MP 7,1 What does Apple's switch to ARM based processors mean for the future of the Mac Pro

[mode11]

The Mac transition to ARM has obviously been in the planning stages for many years. The 2019 Mac Pro was designed in this context, so will almost certainly inherit the case and general platform architecture. It’s possible that design decisions that raised eyebrows for the 2019 MP (e.g. the separate card for USB) will actually make greater sense when the ARM version comes out. Perhaps part of the delay was to coordinate with the changing requirements of an in-development ARM version.

A major part of the ludicrous cost of the 2019 MP is the expensive Xeon CPUs. If using their own silicon lowers the BOM significantly, it’s possible the new MP will be cheaper, whilst still maintaining margins. This could increase the market for the model, making it more worthwhile from Apple’s point of view. Perhaps Apple are even aware the current MP is overpriced, but aren’t too concerned because it was always going to be a temporary situation.

Apple made a huge, public commitment to the Mac Pro tier of the range by releasing the 2019 version. Perhaps for many years, it seemed unlikely they would have been able to make an ARM chip that could compete at this end of the market, so had decided to just let the Mac Pro die off with the intended ARM transition. At some point in the last few years, though, they must have realised it was practical after all. The fact they have announced a transition of the complete line up within two years indicates this is a completely solved problem - and have a compelling roadmap for years to come. It is inconceivable they would announce the transition, then worry about how they were going to replace the Mac Pro later.

 

[Moonjumper]

My first Mac was nearly a cheese grater Mac Pro in 2006, but got a top spec iMac instead. It was a great machine, but was lacking ergonomically due to no way to lower the screen, plus it was not possible to replace parts quickly and keep working (how I wished I had a Mac Pro when it spent a month being repaired at the Apple Store), and lacked the flexibility of upgrading parts later. I am on a 2015 iMac now as the trashcan wasn't the right machine for me either. The current Mac Pro is a fine machine, but the entry price moving so much higher means it is not the machine for me. I have said this to make it clear my position is not the same as current owners.

I think things will change with Apple Silicon, but hopefully it will serve all existing users, but hopefully also all those left behind. The Xeon boundary disappears when Apple design their own chips, so hopefully the Mac Pro will be powerful enough to serve existing users, but also be more flexible in scaling down to lower price points. Having a larger market would make it more worthwhile for Apple to invest in keeping it relevant, to the benefit of all.

Many are looking at the mobile chips for an expectation of where Mac performance will be, but the Arm server chips should be a more useful guide, especially for the Mac Pro. A lot more cores are likely.

Integrated or discrete graphics have been a talking point for years, and Apple only talking about SoC is a big potential change for the Mac Pro, possibly the most important. But how much of the difference is driven by PC vendors and market conditions? Apple are free of those constraints. They can answer the questions in different ways. Is there a technical reason why separate is better? Separating the heat sources is good, but having to connect CPU to GPU is detrimental.

On current Apple Arm chips, the CPU core counts are in single figures, and so are the GPU counts. I expect the Mac Pro will see CPU core counts a long way into double figures, and GPU maybe even in triple figures. With the way Apple GPU cores work, I'm not expecting anything gimped.

 

[ticctacc]

I think those who have a Mac Pro bought it cause they needed a new Mac. That is what I did with my old iMac5k late 2014 was not cutting it and was going to black while using final cut.(does any one know why I heard it's the PSU?)
I just finished my album https://distrokid.com/hyperfollow/thomaspiper/permission-to-live and needed a powerful Mac for the videos I have 3 4k cameras and needed something that could run Multicam video no proxies.
I Also wanted to use some 3d software using boot camp. and different video cards.
I said all that to say if you need a powerful Mac and your current model doesn't make Then buy a Mac Pro if you want expansion. If you are going to wait two years you don't need a computer.
I feel a lot of people who said they where going to buy a Mac Pro but then changed their mind after the announcement (which we all knew was coming) didn't really need one and ARM gave them an excuse not to buy one.

