Apple Transitioning Macs to Custom Chips With 'Incredible' Performance Starting Later This Year

[cmaier]

Not sure that open source RISC-V chips are where the innovation will be, but I guess we will see.

Nothing technically superior about RISC-V, so it’s all about license fees. But Arm license fees seem to buy you indemnification - if you get sued for infringing a patent and the reason is you were using Arm, Arm seems to take care of it. So I;m not sure the RISC-V model is all that compelling.

That said, many of my friends are interested in designing RISC-V chips just for the challenge of it

 

[Gerdi]

Looking at the top Geekbench 5 results for an A12z (4615 multicore 8 cores), it has the same performance of an i7-8750H (6 core/12 thread @2.5Ghz). Personally, I don't view that as "fast".

It does have enough computing power to compete with low-end Intel or AMD laptops.

It is 1/3 the performance of a 3900XT, which will be replaced by a 4900X this fall.

Lot of work ahead of them to compete in the desktop marketplace.

You know, "fast" is relative to the used power and never absolute. The A12z is extremely fast - faster than anything Intel or AMD at the respective power envelope.

To illustrate this, if Apple triple the cores it will already match 3900XT - and they are still below 30W at this point...now they can start to increase the frequency/voltage with the remaining power budget availble for desktops for instance - not much work ahead really.

 

[Nugget]

No. Lots of work behind them. They did it already. The a12z is a 2 year old processor. that means the design of it started around 4 years ago.

The i7-8750h is also a two year old processor.

 

[Gerdi]

The i7-8750h is also a two year old processor.

Last time i checked it is also 45 TDP. It should tell you something that for a match you need to chose a 5x higher power Intel design.

 

[ssgbryan]

You know, "fast" is relative to the used power and never absolute. The A12z is extremely fast - faster than anything Intel or AMD at the respective power envelope.

To illustrate this, if Apple triple the cores it will already match 3900XT - and they are still below 30W at this point...now they can start to increase the frequency/voltage with the remaining power budget availble for desktops for instance - not much work ahead really.

I guess you have to focus on power envelope, because Apple isn't delivering performance.

 

[cmaier]

I guess you have to focus on power envelope, because Apple isn't delivering performance.

His point is that the chip throttles in a phone because of its power budget. Allow it a bigger power budget and the frequency can be at max at all times. And then you can increase the voltage, and also the frequency. Power and frequency are linear. (Power is proportional to square of voltage, though).

 

[Gerdi]

I guess you have to focus on power envelope, because Apple isn't delivering performance.

I do not focus on power at all. But since power is literally the main limiter to performance, you would have to compare at iso power when comparing performance - everything else is meaningless.
It is not an architectural challenge to increase the performance at the expense of power after all - see cmaiers comment above.

 

[venom600]

At least no one has posted "You can tear bootcamp from my cold dead hands" yet.

I know its going to be a loss for a few, but a very very few. Apple did the math here. they know the active use rate for Bootcamp is probably in my guess around 1% of total active mac users. So for the near term, stick with Intel and who knows maybe someday Microsoft will ship a ARM version that is not a total train wreck*

*would not put money on that.

Where are you getting these figures from? Idle conjecture doesn't help anything. You can't know how many users use Base Camp because Apple hasn't talked about it. I think 50% of users rely on Base Camp. I made up that number just like you did, so our statistics are equally valid.

 

[Gerdi]

His point is that the chip throttles in a phone because of its power budget. Allow it a bigger power budget and the frequency can be at max at all times. And then you can increase the voltage, and also the frequency. Power and frequency are linear. (Power is proportional to square of voltage, though).

Indeed. When we are comparing used power of different architecures, we typically use switching capacitance (aka cdyn) = power/(voltage^2*frequency) as metric. For CPUs this can be scaled by CPI, which gives you something like switching capacitance per instruction - which reflects the power efficiency.

 

[Kpjoslee]

I guess you have to focus on power envelope, because Apple isn't delivering performance.

