Solved! ARM Apple High-End CPU - Intel replacement

[Richie Rich]

There is a first rumor about Intel replacement in Apple products:

• ARM based high-end CPU
• 8 cores, no SMT
• IPC +30% over Cortex A77
• desktop performance (Core i7/Ryzen R7) with much lower power consumption
• introduction with new gen MacBook Air in mid 2020 (considering also MacBook PRO and iMac)
• massive AI accelerator

Source Coreteks:

 

[defferoo]

Like making sure design flaws kill unreplaceable keyboard, screen cable or something else within a few years and then the glorious T2 chip causes people to either spend 50% extra on top of that sweet initial price on authorised service or to buy a replacement new unit?

Gotta love that Rippoff As A Service model - if you own the company and not their PCs

they ain’t perfect, their recent macbooks actually have many flaws. which is why i haven’t purchased a new mac since 2009. their butterfly keyboards sucked, their thermals suck, their cooling solution for the current Air is a joke, their upgrade pricing sucks, their lack of repairability and upgradability sucks.

but i’m still able to use my old 2009 Macbook pro without much issue today, which is more than i can say for any Windows laptop i’ve owned. despite the flaws i listed above, i can appreciate the effort they put into their laptops and understand why they would charge a premium for it.

anyway, enough of that, like i said, there’s really no point in discussing it. i still think their silicon is going to be very impressive when it comes out later this year.

 

[coercitiv]

but i’m still able to use my old 2009 Macbook pro without much issue today, which is more than i can say for any Windows laptop i’ve owned. despite the flaws i listed above, i can appreciate the effort they put into their laptops and understand why they would charge a premium for it.

My father is still using a Dell Latitude I bought 10+ years ago and used myself daily for a few years, we can provide anecdotal examples from each side with ease. What I find very interesting is people who praise the quality of Apple products even as they intentionally avoid buying new products and focus on finding older but proven models. The conflict here is palpable, and brings back echoes from a time when the iPhone had a "correct" way of being held.

I can understand why someone who is both invested and accustomed to the Apple ecosystem would stick with their products even as they botch one or two product generations: it's a lot easier to stay in the ecosystem and also check product reliability from one manufacturer only. That is a calculated expense in both time in money. But that speaks about the consumer type Apple is able to (fully) satisfy, not the quality and/or reliability of well built Windows PCs. This type of behavior is also very interesting as it generally ignores bleeding edge performance in favor of reliability and comfort, which leads us back towards the main topic of the thread:

anyway, enough of that, like i said, there’s really no point in discussing it. i still think their silicon is going to be very impressive when it comes out later this year.

That's the problem here, we simply can't get our priorities straight.

On one side, we praise the fact that Apple can and will deliver more performance with custom ARM silicon, but on the other side when someone like @beginner99 points out Apple intentionally held out on performance in their existing lineups, the immediate response is to focus on other product attributes as main benefits and sale drivers. It's almost as if performance is important as differentiating factor, not as objective metric, as requirement for enjoying Apple products.

 

[DrMrLordX]

I don't see them doing HBM on the lower end Macs. It is too expensive and too niche

Okay, what about HBM2e?

 

[Eug]

A respected leaker is indicating some Apple hardware is ready to ship.

I personally believe one of the first to ship this quarter will be the new iMac (non-Pro). However, I also believe it will be Intel based, probably the last Intel iMac Apple releases. I also believe it will have a semi-custom Comet Lake S CPU that is 95 W TDP and 10-core at the high end.

 

[soresu]

The conflict here is palpable, and brings back echoes from a time when the iPhone had a "correct" way of being held.

As much as I love to sit back and watch the flames burn where Apple is concerned, do I remember correctly that some Android vendors also had antenna problems of a similar nature at some time or another?

 

[coercitiv]

As much as I love to sit back and watch the flames burn where Apple is concerned, do I remember correctly that some Android vendors also had antenna problems of a similar nature at some time or another?

Yup, this was a radio problem that probably gave headaches to engineers from multiple phone makers. The difference was Apple was selling the best phone experience, hence it could have none of that... and almost talked down on users for not knowing how to hold a phone properly.

 

[Doug S]

Okay, what about HBM2e?

Not any form of HBM.

 

[awesomedeluxe]

Okay, what about HBM2e?

I think it's safe to assume anyone talking about HBM is talking about HBM2E. But I believe HBM2E actually uses more power than HBM and is even less appropriate for an Air.

A respected leaker is indicating some Apple hardware is ready to ship.

I personally believe one of the first to ship this quarter will be the new iMac (non-Pro). However, I also believe it will be Intel based, probably the last Intel iMac Apple releases. I also believe it will have a semi-custom Comet Lake S CPU that is 95 W TDP and 10-core at the high end.

I agree that iMac is likely to be soon and that it probably won't be AS. Though, I was surprised to find the entry level iMac actually doesn't use external graphics, and all models top out at six-core coffee lake parts.

So I wouldn't be stunned to see the 8+4 part originally reported by Bloomberg here, rather than Apple setting the bar for themselves higher in the future with a better Intel part. It wouldn't be a machine that strives to be competitive with AMD graphics in higher-end iMacs, but if there's just one model and they say "this is the iMac, iMac Pro later" they might get away with it.

