing anything here. I am just stating facts and expressing my discontent with how so many people are picking sides and denigrating others based on dreams that aren't yet realized. I get being excited about things, speculating, anticipating. But it's this idea of calling AMD or Intel less smart, or holding up Apple as this paragon of the mastery of computer engineering is a bit out of hand.

No, this is a scenario where speculation is the only possibility, but given the performance of their lower power entry level parts, I don't think it's wrong to think Apple may do quite well in their desktop parts.

I fully agree that the M1 and Apple's work on core design will permit them to move over to Arm with a nice performance/efficiency gain for the workloads most people use their Mac minis, iMacs and Macbooks for. The A12Z was more than enough for this already. But remember that Apple is already a walled-off world in many ways, and moving over to Arm will only exacerbate that. This doesn't move the needle much - in that those not wanting to pay the Apple tax will still end up using x86-based laptops and desktops for their work. At the pace with which AMD are increasing IPC, I have reservations about other Arm-based chip designers' abilities to catch up quickly enough to make a difference to x86 in the near or mid term. Then there's the whole Windows barrier and business environment. One would hope Windows on Arm would mature quickly, so that we could see a lot more competition all around.

Now what about the very top end of the multi-core designs? Well, you have to ask yourself, does Apple even care? Apple is unlikely to start becoming a server CPU supplier any time soon, so it's probably not high on their priority list. So, yeah, maybe x86 will retain the crown in these high end server parts, but to scream this out as an x86 win is just to completely miss the point.

I never said they care. I was just emphasizing that @Thala's claim that Arm can do anything x86 can do, and do it better, is false as it stands right now. From single core to multicore, the performance crown still rests with x86.

Anyhow, I'm just impressed that my 2017 iMac scores lower in Geekbench 5 than my frickin' iphone just 3 years later. Geekbench is by no means the final judge, but still... Damn!

Same! My iPhone 11 Pro gets 1331/3361 and my 3600 on eco mode gets 1215/6152 (GB 5.2.5). Both are snappy, both do everything I expect. It's amazing the world we live in!

 

[Carfax83]

That isn't a victory to harp about. $800, Desktop Monster CPU, 5950x scrapes a small single threaded win running ~5GHz against a phone running ~3GHz. Closer to embarrassment than something to brag about.

Must I remind you there are up to sixteen of those cores on a single die, and also it's on an older process node. It won't be a completely fair comparison until we see what AMD does on 5nm.

Certainly not what I would call that evidence that "they still can't produce a chip faster than Zen3". It' looks more like Apple could do that in their sleep if that was what they were aiming for. But they weren't.

We haven't seen M1 disected yet. But looking the Geekbench improvements, it probably already exceeds Zen3 ST without even really trying.

Geekbench favors mobile chips due to its burst centric workloads, which is why we need to see the CB R23 results for the M1 to get a clearer picture of how it stacks up against x86-64 desktop CPUs.

 

[guidryp]

Must I remind you there are up to sixteen of those cores on a single die, and also it's on an older process node. It won't be a completely fair comparison until we see what AMD does on 5nm.

We are discussing core performance. Adding more cores is trivial compared to increasing the IPC of each core. And process doesn't really change IPC either.

The fact that ZEN 3, has to clock 66% higher to beat A14 shows the massive IPC advantage Apple has.

 

[Carfax83]

We are discussing core performance. Adding more cores is trivial compared to increasing the IPC of each core.

Are you a semiconductor engineer by chance? That's not what I've heard. Adding more cores dramatically increases the complexity of the uncore design, because having a big multicore CPU isn't beneficial if the CPU doesn't scale well due to poor core to core communication or not being fed enough bandwidth.

And process doesn't really change IPC either.

Not sure if you're being serious here. You think it's just a coincidence that Apple has managed these massive IPC increases over the years practically in lockstep with TMSC's process node advancements? Or that Intel's inability to increase the IPC of their designs wasn't correlated with the disaster that was their 10nm process node?

The fact that ZEN 3, has to clock 66% higher to beat A14 shows the massive IPC advantage Apple has.

Zen 3 was designed to achieve high clock speeds, and the A14 wasn't so I don't see what your point is. In the end, CPU performance is a function of IPC x frequency.

A CPU can have extremely high IPC, yet perform poorly due to being only 1ghz. Same thing for a CPU that can be 10ghz, but have extremely low IPC.

End performance is what matters, and IPC and frequency are both necessary to achieve that.