Why don't you hold onto your judgement until We actually see how Apple Silicon scales up on higher power envelope. It is more than likely that we will see higher core counts and clockspeed on Macs so stop throwing around baseless assumptions that it won't deliver on performance.

 

[confusedpurple]

Ah, that’s where you are confused, the Intel transition was announced in 2005 – you got the wrong decade!

“The game” hasn’t changed much at all, desktop machines are still fundamentally the same as they were in the 1980s - Classic Macintosh. The difference is that the dominant platforms are not now desktop PCs, but handheld computers, AKA “phones”. That phones are more important to Apple is neither here nor there in terms of desktops - they are instrinsically different things. The more Apple tries to make them the same, the worse the experience on the desktop will get. And I am of the opinion that this move is mostly likely to herald a further “integration” of palmtop and desktop OSes.

Wasn't confused at all Apple used PPC in the 90s. Your post was talking about how PPC was supposed to trounce intel - it didnt, hence the switch to intel.

The difference we have in this day and age is deliverability of software is vastly cheaper and more efficient, the tools we all use to build software are more readily available across platforms, and overall theres much less of a difference between platforms.

So yes it is a diferent game. A vastly different one.
[automerge]1593262929[/automerge]
Amazing that people are still being stupid enough to think the same chips in an iphone are going to power their mac. You've not got a clue what apple is going to use for their first mac on arm - it wont be an ipad/iphone chip, it'll be something that can take advantage of the fact that it doesnt have the power and thermal restriction of a portable.

 

[dysamoria]

Maybe watch the state of the union video and then come back and comment. Drivers and plugins were specifically addressed in that video along with everybody else’s concerns such as “this is the end of the world, Apple are locking down the App Store” blah blah.

No, they weren’t addressed, because, yes, I watched it. Once Apple decides to dump the x86 emulation, it’s all going away. Developers won’t port to ARM, certainly not if there’s an emulation in there for a while.

The only people who are taken care of here are average users and people that replace their Mac every few years. They mostly don’t use heavy-duty engineering, design, or content-creation software, and certainly don’t have lots of third-party tools and hardware.

Apple will ditch Rosetta just as quickly as they think they can. Just like they did last time. So much stuff was lost last time. It’s taken ten years to fill in those holes.

 

[Stephen.R]

Developers won’t port to ARM, certainly not if there’s an emulation in there for a while.

The very existence of intel Mac software following the PPC > Intel transition seems to refute your theory.

 

[cmaier]

No, they weren’t addressed, because, yes, I watched it. Once Apple decides to dump the x86 emulation, it’s all going away. Developers won’t port to ARM, certainly not if there’s an emulation in there for a while.

The only people who are taken care of here are average users and people that replace their Mac every few years. They mostly don’t use heavy-duty engineering, design, or content-creation software, and certainly don’t have lots of third-party tools and hardware.

Apple will ditch Rosetta just as quickly as they think they can. Just like they did last time. So much stuff was lost last time. It’s taken ten years to fill in those holes.

Developers ported from ppc->x86 despite the presence of rosetta. Why wouldn’t they do so this time, particularly given that Mac is more important than it was back then? (Market share is not much higher, but developer *profit* share is much higher on mac)

 

[ipponrg]

The very existence of intel Mac software following the PPC > Intel transition seems to refute your theory.

Look at it more of RISC to CISC back to RISC. There were a lot of advantages of going to x86 back then. IMO it was what grew MacOS into what it is today. Going to x86 was what attracted a lot of PC devs such as myself to be excited about the Mac ecosystem and jump over. Seeing news of how we can’t run Windows VMs anymore at the moment is a step back.

Will have to see what happens in the next few years.

 

[Alan Wynn]

I guess you have to focus on power envelope, because Apple isn't delivering performance.

