iMac 27" ARM - Will it have different speed processor options like Intel (i5 vs i9)

[NT1440]

Was there a video where he said that?

Yea, the Platform State of the Union (I think it’s available online now). The main presenter facetime’s the head chip designer. That section is only a couple minutes but the language was clear.

I’ll see if I can find what time that section starts at.

8 minutes in to the PSOTU is where it starts.

It looks like the video isn’t on YouTube, but if you download the Developer app on your phone you can watch it there.

 

[Spectrum]

Apple make a lot of money charging for upgrades. I'd be amazed if they only offer a single model. It will really put off some customers that want to have the "best". The only potential benefit (to Apple, not the consumer), Is that a single CPU offering might lead people that want the best to upgrade more frequently...

But for a second reason: yields. It is inevitable that the CPUs will Have different quality yields, and thus will be binned, and then sold with different performance. I definitely expect to see variable core counts and frequencies.

GPUs are an unknown at present. I find it hard to believe that an Apple GPU design can match the top offerings from AMD or NVIDIA.

 

[verticalines]

I think they'd standardize a certain core count per machine tier. Say the Pros all have 16 cores, the middle 8, and the lower 4. Plus accelerators. Core speed is the same across the board determined by demand and cooling capacity. With tiers they can also upgrade each without disrupting the basic segmentation.

Clockspeed you see is mostly a binning thing. There isn't a specific reason why a i5 is slower than i7 in most cases, it's just they may shift the lower-binned stuff downwards and keeping the premium at higher. Goes for clockspeed and power efficiency.

 

[Krevnik]

If the ARM Macs are all integrated GPU's then It's up to Apple to prove to us how they are supposed to be able to compete with AMD and NVIDIA. I don't think any i-GPU can compete with that. The technology is not there yet.

GPUs are an unknown at present. I find it hard to believe that an Apple GPU design can match the top offerings from AMD or NVIDIA.

Honestly, the big hurdle is that an SoC will effectively always have less die space to devote to GPU CUs than a dedicated GPU. Which means you have to either hit much better perf/mm^2 than a dGPU, or just build a dGPU yourself.

If they dedicate the resources there is no reason they can’t have the gpu performance of an AMD.

If I was Apple, I’d be focusing on making sure the CPU/iGPU side is well handled, and use AMD GPUs where needed like they do today. Worry about building dGPU replacements later, since you’ll have time later to go tackle that if you really want to.

If the first round of SoCs can at least make Intel iGPUs look bad, that’s not a bad place to be looking forward 5 years, really.

 

[windowsblowsass]

Honestly, the big hurdle is that an SoC will effectively always have less die space to devote to GPU CUs than a dedicated GPU. Which means you have to either hit much better perf/mm^2 than a dGPU, or just build a dGPU yourself.



If I was Apple, I’d be focusing on making sure the CPU/iGPU side is well handled, and use AMD GPUs where needed like they do today. Worry about building dGPU replacements later, since you’ll have time later to go tackle that if you really want to.

If the first round of SoCs can at least make Intel iGPUs look bad, that’s not a bad place to be looking forward 5 years, really.

An as is a12z already bears the hell out of intel igpus

 

[KPOM]

It's the battery powered mobile computers that will benefit the easiest from the the performance-per-watt advantage of Apple Silicon. Note that Johny Srouji made performance-per-watt the key point in his segment during Monday's keynote; he mentioned it in the first minute.

Apple's Mac sales are >85% notebooks. Those are the units that will see Apple Silicon the first, probably starting with the MacBook Air or possibly an Apple Silicon revival of the discontinued Retina MacBook.

I’m guessing the 13” Pro might receive Apple Silicon first, since it has gone the longest without a redesign (other than retrofitting for the “Magic Keyboard”). Apple probably won’t want to sell a 13” Pro alongside an Air that outperforms it. That said, I would like to see a 12” MacBook return (but with a Magic Keyboard).

 

[bmot]

This whole ideology of choosing processors types is a Intel thing. Look at iOS/iPadOS. No such thing there and I think Apple will stick to that with the ARM based Macs.