EDIT: I don't know how useful these are, but the A13 Bionic does not appear to get completely blown out by the 560X. In terms of size/yield, 16 GPU cores on a single-die is very doable, this breakdown looks OK to me for an iMac APU die. Put 4-8GB of HBM2E on there as last-level cache for the GPU and I think you have the current gen iMac 21 beat.

All of this is just to say an AS iMac is achievable without taking complicated steps like a chiplet design or some kind of in-house dGPU solution. I'm not sure I'd bet on it, but it's a lot more realizable than I'd first assumed.

 

[RasCas99]

This thread has diverged wildly into apple-hate territory. We should get back on track discussing ideas for Apple's new silicon - and ignore the people who are just on here to start fights about whether Apple's products are a good value.

Someone shared an interesting idea on macrumors that I thought I'd pass along here. What if Apple used HBM2E as cache for their high-end graphics parts, instead of treating it as a separate memory pool? Think Vega high bandwith cache controller: the HBM2E is last-level cache, and system memory is VRAM. The HBM2E would be invisible to the developer. This would allow Apple to use "unified memory" for all their systems without sacrificing GPU performance or having to go whole-hog and use HBM as system memory.

I think HBM2E as system memory is a really interesting idea, I'm not totally sold on it being practical. You would need two stacks of it, and it would have to be close-ish to the CPU cores (and still right up next to the GPU cores). On the other hand, I don't think just using LPDDR5 for everything is acceptable for the MBP16 and up. This struck me as an interesting middle ground if Apple really is married to the "unified memory" design.

Agreed things got sidetracked , I can

A respected leaker is indicating some Apple hardware is ready to ship.

I personally believe one of the first to ship this quarter will be the new iMac (non-Pro). However, I also believe it will be Intel based, probably the last Intel iMac Apple releases. I also believe it will have a semi-custom Comet Lake S CPU that is 95 W TDP and 10-core at the high end.

You believe or you dream

 

[soresu]

and almost talked down on users for not knowing how to hold a phone properly.

Oh damn I didn't know about that.

That's just plain insulting, to say nothing of shooting themselves in the foot from a PR POV.

 

[Eug]

I agree that iMac is likely to be soon and that it probably won't be AS. Though, I was surprised to find the entry level iMac actually doesn't use external graphics, and all models top out at six-core coffee lake parts.

That is not correct. The 2019 iMacs top out with the 8-core Core i9-9900K.

---

You believe or you dream

---

That "you're holding it wrong" thing was a total embarrassment.

Apple responds to iPhone 4 reception issues: you're holding the phone the wrong way

So, we just spoke with Apple and got the straight dirt on the reception issues that have been plaguing users today... and it's a little surprising. In essence, Apple cops to the fact there are reception issues with the new iPhone -- namely, that if you cover the bottom-left corner of the phone...

www.engadget.com

 

[awesomedeluxe]

That is not correct. The 2019 iMacs top out with the 8-core Core i9-9900K.

Oh, whoops. I guess the 27" does go to 8 cores.

I don't think it changes my message much though. The design challenges for AS on iMac in the near-term aren't insurmountable and it's reasonable to assume Apple might prefer to jump the low bar they've set for themselves now rather than clear a higher bar later.

I would still bet on Intel + AMD for the next iMac, but I'd say it's 2:1. My first impression was more like 5:1.

 

[Eug]

Oh, whoops. I guess the 27" does go to 8 cores.

I don't think it changes my message much though. The design challenges for AS on iMac in the near-term aren't insurmountable and it's reasonable to assume Apple might prefer to jump the low bar they've set for themselves now rather than clear a higher bar later.

I would still bet on Intel + AMD for the next iMac, but I'd say it's 2:1. My first impression was more like 5:1.

Well, the reason I'm betting on an Intel iMac very soon is partially because Apple has already said they are continuing to release Intel Macs, and the iMacs are due for an upgrade (as they were last updated 17 months ago). Furthermore, the iMacs are arguably Apple's most important "pro" desktop. Very few people buy the iMac Pro or the Mac Pro. Most so-called pros with Mac desktops actually own the iMac.

It is this machine (along with MacBook Pros) which needs an Intel update to maintain proper workflows with 3rd party software built for macOS Intel.

Apple announces Mac transition to Apple silicon

Apple today announced it will transition the Mac to its custom silicon to deliver industry-leading performance and powerful new technologies.

www.apple.com

Apple plans to ship the first Mac with Apple silicon by the end of the year and complete the transition in about two years. Apple will continue to support and release new versions of macOS for Intel-based Macs for years to come, and has exciting new Intel-based Macs in development. The transition to Apple silicon represents the biggest leap ever for the Mac.

 

[Eug]

https://pbs.twimg.com/media/ETI4ymHUYAEswK-?format=jpg&name=small

Probably fake but a GB5 score has appeared for the 2020 iPhone. 1658/4612 is really good for having 2 big cores and 4 small.

I can't read this since it is in Korean, but apparently it is indeed fake.