How about you get on the record with answers to these questions:

1. What set of benchmarks will you consider as the basis for comparison between the released Apple Silicon Mac systems and competitive Intel/AMD machines?
2. How will you pick the comparable Intel/AMD system? Machines at the same price point? Machines with the same max TDP? Something else?
3. What would you define as success for Apple Silicon (10% faster? 25%? 10% better battery life? 25%? Something else?)
4. When did you purchase your most recent Mac?
5. What would be required for you to purchase an Apple Silicon-based system?

Easy to sit on the sidelines and throw rocks. Nugget, IPProng, Cmaier, DJJeff, and Gerdi, same questions to all of you.

For me:

1. Latest version of Geek Bench.
2. Same max TDP
3. 10% better performance
4. January 2020 for a Mac Pro, January 2019 for a MacMini.
5. For a Mac Pro replacement:
   1. Native DaVinci Resolve, Blackmagic Fusion, Affinity Suite, Blender
   2. Bonus: Native Creative Cloud
   3. 25% performance boost, or 25% power drop (or a combination), and/or a good enough trade in price.
6. For a Mac Mini:
   1. 20% better GPU
   2. Native Affinity Suite, MS Office
7. For a Laptop:
   1. Support for at least 64GB RAM
   2. Native Affinity Suite, MS Office
8. For an AppleTV:
   1. Release it. (about to purchase some new 4K sets and would prefer to get the next generation AppleTV).

 

[cmaier]

How about you get on the record with answers to these questions:

1. What set of benchmarks will you consider as the basis for comparison between the released Apple Silicon Mac systems and competitive Intel/AMD machines?
2. How will you pick the comparable Intel/AMD system? Machines at the same price point? Machines with the same max TDP? Something else?
3. What would you define as success for Apple Silicon (10% faster? 25%? 10% better battery life? 25%? Something else?)
4. When did you purchase your most recent Mac?
5. What would be required for you to purchase an Apple Silicon-based system?

Easy to sit on the sidelines and throw rocks. Nugget, IPProng, Cmaier, DJJeff, and Gerdi, same questions to all of you.

Well, since you called me out:

1) don’t care about benchmarks. I stopped caring about benchmarks when I stopped designing CPUs. My dissertation is full of benchmarks, but these are the same benchmarks I used in industry, but they are not benchmarks that forum posters know about (spice, anyone?). They’re swell and all, but they only mean something if you use a lot of them, or if they are representative of what you’re doing. If people who buy Arm macs feel like they don’t miss Intel, I consider that a success. That said, I figure Apple will win on geekbench, superpi, cinebench, or whatever. (CPU designers don’t use these).

2) I think I’d compare based on market position. Apple may well release a base 13” Arm MBP that has a lower TDP than the old one, and costs $50 less. I’d compare that to the latest 13” Intel MBP. If you need to compare across manufacturers, I guess you have to do a little untangling, but comparing a dell machine to an apple machine is always a little fraught. If you need to do that, it’s always hard.

3) If they match contemporaneous Intel speeds with 20% more battery life, or they are 20% faster with the same battery life, I would consider that more than a win. Honestly, even matching intel in both categories would really be success, given that this is their first attempt and everyone seems to think they can’t. I expect they will FAR exceed thes specifications.

4. 2018

5. I will purchase one regardless, as a development tool and to tinker with. As long as it has a scissor keyboard, of course! (I will purchase the 16“ MBP variety)

 

[pankajdoharey]

If the performance is more than 30% greater compared to an equivalent Intel CPU at the time of release, there is no way Apple will lower selling prices. If it's 100%, look out.

Could be I am no expert at Apple pricing or pricing in general, I am developer and my reasons are mostly frugality and honesty. I mean how much of a top-up should a business really pocket? If if were a business 100% is probably more than enough, 400% marginal costs to profit seems a bit too much but hey its apple and you could be right.

 

[cmaier]

1. For an AppleTV:
   1. Release it. (about to purchase some new 4K sets and would prefer to get the next generation AppleTV).