 

[seangrimes590]

This whole ideology of choosing processors types is a Intel thing. Look at iOS/iPadOS. No such thing there and I think Apple will stick to that with the ARM based Macs.

Apple allowed you to choose processor speed back in the power pc days. No good reason not to do the same now, it improves their yields, their profit margin on the top spec chips, and it makes consumers happy because they feel better about the +$1800 spec machine with extra RAM and storage because they can also get a CPU that is a few hundred MHz faster than the "regulars".

 

[jdb8167]

There is no way Apple can rely solely on iGPU built into the SoC. Just think about the Mac Pro that can now be equipped with 4 powerful AMD GPUs. They need to be able to provide this kind of GPU power by the end of the 2-years transition period and I don't see them doing that by themselves so soon.

I thought they were fairly clear in the Apple Silicon sessions that they are going to use a unified memory model for the Apple Silicon SoCs. That implies that the first round of chips will not have external GPU options. They can always add discrete graphics at a later date but the way they talked, it seemed like it was going to be integrated GPUs for the first chips.

 

[Danny82]

I really hope not.. I would like it if they do it the same way as rumoured iphone business model way this year like 2 normal macbook which is the air and 2 pro macbook 14 and 16.. macbook air to be 12 and 14 with SOC similiar to ipad as no fan and the pro model to get a laptop specific SOC..

 

[cmaier]

I thought they were fairly clear in the Apple Silicon sessions that they are going to use a unified memory model for the Apple Silicon SoCs. That implies that the first round of chips will not have external GPU options. They can always add discrete graphics at a later date but the way they talked, it seemed like it was going to be integrated GPUs for the first chips.

I think the integrated/discrete distinction won’t work so well for these. In the sessions they said that developers should treat the Apple GPU as “discrete.”. But they also said it will be on the SoC. (Could be they were speaking loosely and it will be in the package but not on the same chip as the CPU, but i doubt it.)

 

[Jorbanead]

I think they will totally have different tiers for processors.

For one, they have stated they are creating a family of SoC's for the Mac. But also, they can easily take the same processor, under clock it, or remove a few cores, and give it a new name. Sort of how i5 and i7 are the same chips but re-binned. I see them making 3 groups of chips (Two groups are the same chips, just rebinned). Here is my theory:

Laptops: All laptops will come with special laptop-designed chips by default. These are designed for passive cooling. The MBP models will come with an option to upgrade to the Desktop-class chips, though the higher-core desktop chips will not be an option. Expect 8 - 12 cores standard. 12 - 18 cores upgrade (MBP only).

Desktops: All non-pro desktops will come with these desktop-class chips. These are designed for active cooling. They are essentially just higher-clocked Laptop-class chips, but with more cores activated. Desktop Macs will have an upgrade option for workstation chips though you will have limited options for core counts (Gotta give buyers an incentive for the Mac Pro). Expect 12 - 18 cores standard. 18 - 24 cores upgrade.

Workstation: The iMac Pro and Mac Pro will use workstation-class chips. These are designed for active cooling and demanding workflows like 8K video editing. iMac and Mac mini will have access to some of these chips too, but they will save the high core counts and some pro features only for the pro desktops. Expect 18 - 48 cores.

I imagine all processors will hav 4 little cores. So an 8-core processor is 4 big + 4 little. A 48-core processor is 44 big + 4 little. Etc.

 

[Boil]

I think they will totally have different tiers for processors.

For one, they have stated they are creating a family of SoC's for the Mac. But also, they can easily take the same processor, under clock it, or remove a few cores, and give it a new name. Sort of how i5 and i7 are the same chips but re-binned. I see them making 3 groups of chips (Two groups are the same chips, just rebinned). Here is my theory:

Laptops: All laptops will come with special laptop-designed chips by default. These are designed for passive cooling. The MBP models will come with an option to upgrade to the Desktop-class chips, though the higher-core desktop chips will not be an option. Expect 8 - 12 cores standard. 12 - 18 cores upgrade (MBP only).