A14 긱벤치 내가 조작한건데 다 퍼져나가네;; - 스마트폰 갤러리

ㅋㅋㅋㅋㅋㅋ 시벌 어떡하지 맨 처음에 장난치려고 개발자도구로 수정해서 갤러리에 올렸는데 수출되네

gall.dcinside.com

 

[lobz]

Yup, this was a radio problem that probably gave headaches to engineers from multiple phone makers. The difference was Apple was selling the best phone experience, hence it could have none of that... and almost talked down on users for not knowing how to hold a phone properly.

Many people seem to passionately love arrogance. Maybe it gives a false sense of being superior?

 

[DrMrLordX]

I think it's safe to assume anyone talking about HBM is talking about HBM2E. But I believe HBM2E actually uses more power than HBM and is even less appropriate for an Air.

HBM3 is also potentially in play. HBM2e is supposed to be cheaper than HBM2 though.

 

[awesomedeluxe]

HBM3 is also potentially in play. HBM2e is supposed to be cheaper than HBM2 though.

I've heard a few people whispering about HBM3. Are there any firm manufacturing commitments? As well as I'm aware HBM3 is borderline vaporware right now.

It is theoretically a game-changer. You could theoretically fit a 32GB stack of HBM3 in the same physical/thermal/cost space as a 16GB HBM2E stack. This would enable a lot of designs that are only narrowly possible right now.

But I don't want to start thinking about it until someone can prove to me that it exists.

 

[DrMrLordX]

But I don't want to start thinking about it until someone can prove to me that it exists.

In the real world I have seen nothing. In the Apple universe . . . you never really know.

 

[IvanKaramazov]

It's hard to track down where they're coming from, but it looks like initial native Geekbench 5 results are coming in on the DTK. It scores 1098 / 4555, which suggests the initial scores got roughly 70% efficiency under Rosetta 2 (for single-core, multi-core is complicated by the fact that the Rosetta version couldn't access the 4 little cores).

Big Sur Beta 3 added iOS and iPadOS app support on the DTK, and it looks like someone essentially used the iPad Geekbench suite, so it treats it as an iPad Pro with 16GB of RAM.

 

[Doug S]

Wonder why the A12Z in the DTZ scores better despite having (or at least reporting) the same clock rate? More RAM wouldn't make a difference, but maybe that RAM is organized differently and helps performance on bandwidth and/or latency sensitive subtests? Too bad we can't see the individual results, knowing where it does better would help determine why it does better.

 

[IvanKaramazov]

It's not better on CPU, it's close to identical. The 2020 iPad Pro scores 1118 / 4626, so actually marginally higher. The DTK does have a higher Metal compute score though, at 12,610 v 9853 on the iPad. Maybe OS related?

 

[name99]

Wonder why the A12Z in the DTZ scores better despite having (or at least reporting) the same clock rate? More RAM wouldn't make a difference, but maybe that RAM is organized differently and helps performance on bandwidth and/or latency sensitive subtests? Too bad we can't see the individual results, knowing where it does better would help determine why it does better.

Has anyone dared to open up their DTK and see what the hardware looks like?
Remember the way RAM is mounted on a "traditional" A12X or A12Z:

The Packaging of Apple’s A12X is… Weird | Siliconica

sst.semiconductor-digest.com

But a DTK has quite a bit more RAM. Mounted in the same way, or as DIMMs, or???
It's possible that the mounting further from the core requires the DRAM to run slightly slower, and that explains the difference?

My guess is that tight on-package mounting (with implications for non-ability to alter RAM after purchase...) will be the standard for the real Macs, but for the DTK they may well have put together a quick sub-optimal scheme that both burns extra power and runs slower.

 

[name99]

Has anyone dared to open up their DTK and see what the hardware looks like?
Remember the way RAM is mounted on a "traditional" A12X or A12Z:

The Packaging of Apple’s A12X is… Weird | Siliconica

sst.semiconductor-digest.com

But a DTK has quite a bit more RAM. Mounted in the same way, or as DIMMs, or???
It's possible that the mounting further from the core requires the DRAM to run slightly slower, and that explains the difference?

My guess is that tight on-package mounting (with implications for non-ability to alter RAM after purchase...) will be the standard for the real Macs, but for the DTK they may well have put together a quick sub-optimal scheme that both burns extra power and runs slower.

BTW look at the metal score reported! That is interesting: 12610 compared to about 9900 for an iPad Pro w/ A12Z, so ~25% higher...
One possibility is the power/thermal envelope allows them to run the GPU quite a bit faster.
A second possibility is that the RAM system is indeed different from the iPad, so perhaps slightly higher latency (hurts CPU) but a lot higher bandwidth (helps GPU)?

 

[LightningZ71]

With the overall score not changing a lot, but with metal that much higher, it feels like the memory subsystem is roughly as capable as it is in the iPad pro, but it has a lot higher thermal envelope to keep the gpu section running at its boost clock for far longer (maybe permanently?). Given how memory sensitive we know Geekbench to be, any change in the memory bandwidth and latency would show itself in a big way.

 

[Eug]

It's not better on CPU, it's close to identical.

So does this mean Geekbench is OK as a cross platform CPU benchmark for iOS and macOS?