Around 15 years ago we did an extensive remodel of our house, gutting almost the entire downstairs, removing walls, etc. As part of that I convinced my wife to let me install a home theater system in the family room, with a projector on the ceiling, a screen that comes down automatically, surround speakers in the ceiling, an 1080p set behind the screen for everyday use, and one remote to control everything. It helped that she didn’t know what it cost.

Anyway, in late february, just before the lockdowns, the 1080p set got fried. The company that installed all my stuff just went out of business, the remote I use no longer allows consumers to have the programming software, and all my cables, the amp, the HDMI splitter, etc. all were so old they couldn’t handle 4k. Took awhile, but I found a great guy to come replace all my stuff, route new HDMI cables through the walls (that took a full day because the old cables were routed when the walls were down, and the dummies stapled them to the beams).

I removed stuff I no longer use (ps3, a HD-DVD player, etc.), bought a 4k Apple TV, and got a 2019 LG OLED screen. Needed a special amp with the ability to handle one 1080p output and one 4k output. I kept the old HD Apple TV for use when I use the projector, because the 4k gets a little funky when the monitor switches resolutions.

Anyway, long story short, the 4k Apple TV works really well. It took me a week or two of messing with LG settings (LG *really* likes to put you in soap opera mode) and Apple TV settings, but it’s all very smooth now.

I’m just happy it didn’t happen *during* the shelter in place, because my family has been watching a lot more tv the last few months...

 

[Alan Wynn]

Could be I am no expert at Apple pricing or pricing in general, I am developer and my reasons are mostly frugality and honesty. I mean how much of a top-up should a business really pocket? If if were a business 100% is probably more than enough, 400% marginal costs to profit seems a bit too much but hey its apple and you could be right.

Apple’s profit margins have remained remarkably consistent over the years for the last 8 years, its net margin has been between 20% and 27.13% (for the last five years it has been between 20.73% and 22.87%). This idea that Apple’s margins are so high is completely false.

 

[Gerdi]

How about you get on the record with answers to these questions:

1. What set of benchmarks will you consider as the basis for comparison between the released Apple Silicon Mac systems and competitive Intel/AMD machines?
2. How will you pick the comparable Intel/AMD system? Machines at the same price point? Machines with the same max TDP? Something else?
3. What would you define as success for Apple Silicon (10% faster? 25%? 10% better battery life? 25%? Something else?)
4. When did you purchase your most recent Mac?
5. What would be required for you to purchase an Apple Silicon-based system?

Easy to sit on the sidelines and throw rocks. Nugget, IPProng, Cmaier, DJJeff, and Gerdi, same questions to all of you.

For me:

1. Latest version of Geek Bench.
2. Same max TDP
3. 10% better performance
4. January 2020 for a Mac Pro, January 2019 for a MacMini.
5. For a Mac Pro replacement:
   1. Native DaVinci Resolve, Blackmagic Fusion, Affinity Suite, Blender
   2. Bonus: Native Creative Cloud
   3. 25% performance boost, or 25% power drop (or a combination), and/or a good enough trade in price.
6. For a Mac Mini:
   1. 20% better GPU
   2. Native Affinity Suite, MS Office
7. For a Laptop:
   1. Support for at least 64GB RAM
   2. Native Affinity Suite, MS Office
8. For an AppleTV:
   1. Release it. (about to purchase some new 4K sets and would prefer to get the next generation AppleTV).

1) Geekbench5 or SPEC2017 - both are pure CPU benchmarks - both do not contain special code pathes for a particular CPU architecture - and in the end both more or less agree when it comes to relative performance.

2) Same TDP - and we would have to consider multi-core benchmarks as single core does not use the available TDP

3) My expectation is 25% faster at same power. I could imagine it is even more.

4) I have 2019 Macbook Pro i7/8 core (I also have a Surface Pro X)

5) I would like to have Windows Virtualization or Bootcamp (native ARM Windows of course)

 

[Woochoo]

Well, I don't believe in magic. Lower power consumption = less powerful A/G/C/PU, it's general rule for more or less similar tech process. You need power for transistors. Lesser number of transistors - lesser computing power.