Desktops: All non-pro desktops will come with these desktop-class chips. These are designed for active cooling. They are essentially just higher-clocked Laptop-class chips, but with more cores activated. Desktop Macs will have an upgrade option for workstation chips though you will have limited options for core counts (Gotta give buyers an incentive for the Mac Pro). Expect 12 - 18 cores standard. 18 - 24 cores upgrade.

Workstation: The iMac Pro and Mac Pro will use workstation-class chips. These are designed for active cooling and demanding workflows like 8K video editing. iMac and Mac mini will have access to some of these chips too, but they will save the high core counts and some pro features only for the pro desktops. Expect 18 - 48 cores.

I imagine all processors will hav 4 little cores. So an 8-core processor is 4 big + 4 little. A 48-core processor is 44 big + 4 little. Etc.

Great post! My thoughts exactly, not everything will be based on an iPad Pro APU. There will be larger / more complex APUs for use in desktop / workstation Macs. My ideas towards workstation APUs:

32 P cores / 4 E cores / 48 GPU cores / 16GB HBM2e UMA

48 P cores / 6 E cores / 64 GPU cores / 24GB HBM2e UMA

64 P cores / 8 E cores / 80 GPU cores / 32GB HBM2e UMA

I give high P core counts because these will not have SMT.

More RAM (DDR5) & NVMe storage on logic board.

Mac Pro line-up will be threefold:

Mac Pro Cube / iMac Pro / Mac (Big Chungus) Pro

The Big Chungus Mac Pro will have PCIe slots for assorted usage (mainly A/V DSPs & I/O), Apple will make compute GPUs for more rendering power. These compute GPUs will be two or four APUs with low P & E cores (enough to control things) & a high GPU core count. Think of a compute GPU with up 320 GPU cores! No video output, that is handled by the main workstation APU.

APUzilla...!!!

 

[Waragainstsleep]

I believe their current fastest GPU in the iPad Pro is roughly around the level of a lower mid range PC card (e.g. GTX 750) from 2013 or an Xbox One/One S from that same era (impressive for a mobile device but that is an eternity ago tech wise and performance wise compared to the latest and greatest GPUs).

I think you can make a pretty big jump already by applying some similar logic as we do the state of their CPU performance. That A12Z which is equivalent to an Xbox One has 8 GPU cores and 6GB of LPDDR4X RAM. In a desktop, you can immediately ditch the low power part of this RAM, up the amount available and I'm guessing multiply the speed of it by quite a lot. My 16" MacBook Pro is running 2667MHz which is a factor of 10 over LPDDR4. That sounds like quite a boost to me.
I don't know how to compare the GPU cores in an A12 to the crazy numbers of specialised cores in a modern GPU so the best I can do is some very dirty extrapolation.
The GTX750 has 512 Cuda cores, Power = 55W
The A12Z has 8 cores, Power = 15W
These two are supposedly equivalent at 1.4tflops

I don't know how many GPU cores we can add in an iMac but I reckon its a few. The iMac Pro has a Radeon Pro Vega 64, the desktop variant of which has a TDP of 250W. I'm sure the smaller version in the iMac is probably a lot less so lets halve it to 125W. (The iMac Pro supposedly uses ~430W all in. The top end Xeon in it is 140W, allow another 100W for the rest of the unit and you still have ~200W left for the GPU so 125W seems potentially conservative, at worst its reasonable)
The A12Z cores look to be around ~1W each, assuming the active cooling can compensate for any non-linear scaling in watts per core, We could add over 100 GPU cores to an iMac SoC. Lets call it 80 for simplicity. Thats 10 times what the A12Z has and with 10x faster RAM. Potentially it could also have 10x more RAM for all I know.
That gives us 14tflops which would put it equivalent to an RTX 2080ti.

Like I say, this is a very dirty calculation with some pretty optimistic assumptions regarding core scaling, TDP, cooling etc. It doesn't include any improvements in the actual cores or a clock boost for them. It also doesn't factor in any equivalent Apple might have for the RT or Tensor cores that Nvidia uses or anything else they might have up their sleeves.

Even if I'm out by 20% or so, 11tflops is the regular RTX 2080. At any rate, its not a preposterously massive stretch that Apple can get some good graphics performance from their new silicon in reasonably short order.