 

[amrnuke]

That isn't a victory to harp about. $800, Desktop Monster CPU, 5950x scrapes a small single threaded win running ~5GHz against a phone running ~3GHz. Closer to embarrassment than something to brag about.

Certainly not what I would call that evidence that "they still can't produce a chip faster than Zen3". It' looks more like Apple could do that in their sleep if that was what they were aiming for. But they weren't.

We haven't seen M1 disected yet. But looking the Geekbench improvements, it probably already exceeds Zen3 ST without even really trying.

Put another way, a monster desktop CPU that whips the crap out of the A14 in anything more than single-threaded apps, still wins also at single-thread apps.

Core power consumption is 20W on the 5950X at 5.05 GHz for ST loads, package power is 49W, and a lot of that comes from the fact that there is a lot more this chip is designed to be able to do than the A14. Again, as I've said all along, Firestorm and Zen3 are designed for different purposes and to expound on that, the full chips / SoC are designed for different purposes. You can't add RAM or a graphics card or multiple NVMe drives or attach USB drives or gigabit ethernet to an A14 SoC. But the 5950X needs to be able to do that, and it comes at a cost.

But we can look to the 5600X, which has a SPEC2006 score profile that still beats the A14 - its package power consumption is 28W for ST apps, 11W for the core. As you plug all this in, you get a sense of scaling. It's not really that far off. A14 has 2 big cores and 4 small cores. 5600X has 6 big cores and SMT. And small gains at the high end come at a power cost, as you can see with power-frequency scaling on the A12, A13, and A14.

(Keep in mind also that Andrei ran a version of SPEC compiled on the latest version of LLVM in Xcode 12 for the A14, but didn't update the A13 numbers. And the optimizations seen in that version haven't carried over to the x86 side either, so it's possible Zen3 is actually further ahead, as the gains in 456.hmmer are 80%+.)

 

[gdansk]

Ultimately, it seems people are operating on the assumption that Apple will scale core counts and clock speeds faster than AMD and Intel can increase IPC. We'll see.

 

[amrnuke]

We are discussing core performance. Adding more cores is trivial compared to increasing the IPC of each core. And process doesn't really change IPC either.

The fact that ZEN 3, has to clock 66% higher to beat A14 shows the massive IPC advantage Apple has.

Yes we are discussing core performance. Core performance of the 5950X, 5900X, 5800X, and 5600X are all higher than the Firestorm core.

Where does IPC come into play? How many of your programs require high IPC to work well? Isn't actual performance, or even efficiency, the best metric?

 

[Eug]

Ultimately, it seems people are operating on the assumption that Apple will scale core counts and clock speeds faster than AMD and Intel can increase IPC. We'll see.

Who said Apple will scale clock speeds fast?

Also, who said that Apple will scale core counts fast?

Personally I think Apple won't raise clock speeds that much, and I also personally think that Apple will only scale core counts up to a certain level that fits their business model. Apple is not in the server or HPC business.

 

[jeanlain]

Ultimately, it seems people are operating on the assumption that Apple will scale core counts and clock speeds faster than AMD and Intel can increase IPC. We'll see.

Apple doesn't need to increase clock speed to be competitive on the desktop, if we trust what we're seeing on the M1.
Can they increase the core count faster than the other increase IPC? I believe so yes. Apple SoCs already have large caches and high bandwidth. I don't see what particular challenge Apple could not solve.
As for higher IPC achieved by the competition. Apparently, it's not so easy to do with the X86 ISA and its variable instruction length limiting the width of the decode part of the CPU. And as far as intel in concerned, we haven't see a notable leap in IPC since Sandy Bridge, have we?
AMD CPUs have achieved nice IPC improvements by better using the caches. Do they still have a large margin of progression?

 

[guidryp]

Core power consumption is 20W on the 5950X at 5.05 GHz for ST loads, package power is 49W, and a lot of that comes from the fact that there is a lot more this chip is designed to be able to do than the A14.

This looks more like AMD advocacy, and the opposite of reality.

The 5950X isn't designed for more, it does significantly less, than an A14.

No GPU.
No Media encoder.
No AI engine.
No Flash controller.

The A14 SoC includes all that and does WAY more than the 5950x, which is just a CPU.

 

[guidryp]

Ultimately, it seems people are operating on the assumption that Apple will scale core counts and clock speeds faster than AMD and Intel can increase IPC. We'll see.

Easy bet IMO.

 

[SarahKerrigan]

Must I remind you there are up to sixteen of those cores on a single die, and also it's on an older process node. It won't be a completely fair comparison until we see what AMD does on 5nm.

What Zen3 part has sixteen cores on a die?