The real question is will apple even make a 8.1 Mac Pro. will Mac even support GPU cards. external or internal. I also feel with no ARM boot camp the 7.1 Mac Pro will be sought after. but who knows. I have a feeling lot of people will be holding on to the 7.1 to run apps that don't make the switch. The reality is a iMac Pro is still very powerful for most people. Alot like me bought the 7.1 Mac Pro just for the PCI slots (also for me the thunderbolt 3 ports cause my previous desktop did not have them). I Think it will have a long life span 5 years is all I need and I don't usually sell my desktops just my laptops. I think you will get great deals on used ones too.

 

[ssgbryan]

You think ARM is going to be a LOT less money? lol. I have a great deal on a unicorn I can sell you.

Thanks for the chuckle.

You misunderstood. By dumping OSX, I got more (a whole hell of a lot more) performance, for a lot less money.

Apple isn't going to lower prices - the true believers will pay whatever Timmy wants them to pay.

 

[mode11]

I’d be surprised if Apple don’t make an 8,1, and not just because they just introduced the 7,1 after much delay and fanfare. To rely exclusively on SoCs is not practical - high end GPUs are massive chips, consuming hundreds of watts. And unlike Intel CPUs, they are already built using cutting edge TSMC fabrication, so they don’t get an easy win there. It would be crazy to stuff one into an SoC with lots of CPU cores as well. Even if they did, it would be so far from an iPhone SoC that there would be little benefit to Apple anyway.

It’s also not realistic to expect Apple to design discrete GPUs that can keep up with the best AMD and Nvidia have to offer, and to keep doing so year on year. Especially for a machine that represents only 1% of the sales of the company, that itself has 5% or so of the PC market. They could certainly make great SoCs for MBPs and iMacs, but that’s very different from e.g. an MPX module with dual Radeon 7s.

 

[StuAff]

The notion that Apple's going to junk discrete GPU support (internal, let alone external) in favour of just that on an SoC is plain nuts. One slide on a WWDC presentation is not evidence to the contrary. No, it doesn't mention AMD as being an option for Apple Sillicon systems, but they haven't suggested what they're going to use on actual for-sale Macs either, apart from the SoC GPU, which is the clear replacement for Intel's integrated graphics. Not a replacement for a couple of Pro Vega II Duos…Apple's not thinking it is, BTW. The SoC is a baseline spec for developers to work with, the future AMD (or Intel, or…) GPU will use that as a starting point.

 

[deconstruct60]

It took 7 months in 2006 (from the MacBook Pro release on January 10 to the release of the Mac Pro on August 7).

with reportedly mentioned on this forum largely outsourcing much of the Mac Pro work out to Intel ( in part why rev 1 of Mac pro was more "drop ATX board into container" solution) . The 12 months was a 'hard' dealine that Jobs threw out there and knew up front that was throwing work outside to get it done. Speed was an issue in being competitive.

Intel was already making a full line up of chips and more importantly a wider set of customers to sell that broader mix to. ( e.g. more than several 100K worth of workstation chips would be sold by Intel whether Mac Pro 2006 shipped or not. ). Apple is still growing theirs, so the timeline isn't going to be the same at all. Nor should the expectation be that it would be the same. Different circumstances probably will lead to different outcomes.

Speed applied to doing a Mac Pro class processor could be a major problem if Apple screws that up by going too fast (and end up with bugs in silicon. ) Managing the expense of the package is going to matter because there is exceeding little volume to spread the costs over. ( it is a dual edge sword that Apple is the only buyer of the chip. At high volumes less so. At super low volumes more so. )

3-4 years is time lots of processors take to get out the door. Revisions appear closer to a 1 year rate when there are multiple designs being pipelined and run concurrently . That way a processor can get retired out of the design pipeline at each year 'clock' while keeping multiple in flight.

The Apple silicon team is much larger than the Industrial design team but at some rate of product flow they'll likely become a choke point also. It won't be surprising if the new Mac processors are rolled out over the range at extended intervals (while the ones used in more high volume products maintain a 3-4 deep pipleline so hit every year )

 

[deconstruct60]

I hope they’ll just release an upgrade board/option for 7,1 users. I don’t think they designed they case just for one revision.