 

[JacobHarvey]

I think you can make a pretty big jump already by applying some similar logic as we do the state of their CPU performance. That A12Z which is equivalent to an Xbox One has 8 GPU cores and 6GB of LPDDR4X RAM. In a desktop, you can immediately ditch the low power part of this RAM, up the amount available and I'm guessing multiply the speed of it by quite a lot. My 16" MacBook Pro is running 2667MHz which is a factor of 10 over LPDDR4. That sounds like quite a boost to me.
I don't know how to compare the GPU cores in an A12 to the crazy numbers of specialised cores in a modern GPU so the best I can do is some very dirty extrapolation.
The GTX750 has 512 Cuda cores, Power = 55W
The A12Z has 8 cores, Power = 15W
These two are supposedly equivalent at 1.4tflops

I don't know how many GPU cores we can add in an iMac but I reckon its a few. The iMac Pro has a Radeon Pro Vega 64, the desktop variant of which has a TDP of 250W. I'm sure the smaller version in the iMac is probably a lot less so lets halve it to 125W. (The iMac Pro supposedly uses ~430W all in. The top end Xeon in it is 140W, allow another 100W for the rest of the unit and you still have ~200W left for the GPU so 125W seems potentially conservative, at worst its reasonable)
The A12Z cores look to be around ~1W each, assuming the active cooling can compensate for any non-linear scaling in watts per core, We could add over 100 GPU cores to an iMac SoC. Lets call it 80 for simplicity. Thats 10 times what the A12Z has and with 10x faster RAM. Potentially it could also have 10x more RAM for all I know.
That gives us 14tflops which would put it equivalent to an RTX 2080ti.

Like I say, this is a very dirty calculation with some pretty optimistic assumptions regarding core scaling, TDP, cooling etc. It doesn't include any improvements in the actual cores or a clock boost for them. It also doesn't factor in any equivalent Apple might have for the RT or Tensor cores that Nvidia uses or anything else they might have up their sleeves.

Even if I'm out by 20% or so, 11tflops is the regular RTX 2080. At any rate, its not a preposterously massive stretch that Apple can get some good graphics performance from their new silicon in reasonably short order.

I have no doubt that they'll get very good performance out of their iGPUs. However TFLOPs alone is not a good indicator of the 3D graphical performance of a GPU at all. As a quick example a Vega 64 with 12.66 TFLOPS (FP32) would be expected to smash the RTX 2070 Super (9.06 TFLOPs FP32) if you think that's a good indicator of GPU performance. Instead the Vega 64 gets significantly outperformed in practically every 3D game out there by a large margin while also sucking up way more power. There are so many other factors relating to the microarchitecture of the GPU itself etc. that are important to consider in determining how well it performs in 3D tasks like gaming, modelling etc.

So I think that these rough extrapolations using a single measure of GPU performance (and one that isn't very relevant to real-life 3D performance when looked at alone) are not much better than complete guesswork in supporting the notion that Apple could soon reach RTX 2080-like performance in 3D tasks with its iGPUs. There are also limits to what is a realistic die size to be used in a mass production device like a mac. So it is not as simple as just upping the number of GPU cores until you've got 80 A12Z iGPU cores next to the CPU cores on an iMac as part of a future mac A14/15 (or whatever it'll be called). Those Apple GPU cores are totally different to nVidias 'CUDA cores' and are much larger in terms of die area used

We really have to wait and see actual Apple hardware running similar cross-platform programs, games, and benchmarks before we can get a good idea of what Apple can achieve vs AMD and nVidia and where it is heading with its GPUs. (Is Apple going to primarily focus on very good iGPU performance with best in business perf/watt to focus on macbook/macbook pro users? Or is it also likely to devote resources to meet or beat nVidia and AMD in high end GPUs to target gamers and people who use 3D modelling and design?)

I think that Apple is definitely able to compete with nVidia and AMD given enough time, focus, and resources. But they will have to separate the GPU silicon from the CPU silicon and improve its architecture for any future high end GPU to beat what nVidia can create (which is also constantly innovating).

