I think we agree more than not, though it's probably how I said it that led you in the wrong direction. I hate Mondays, and it shows.
Yeah, I sill like that idea, I just was grumbling about only 4 hp cores.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Very Interesting M1 Analysis
- Thread starter leons
- Start date
- Sort by reaction score
POWER9 is not arm, it is Power ISA. IBM POWER series are not well known at extremely fast 1T performance, so do you have any source to back your claim?
Technically, @Gnattu's post was accurate. At the moment of it's announcement on 10 November 2020, M1 was the fastest single-core CPU (Ryzen 9 5900X was released in January and Rocket lake is less than a month old).
It's a bit strange seeing you mentioning Power9... those are really not the fastest of the bunch.
So what? Neither is x86.
Well, a 40 core POWER9 system clocks in at 392 SPECrate2017_int_base, that's nearly 10 per core. Even the fastest AMD Epyc needs 48 cores to just get past that, and still doesn't get close to the POWER9 in SPECrate2017_int_peak.
Why is x86 so special? Different architectures are different architectures, neither is x86 means nothing.
POWER9 combines multiple single-threaded processing core into one cluster and called that a "core", which means the only way to use up that "core" is running multi-threaded tasks. A core may contain 2 or 4 "super-slices", each super-slices consists 2 single-threaded processing core. So the number of nearly '10 per core' is actually the score of 4 single-threaded processing core combined. IBM calls that cluster a 'core' because they are sharing cache, instruction fetch and sequencing unit.
This is the second year in which we get new Macs and even iPads with last years M1 processor. Apple will want to complete the transition to Apple Silicone as soon as possible, so updating the M1 machines has low priority. M2 (or however they will name their next chip) will focus on the large iMacs and MacBook Pros. Only when all these machines are released, work on a redesigned MacBook Air for next year can begin. The true M1 successor will then return to the usual 10-20% performance increase, but the new form factor will make these machines brilliant nonetheless.
M2 volume production has already begun… Looks to be based on next generation core, so expect 20% single core performance bump.
We will have to see if that is a successor to the M1 or a higher-end processor for the 30” iMac/16” MBP.
Possibly both! This year’s 30” iMac/16” MBP processor is next year’s 13” MBP/MBA/24” iMac processor?
The thing my sources will never ever tell me is whether they are planning on having two “current gen” chips with different numbers of cores or not. I still suspect they are, because I know they are working on a discrete GPU, and it just makes sense that if you have a separate GPU chip you can use some of the space on the CPU die for more CPU cores. No need to switch back to the small GPU to reduce power - now that Apple is in control of the whole architecture they could just reduce power, when appropriate, by turning off cores in the big GPU.
I’ve hypothesized that we’ll see a “return” of TDM as a port of Sidecar. Cue the disappointment of many who wanted to use the iMac with their PC or such.
It seems possible that the M1 generation might differ from M2 (etc.) — perhaps they’ll just have one variant for the first generation and then more beginning with M2.
I think that it’s possible that the larger iMac will get started with an M2 variant.
That's an extremely small segment of the market though. The bulk of PC sales are in the low and mid-range models, the vast majority of which are running relatively underpowered CPU/iGPU combos. Most people will never take advantage of a dGPU or the additional power that a gaming-class system offers.
If the M1 was made up of nothing more than the CPU and GPU cores, you may have been accurate here. However, the bulk of the chip is actually utilized by the Neural Engine, onboard caches, and controllers for I/O, TB4, etc. So while adding cores would increase the overall size of the SoC, it would not be nearly as much a difference in size as you stated.
POWER9 is a niche product at best, and really doesn't even factor into the M1 vs. x86 discussion. That's IBM doing its own thing in its own way and basically ignoring the rest of the industry.
Simple answer here: the CPU is not the only distinguishing factor between the machines. In fact, as you pointed out it's not a distinguishing factor at all between all of the low-end Macs and high-end iPads.
Those devices are still greatly differentiated though, by form factor, screen size, display technology, speakers, camera systems, microphones, touch capability, operating system, etc etc etc. Now you can not worry about getting great performance out of the processor and instead focus on the other attributes of the computers.
For the M1s initial release, it made sense to stick to existing and proven designs, if for no other reason than to eliminate one variable in the rollout. Given the sheer volume of leaks regarding the new MBP that will replace both the 4-port 13" and the 16" models, the redesign has already taken place. In fact, I would go so far as to say that the new Magsafe setup the 24" iMac is using may be the same setup those new MBPs use, including having the ability to run Ethernet through the adapter. From Apple's perspective, using the same adapter across multiple models would achieve economies of scale (which is already done with the M1 being used across five product lines now - base MBP, MBA, 24" iMac, Mac Mini, and iPad Pro), while also making it easier for the customer to replace their adapter or simply pick up a second one as a backup or to keep one at the office/one at home.