No chance of that. The architecture will be too fundamentally different.

the internal architecture of the CPU package doesn't necessarily have a huge influence on the logic board design.
If Apple went to a whatever Intel ( or AMD) had current in 2022 it would be just as new logic board also. What is likely to happen is that the logic board changes. That could easily change the mount screw points. ( in part because Apple really doesn't care about selling raw motherboards in the slightest. )

that doesn't require a radical case design , but it would need some minor tweaks. Which is largely exactly just what was done in the transition from 2007 to 2008 to 2009 (looks mostly the same on outside , not the same on the inside. ).

With PCI-e v4 (and higher) it isn't really the best design to move the highest bandwidth slots the farthest away from the CPU package as possible. So perhaps keep slot 8 where it is now (just to perhaps keep those USB slots closer to the likely desktop surface above where the Mac pro is placed. ) but move what is now slots 6-7 to the other side of MPX bays.

There is also a I/O board embeeded in the frame to provision (top / front) slots . those boards too would have to come out also. So selling a two board combo set is even less likely.

There were 68K to PPC upgrade kits (from the last of the latter to the first of the former) because the PowerPC 601 processor Macs, in both hardware and software, were designed with backwards compatibility in mind- they had to be otherwise the OS wouldn't have worked. That, and Apple's dire financial straits at the time.

In dire straits, in part because were overly focused on maximizing box with slots line up. Not happening this time.
Apple I was sold as an unassembled kit at one point. That isn't really Apple anymore either (never was where Apple wanted to be) . When have to jump back into the last century for last time Apple did something like this , then that is more a trip in the 'way back machine with Sherman' than accurate predictors of the future.

 

[Moonjumper]

I’d be surprised if Apple don’t make an 8,1, and not just because they just introduced the 7,1 after much delay and fanfare. To rely exclusively on SoCs is not practical - high end GPUs are massive chips, consuming hundreds of watts. And unlike Intel CPUs, they are already built using cutting edge TSMC fabrication, so they don’t get an easy win there. It would be crazy to stuff one into an SoC with lots of CPU cores as well. Even if they did, it would be so far from an iPhone SoC that there would be little benefit to Apple anyway.

It’s also not realistic to expect Apple to design discrete GPUs that can keep up with the best AMD and Nvidia have to offer, and to keep doing so year on year. Especially for a machine that represents only 1% of the sales of the company, that itself has 5% or so of the PC market. They could certainly make great SoCs for MBPs and iMacs, but that’s very different from e.g. an MPX module with dual Radeon 7s.

There are Intel processors also in the hundreds of watts. Apple intend to compete there, but taking an alternative approach that doesn't require that power draw.

The same is just as likely to apply to GPUs. Tile-Based Deferred Rendering and Metal are two of the ways Apple Silicon are different on the graphics front.

 

[deconstruct60]

PCIe already goes directly to the CPU. New Intel chips will have the Thunderbolt logic built directly into the CPU.

On laptop chips. On Desktop chips they do not.

Exclusive: Intel Rocket Lake-S features PCI-Express 4.0, Xe Graphics - VideoCardz.com

Intel Rocket Lake-S might provide the biggest upgrade to the mainstream desktop platform in years. Intel Rocket Lake-S gets PCIe 4.0 The platform codenamed Rocket Lake-S is currently expected in late 2020. The rumors about the RLK-S platform have been floating around the web for a while now, but...

videocardz.com

Even with a iGPU they don't here because pushing the 10nm design back to 14nm soaks up lots of floorplan space. So THunderbolt is discrete (see lower right of diagram above).

In completely iGPU excised CPU packages, Intel has given ZERO indication that they are putting Thunderbolt on the die. None. That's largely because there is no good motivation there at all. Zero. That means running running what are effectively unidirectional pin connections in just so can run them back out of the CPU again as opposed to sending them closer to where they need to eventually need to go in the first place. Unde misdirection of display output is just Rube Goldberg exercise. It isn't good CPU packaging.