 

[Waragainstsleep]

I did say it was dirty and there wasn't really any other way to compare a theoretical device to real world ones. Apple has the resources to compete with pretty much anyone they want, its all a question of time really.

 

[YoitsTmac]

A lot of people here are saying there won't be chip variation because of the way Apple handles iPhone and iPad. But, the iPad is completely segmented by chip: the iPad with A10, the iPad Air with A12, and the iPad Pro with A12Z.


And as someone else pointed out, Apple will have different chip options for the profitability alone. Joining the speculation bandwagon, I had a discussion with my friend who asked me where I thought Apple silicon could go. He's not as tech savvy and I immediately began to think. Among low power benefits that Apple has already shown off in iOS (my Mac drops battery fast just sitting for App Nap), I think Apple will segregate product performance benefits deeper than ever, hear me out. Apple has shown they can make chips that are heavily specialized. The chip in the MBP can accelerate encoding and other video features. I suspect each Mac silicon will have its own specialities.

Computer like the MacBook Air will have no bells and whistles. It's the daily driver with great entry graphics: maybe one high-performance core and very well done low-power cores. The high power core is for tasks that need to be completed quickly much like intel powerboost, and then goes back to the low-power cores for watching the movie, scrolling through emails, etc.

The 13" MacBook Pro will have more performance cores or as someone else said, just more performance cores turned on compared to the Air. Again with Apple doing great specialization (T2 chip in Macs, Neural Engine in iPhone), they will add some features to the silicon to target pro users. Perhaps an audio encoder that can optimize multiple audio tracks. The 16" MacBook Pro gets these features amplified. A mini afterburner card that can accelerate graphics in FCPX or other metal tasks alongside an option for a more powerful GPU. It's 3AM and I can't think of other specialized tasks Apple does with their silicon, but I think they'll do things like this, parts of the chips highly specialized, to greater differentiate models in ways beyond cores and clocks.

 

[jdb8167]

Computer like the MacBook Air will have no bells and whistles. It's the daily driver with great entry graphics: maybe one high-performance core and very well done low-power cores. The high power core is for tasks that need to be completed quickly much like intel powerboost, and then goes back to the low-power cores for watching the movie, scrolling through emails, etc.

Remember it doesn’t cost Apple much more to have more high performance cores; except some SoC die space. I doubt they would ever release a single high performance core SoC. It wouldn’t make much sense. Expect the low end to be at least similar to the A12: 4 high performance cores and 4 efficiency cores. It also makes very little sense that I can see to have more than 4 efficiency cores. Anything needing more low power work would be transferred to a high performance core(s) which will finish whatever task is required much faster. This saves more battery than it uses. The efficiency cores will likely be used more for situations that are IO bound with a lot of waiting for IO events. These types of events are very easy to distribute across multiple, low frequency threads.

 

[xgman]

This whole ideology of choosing processors types is a Intel thing. Look at iOS/iPadOS. No such thing there and I think Apple will stick to that with the ARM based Macs.

But there is certainly a performance difference in the various iphone and ipad models. I would expect the same on ARM macs.

 

[Waragainstsleep]

They will have thunderbolt on the new ARM macs. You can’t replace thunderbolt with USB C in all cases. The investment that some Mac uses have made in thunderbolt 3 over the years would absolutely kill part of their pro market.

Apple confirmed TB support for new Apple Silicon Macs.

Computer like the MacBook Air will have no bells and whistles. It's the daily driver with great entry graphics: maybe one high-performance core and very well done low-power cores. The high power core is for tasks that need to be completed quickly much like intel powerboost, and then goes back to the low-power cores for watching the movie, scrolling through emails, etc.

Why would they make a chip with only one high performance core for a MacBook when an iPad already has 4 plus 4 more efficiency cores for running background tasks? Doesn't make any sense.

 

[groove-agent]

Apple already offers different CPUs amongst their phones so there's no reason why they couldn't do that with their Macs. Perhaps the cheaper Macs will have older CPUs and the more expensive Macs will have the bleeding edge ones.

No ones know this but just putting it out there. With an iMac 27 you can choose i5 vs i9. Do you think when the iMac 27" ARM does come out it will have an option for more than one SOC processor type?