But you can actually believe in science and technology. To illustrate this, let's compare actual Nvidia vs AMD GPUs:
- Nvidia RTX 2070: optimized 14nm node called 12 nm FinFET from TSMC. 10.8 billion transistor count. 185W TDP.
- AMD 5700XT: 7nm node from TSMC. 10.3 billion transistor count. 225W TDP.

Both similar performance. So how can it be that Nvidia has a similar performing product, with less power consumption and in a way worse and older node? Well, architecture matters.

Now if you compare an x86 on 10nm that performs bad (sometimes even worse that their refined 14nm) vs a 5nm ARM that will be the A14, you can start to imagine some differences.

 

[ssgbryan]

How about you get on the record with answers to these questions:

1. What set of benchmarks will you consider as the basis for comparison between the released Apple Silicon Mac systems and competitive Intel/AMD machines?
2. How will you pick the comparable Intel/AMD system? Machines at the same price point? Machines with the same max TDP? Something else?
3. What would you define as success for Apple Silicon (10% faster? 25%? 10% better battery life? 25%? Something else?)
4. When did you purchase your most recent Mac?
5. What would be required for you to purchase an Apple Silicon-based system?

Since you were quoting me.....

1. The highest multi-core Geekbench 5 score for a A12z is 4615, or about 1/3 of a Ryzen 7 3900x. The 3000 series will be replaced by the 4000 series by the end of the year. As of right now, we are looking at a 15% or so performance uplift over the 3000 series chips.

Cinebench 20 is a good all round benchmark for what I use my machine for (3d Art.) Blender render engine tests are also good, since one of my primary render engines is a fork of the Cycles render engine. My workflow is built around cores & ram.

2. Performance drivers for a new system
A. Is all of my mission critical software available for this CPU? Poser, Daz Studio, Blender, ZBrush, Adobe Acrobat (not the reader), Libre Office, Project Libre, Hexagon, VLC, Mozilla, Steam, Battle.net, MSI Afterburner, etc......
(It will be no for a minimum of 36 months - Poser 12 is about to drop, and I don't see Daz Studio coming to ARM - the iRay render engine is CUDA based. Poser 12 will add GPU rendering with Nvidia cards - important for me.).
B. How many cores does the CPU have? (See 1.) 16 is the new minimum for me.
C. How much ECC ram can the motherboard support? Right now 128Gb is my sweet spot.
D. How many M.2 drives can I install? Are both of those M.2s on the same PCIe 4.0 lanes? (RAID 0).
E. Is there an upgrade path for the CPU? I tend to do a mid-life cycle upgrade. (The AM4 socket lasted for 4 years, and 3 CPU cycles - Zen, Zen+, Zen2, and Zen 3.). Zen 5 will have 4 way SMT. I could use a 16 core/64 thread system, since my main driver is tile based rendering.
F. Is there an upgrade path for the GPU (AMD or Nvidia - the AMD ProRender Engine will use either[or both at the same time]? A LOT of stuff is coming down the pike in the next 24 months.
G. Is there a way to run Windows? As the base OS or in a virtual box - either will do - I have a lot of poorly coded Adobe apps.
H. How many PCIe 4.0 (or 5.0 for 2021) slots available? If the motherboard doesn't come with 10Gb network connection, I want the ability to add it. I also have an e-Sata card (connects to my backup).
I. How many USB connectors? - I have a scanner, as well as other peripherals.
J. How many monitors can I connect? I have been running a dual monitor setup for over a decade, and I am ready to add a 3rd monitor.
K. How many hard drive bays are available? I need 1 for my blu-ray player, and at least 4 more for my data drives - I have a lot of data - iTunes alone is nearing 10Tb. Yes, I do actually need that - back in the days of 10.2, P.T. Barnum was pushing macs as The hub of your digital lifestyle. I (and a lot of other people) went all-in on that. I have another 10Tb in 3d art assets. Needless to say, the cloud simply isn't an option.