I'm aware AMD isn't doing it for their APU's but that doesn't mean Apple won't, they aren't AMD.

You are the one engaging in "monkey see , monkey do " design copy of AMD. Not me. It makes almost no sense to chop are relatively small-medium die into smaller dies if getting anywhere near decent yields out of the fab process. one mask covers both "half" and can just crank those out quicker with fewer mask shifts on the fab hardware. It would likely cost more.

For the medium-to "max possible" range chiplets can makes sense but you can't put medium to max dies in a laptop form factor. Laptops are logic board space constrained so quite likely will be limited to sub 8-16 cores at given range fab process ( 7nm -> 5nm ) . A big shrink will get more cores, not chiplets. Chiplets actually take up more space. You get cheaper dies but there is a design cost trade off ( it isn't 'free' ).

As soon as Intel has something as fast as an older AMD mobile they dumped that MCM Frankenstein project they had with an AMD die . Intel may do a 10nm GPU + 14nm CPU mash up for Rocket Lake but that mainly due to the fab screw ups more so than whether that is a more energy efficient approach. It is likely another Frankenstein.

Also in conversations like this I wouldn't really say things like "Err no" unless you have facts. It's all just opinion until they ship product, definitive statements can't be made.

Facts are that chiplet solutions are a bigger footprint. Facts are that laptops don't deal well with bigger footprint solutions. Nor do they deal well with higher power consuming solutions. that because physics is physics. It has nothing to do with instruction sets and approaches that IBM took with mainframe dozen years ago or AMD is taking with their desktop solutions now. In those contexts those constraints don't matter as much as they do in laptops.

If have a laptop and have deep need for much bigger GPU die then making it a discete package helps distribute the thermal load out. There could be a slippery slope there where mounting it in the same package with HBMv2 memory could help be a "one shot" leg up on assembly but that is getting out of the "chiplet" space.

 

[mode11]

There are Intel processors also in the hundreds of watts. Apple intend to compete there, but taking an alternative approach that doesn't require that power draw.

The same is just as likely to apply to GPUs. Tile-Based Deferred Rendering and Metal are two of the ways Apple Silicon are different on the graphics front.

Yes, but Intel are stuck on 14nm whereas GPUs are either already on 7nm or soon will be. So Apple won‘t get the easy advantage of simply manufacturing on a smaller process node. Also, x86 desktop chips are at 4-5GHz these days. If Apple Silicon needs to use higher clock speeds than in the iPhone to be competitive in a workstation context, some of that power efficiency will Inevitably diminish. ARM is efficient, but it isn’t magic. Apple has business reasons for making the transition, alongside performance ones.

I also doubt that tile based rendering or a graphics API are suddenly going to give Apple a huge advantage over Nvidia (or even AMD) and Dx12 / Vulkan, somehow making chips that are as fast as the industry leader’s whilst using a fraction of the power. Again, Apple’s A-series advantage is in performance per watt, not absolute performance.

 

[deconstruct60]

....

A major part of the ludicrous cost of the 2019 MP is the expensive Xeon CPUs.

It really is not. The W-3223 lists at $749

Intel® Xeon® W-3223 Processor (16.5M Cache, 3.50 GHz) Product Specifications

Intel® Xeon® W-3223 Processor (16.5M Cache, 3.50 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

That amounts to only about 12% of the entry Mac Pro price.

Now let's look at the entry Mini with a Core I processor that is suppose to be dramatically better. Core i3-8100B is $133

Intel® Core™ i3-8100B Processor (4M Cache, 3.60 GHz) Product Specifications

Intel® Core™ i3-8100B Processor (4M Cache, 3.60 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

That amounts to only about 16% percent of the system price. Two points higher still for the non-Xeon.

Now as creep higher in the line up and absolutely when land on the top two BTO CPU choices that have a 'M' suffix on them with a $3000. " > 1TB RAM" tax slapped on, then you do costs spinning out. But 'Xeon'. No it doesn't drive up the prices in and that simply as a product label prefix. If Apple included the non M models and just stuck to under 1TB of RAM they could have shaved alot off the systems price. Apple didn't. Intel provides the CPUs. Apple skipped them.