My gut tells me gen 1 will have one choice for a processor. Maybe gen 2 or gen 3 will have upgrade options.

Thoughts!

 

[Erehy Dobon]

My guess is that Apple will have two main Apple Silicon product lines: one for notebooks and another for desktop systems primarily for TDP limitations. They will likely differentiate in each category by binning.

Note that while technically a desktop in form factor, the Mac mini is actually a headless/keyboardless notebook computer. Apple has used high-end mobile CPUs in Mac minis for a very long time.

For the notebooks, I expect Apple to consider the same TDP performance envelopes: 15W for very low-end devices, 65W for the high-end Pro line and somewhere around 28W for the mid-range. The latter might be a high-end MBA or a low-end MBP.

That said, I think Apple has given up on the ultrabook low-end. They just want people to buy an iPad Pro instead.

 

[sublunar]

My guess is that Apple will have two main Apple Silicon product lines: one for notebooks and another for desktop systems primarily for TDP limitations. They will likely differentiate in each category by binning.

Note that while technically a desktop in form factor, the Mac mini is actually a headless/keyboardless notebook computer. Apple has used high-end mobile CPUs in Mac minis for a very long time.

For the notebooks, I expect Apple to consider the same TDP performance envelopes: 15W for very low-end devices, 65W for the high-end Pro line and somewhere around 28W for the mid-range. The latter might be a high-end MBA or a low-end MBP.

That said, I think Apple has given up on the ultrabook low-end. They just want people to buy an iPad Pro instead.

I would take your point about the Mac mini in general having mobile CPUs but the 2018 model uses what is designated a desktop cpu by intel (even though it‘s packaged like the mobile CPUs with its BGA format).

Binning may allow Apple to offer different CPU speeds but I would have said number of cores would be more likely to be the performance differentiation that they would go for. They have a number of lower end products that could accept ARM CPUs that fail to meet standards for higher up SKUs.

In terms of benchmarks, we haven’t seen anything that could match the graphics performance from the Macbook Pro 16“ AMD dGPU but there should be no problem matching up performance with an improvement in integrated graphics and potentially battery life for the 13” SKUs.

The real surprise would be how Apple deal with Macs that have discrete graphics provided by AMD. Do they have an in house solution that will fit inside a 50w budget for example?

 

[aeronatis]

In terms of benchmarks, we haven’t seen anything that could match the graphics performance from the Macbook Pro 16“ AMD dGPU but there should be no problem matching up performance with an improvement in integrated graphics and potentially battery life for the 13” SKUs.

The real surprise would be how Apple deal with Macs that have discrete graphics provided by AMD. Do they have an in house solution that will fit inside a 50w budget for example?

I believe MacBook Air will have the regular A14 just like the regular iPhone and iPad Air/Mini as it is the case with the current MacBook Air with less than 10 watt CPU. The only different from iPhone/iPad line-up would be the configurability of the amount of RAM. There is no reason the system memory being anything more than LPDDR5 (which would be fine as the current device already uses LPDDR4 shared memory).

14" and 16" MacBook Pro would have A14X (or whatever it is called) with the rumoured 8P & 4E cores as well as the iPad Pros, still the difference being the amount of RAM (and TDP obviously). This would possibly reduce (if not minimize) the performance gap on 14" and 16" MacBook Pro just like it is the case with iPad Pro 11" and 12.9".

Given Radeon Pro 5500M and 5600M are around the performance level of GTX 1060 Max-Q and RTX 2060 Max-Q respectively, it is really hard to estimate the graphics performance level of the new devices (Maybe less in raw power but similar/better performing due to running an ARM optimised OS and apps). The type of the unified memory they mentioned during the event is a mystery for now.

 

[Joe The Dragon]

I thought they were fairly clear in the Apple Silicon sessions that they are going to use a unified memory model for the Apple Silicon SoCs. That implies that the first round of chips will not have external GPU options. They can always add discrete graphics at a later date but the way they talked, it seemed like it was going to be integrated GPUs for the first chips.

so mac pro / imac pro will poor pci-e lanes and maybe no 10G-E?