Notice that TDP isn't on that list - although the 3950x only hits 105 watts when boosting.

3. Only Apple can define "success" for their products. I would define is as Apple selling enough jumped up iPads to develop an 8,1 Mac Pro. (Which personally, I don't think will happen.)

4. 9 months ago, I picked up another 4,1 for backup (my current system was starting to have hardware issues). When the 7,1 was announced, I picked up an HP Z210 to get myself up to speed on Windows 10. Now I am on a Ryzen/Win10 box. 16 cores/32 threads, 128Gb of ram.

I passed on the trashcan because it was a TCO fail - the 7,1 is even more of a TCO fail than the trashcan (that is before WWDC). None of the other Macs can do much in the way of 3d art (Ask me how I know......)

5. It must outperform what is available from AMD at near price parity - and by near, I mean within 20%.

With all of this, the odds of me slinking back to Apple isn't in the cards - I am NOT replacing all of my software, as well as my hardware, so I can be an unpaid beta tester. Version 1 hardware with Version 1 software - what could possibly go wrong?...........

 

[cmaier]

Well, I don't believe in magic. Lower power consumption = less powerful A/G/C/PU, it's general rule for more or less similar tech process. You need power for transistors. Lesser number of transistors - lesser computing power.

Not true.

I have a transistor. It isn’t busy doing anything. In one chip I leave it source-drain connected to power rails. In another I shut off the VDD power rail so that the transistor has no VDS.

Machines are equally ”powerful.” But one burns more power than the other.

Another example. I design two machines otherwise identical. One uses dynamic circuits for its ALUs. The other uses static circuits. The static circuits include many more transistors than the dynamic. But their functionality is equal. And the dynamic circuits perform the function faster.

Another example. I add an additional pipe stage to increase clock rate. That requires extra transistors. I also increase the complexity of the branch prediction unit to compensate for the fact that because I have extra pipe stages, an incorrectly-predicted conditional branch instruction has a larger penalty.

Which burns more power? The one with the extra pipe stages. Which is more powerful? Not clear.

This is all physics and engineering, not math.

 

[Alan Wynn]

Since you were quoting me.....

Sorry, you seem not to have understood the point of my first three questions so I will clarify it for you. You have repeatedly argued that Apple Silicon will not be competitive with AMD/Intel hardware. One of the biggest issues in these discussion is the constantly shifting goal posts. As such, I am trying to get a set concrete definitions for how to compare these new systems. It is not clear, but I think you are saying that when we compare these new systems upon release, you feel that Geekbench 5 and Cinebench 20 (or the current versions of those two packages?) would be fair.

The second question is more clearly defined as:

2. When doing our comparisons between Apple Silicon-based hardware and AMD/Intel based hardware, how will you pick the AMD/Intel chip to compare? What objective metric would you use to define equivalent systems for comparison?

The third question is more clearly defined as:

3. What objective criteria would Apple Silicon have to meet to be a successful product vs. Intel/AMD’s chips? (While I was not asking at the time, I will add once Apple starts to deliver high-end GPUs, I would be curious as to your answers on those same metrics in that space.)

The fourth question is more clearly defined as:

4. When was your most recent purchase of a Macintosh system from Apple or a third party reseller that was currently shipping at the time you purchased it?

3. Only Apple can define "success" for their products. I would define is as Apple selling enough jumped up iPads to develop an 8,1 Mac Pro. (Which personally, I don't think will happen.)

While as I mentioned, this was not the question I was asking, I just want to confirm that you do not believe that Apple will release an Apple Silicon-based MacPro. Is that correct?

4. 9 months ago, I picked up another 4,1 for backup (my current system was starting to have hardware issues).

So you have not purchased a new machine from Apple for over 10 years, (or was there a more recent Mac that was a currently shipping system when you bought it)?

5. It must outperform what is available from AMD at near price parity - and by near, I mean within 20%.

Got it.