Going back to the MacBook Air, Apple tends to develop products in parallel rather than one product or product family at a time. They have probably had mockups of new designs in the labs for a while now and have been testing them as well as analyzing diagnostic data from the models already sold to customers to make final decisions regarding what changes will stick and which ones will be shelved for now. The other thing here is that Apple is always working on at least two generations of their products at a time, so when the A14 was announced, they were already finalizing plans for the A15 and most likely in early design phase for the A16. There is a 2-3 year lead time for processor/SoC development, so for Apple to be releasing new generations on an annual basis means simultaneous development has to be practiced in Cupertino.
I think this may underestimate the importance of Apple Silicon and the M1. With x86 (both Intel and AMD), it's easy to get bogged down in things such as clock speed, core count, etc. - most of which will never be noticed by the average user. With the M1, the processor itself almost becomes secondary, and things such as battery life and weight (for laptops), screen size, ports, speakers, camera, etc. become the focus of consumers' attention. Since those are the things that truly affect the user experience, it makes sense to focus on those aspects rather than on core count, raw clock speed (which is a horrible indicator of performance today, and has been ever since AMD first started shipping dual and quad-core machines), etc.
From Apple's perspective, switching to their own SoCs achieves cost savings across the board. That allows the company to make quality of life improvements such as upgrading the MBA screen to the same P3 display used in the MBP without increasing the overall MSRP of the device. Keeping with that line of thought, Apple can also make QoL improvements to other products in its lineup, such as what we have seen regarding the changes being made to the upcoming 4-port MacBook Pro. I have seen people both here and elsewhere on the internet complaining that Apple should have slashed the pricing across the board of their products with the shift to Apple Silicon. Personally, I'd rather see Apple make those types of improvements that may not show up in a benchmark (but show up in the day to day user experience) while keeping pricing at current levels or possibly dropping slightly.
As others have pointed out, those results are from running 320 threads (8 per “core”) - this translates to under 1.5 points per thread. M1 is more than twice as fast per thread. In the end, it all boils down to POWER9 being a server-oriented CPU, specifically designed for server applications. It's benchmarks results are certainly impressive when you look at the entire machine, but they are not that impressive compared to other similar devices.
Actually, that's perfectly clear.
And if they want to maintain a unified memory model, this kind of setup allows it.
I can't see this making sense for anything but the Mac Pro though.
The two main reasons to separate the functionality like this are, as far as I'm aware, either that your silicon die size grows outside the reticle limit, or that you can reap substantial benefit from flexibility in terms of how you combine your building blocks. (The main success story here being AMDs Ryzen family of processors where using the same CPU chiplets allowed them to adress markets from mid level consumer desktop all the way up to the biggest iron with very short time to market. Note though - their product needs don't apply to Apple.)
Yield for a large SoC wouldn't seem to be an issue - as I wrote elsewhere here, AMD has made SoCs for the XBox sx and the PS5 that are 360mm2 and 310mm2 respectively on TSMCs 7nm process. More than 10 million of those SoCs have been sold to end users, and they are cheap, as they are put into boxes that also include 16GB of GDDR6, 1TB of fast SSD and bluray drives with full licensing et cetera, package sold at $499.
Making a monolithic die is obviously the most economical way of going about things at those die sizes or Sony and Microsoft wouldn't have chosen it, and TSMC 5nm is closely related. TSMC has made very positive noises about defect rate, and Apple has already manufactured a couple of hundred million A14 + 10-15 million M1s by now on that process. Ergo, it seems reasonable to assume that if Apple breaks the SoC up like this, the CPU and GPU parts need to be really big. We're not talking twice, or even four times as many CPU and GPU cores as M1 as that would still fit inside 200mm2 or so, but significantly beyond that. For the life of me, I can't see Apple going that big for laptops, and I'm not sure "Big iMac" volumes justify the design effort. (Nor that their current iMac design ethos would comfortably deal with the implied thermals.)
Maybe something like this could be justified for the Mac Pro, and if so, maybe a variation or the same parts could find themselves into a Big iMac. But that's a bit too much "if" and "maybe" for my liking.
If "The GPU will indeed be a discrete chip for the highest end product." turns out to be accurate, then I suspect this is Mac Pro only.
Different use cases for folks. The Mini definitely has appeal. It is quite the machine but Apple also let it sit on the back burner for so long, I think people forgot it existed and still do. Apple has been synonymous with "all in one" for a while.
I guess the need for the iPad would be for cellular and pencil support. Other than that I am scratching my head. I wouldn't spend that much on a tablet which is limited by iPad OS. This is the exact reason I purchased the M1 MBP. In the off chance I want to use an iPadOS/iOS app on a larger display I can now. Saved me a few hundred bucks in the process.
I won't knock anyones decision on what device they buy, even if it seems foolish to me.