A decent chunk of the base Mac Pro price is a "low volume" tax. That is highly likely because not relatvely (to rest of Mac market; nothing to do with iPhones or iPads or etc. ) not selling very many and Apple needs that kicker to be interested ( ignore the low volume threshold cutoff metrics ) .

If using their own silicon lowers the BOM significantly, it’s possible the new MP will be cheaper,

For the "RAM tax" models probably. AMD already offers a lower solution. Intel probably will on the next iteration also. [ Notice though how Apple is of so conviently ignoring the price cut on W-2200 chips that could go into a refreshed iMac Pro .... those iMac Pro prices are holding still. So if they repeat that behavor with Apple Silicon there is highly likely no pass through on CPU price cuts. Just a shift to higher BOM costs for non CPU items or used to fill the Scrooge McDuck money pit. ]

They may put better SSD and RAM capacities to start with, but Mac Pro probably isn't going to drop dramatically in price if the expected volume is about the same. If the HP/Lenovo/Dell/Boxx higher end worskstations all move back a bit due to the newer price competition between AMD and Intel , then that would perhaps get Apple to shift but the lower CPU BOM costs are not likely a huge driver here.

There other large assumption there is that Apple can actually make the super low volume, largest die CPUs cheaper than Intel does. Intel's markups on those include lots of profit margin but those chips are harder to make and typically have lots of software (kernel scheduler mods , NUMA work arounds , fancy rare equipment drivers , etc.) costs associated with them too. Once get into the very high core counts that not even most Mac Pro buyers select then get even more smaller. Pretty good chance that Apple too is going to make 8-16 core buyers pay a substantive amount of the 'freight' of the 28-32 core parts too. The smaller the sky high core count volume the bigger 'tax' is going to be the lower range.

whilst still maintaining margins. This could increase the market for the model, making it more worthwhile from Apple’s point of view. Perhaps Apple are even aware the current MP is overpriced, but aren’t too concerned because it was always going to be a temporary situation.

You are probably going to be disappointed. Apple knew fewer folks would buy because that is why the put a "low volume" tax on the system. If the Mac Pro is selling "enough" , they probably are not going to move much.

Apple made a huge, public commitment to the Mac Pro tier of the range by releasing the 2019 version. Perhaps for many years, it seemed unlikely they would have been able to make an ARM chip that could compete at this end of the market, so had decided to just let the Mac Pro die off with the intended ARM transition. At some point in the last few years, though, they must have realised it was practical after all. The fact they have announced a transition of the complete line up within two years indicates this is a completely solved problem - and have a compelling roadmap for years to come. It is inconceivable they would announce the transition, then worry about how they were going to replace the Mac Pro later.

First, they may not be competing at the whole end of the market where the Mac Pro is now. 28 cores in 2018-2019 isn't the same market as 28 cores in 2022-24. If Apple is putting a line under it of "32" as enough for 2-3 years then that Mac Pro will be more detached from where the higher end general market Workstations will be at in 2022. But that may be enough for the subset of the market that Apple intends to hold onto.

They probably don't intend to hold onto everything for everybody from the 2017-2019 workstation market that insatiably always "has to have" the highest possible main CPU core count. At some point, they can cover most workloads with enough. Same way they don't have to cover the $2,500-4,000 box with slots now.

if Apple makes a new CPU package that is 10-15% faster on CPU core only apps than the current ( effectively 2017-18) Intel solution then they probably declare 'victory' and move on. The next Mac Pro will compare the previous (2022) Mac Pro again.

What Intel and AMD are going to be doing in the high double digit core count zone is probably going to be different and Apple probably isn't/won't be really going to try to compete there. It makes no sense. Apple is making maybe around 8-20K of those "top end" processors and Intel and AMD are making one (or two) orders of magnitude more . It is not going to be "fall out of bed in the morning" easy for Apple to compete there. The units cost will go up because the volume is so low. Some of what Intel/AMD would pocket as profit Apple is going to have to pay to cover costs (which the most certainly are going to pass along). If Apple is doing the Mac pro it is probably because the bottom 'half' of the range makes sense; not the very top end. The top end is likely to get pragmatically prune depending upon where the steady state volume level fall become.


Apple Silicon that works better in a lower power consumption zone will help the iMac Pro iterate forward much better than it will the Mac Pro ( which didn't have the problem of being case constrained). Apple will probably put more effort there than on a case that they hadn't painted themselves into a corner on. there is probably enough of that project to get a Mac Pro out of it too with some incremental extra work.


P.S. on the processor ratio thing the iMac Pro is even more illstrative of what is going on with the base Mac Pro price.

Entry iMac Pro has a variation of 8 core W-2145 so use that price. $1113

Intel® Xeon® W-2145 Processor (11M Cache, 3.70 GHz) Product Specifications

Intel® Xeon® W-2145 Processor (11M Cache, 3.70 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

Which comes out to 22% of the iMac Pro entry price. This system has a screen cost in the BOM. The Mac Pro has no screen at all and the CPU costs is almost 10% lower fraction of the system cost. Both of them "Xeon" prefix productions powering the systems. "Xeon" itself isn't the issue. The iMac Pro should be lower but it is largely Apple keeping it higher. Switching to a W-2245 would drop the CPU cost down to $667 ( ~13% ) .

 

[Quu]

On laptop chips. On Desktop chips they do not.

In your own screenshot it shows x16 PCIe from the CPU directly.

 

[LeonPro]

Let's be real. The ludicrous cost of the 2019 Mac Pro is due to the Apple tax. The same way there's a Porsche tax and the likes.

And I'm okay with that.

This includes the ludicrous R&D that goes into the product including the aluminum material, unnecessary ease of disassembly of the chassis, cable-less motherboard/module, silent fans, and excellent airflow. All amounting to an over-engineered workstation.

When you pay for all of that and are only looking for the speed factor, then this product isn't for you. Go build a PC. I did those back in the day and had fun and stress in the process.

Being older, I'm at the point where I like something that just works as well as works reliably and designed beautifully.

When the ARM Mac Pros come out, I'll be waiting all over again.

 

[deconstruct60]

In your own screenshot it shows x16 PCIe from the CPU directly.

Playing move the goal pots away from Thunderbolt ?

 

[Quu]

Playing move the goal pots away from Thunderbolt ?

It is my opinion the I/O die they develop will contain a thunderbolt controller and it'll be used on all the Macs. For the Mac Pro additional controllers will be present on add-in cards (MPX module cards if they continue with that interface which I think they will).

For instance the two top Thunderbolt ports on the Mac Pro I think will use the SoC's controller while the ones on the back will use the graphics cards controller. Perhaps they'll use the SoC one for the add-in card that is currently present on the Mac Pro also.

For the MacBook / MacBook Pro / iMac / Mac Mini I'd expect the SoC Thunderbolt silicon to be used exclusively and for the iMac Pro perhaps an added Thunderbolt chip but it depends how many ports they provide. At the moment Intel uses 1 controller for every 2 ports. But there's no reason Apple has to stay to that hierarchy they could provide more PCIe lanes at the back end and do more ports per controller.

 

[mode11]

A decent chunk of the base Mac Pro price is a "low volume" tax

You make a lot of good points. I hadn't appreciated that the Xeons in the base machine, whilst expensive by desktop PC standards (you'd pay less for a top end 16-Core Ryzen), don't actually make up much of the cost. And Apple may get them for less than list anyway, given the good PR for Intel that comes from them being featured in a high profile machine. In fact, as you mention, it's going to be hard for Apple to make these presumably large and very low volume chips for less than Intel. Perhaps they'll use a chiplet design and put one, two or four iMac processors in one package? Or even go multi-socket.

I agree that Apple will likely just cede the ultra high-end workstation market to the PC and top out with something sufficient for almost anyone (e.g. 32 cores) - as you say, most people aren't buying the highest end Mac Pros at the moment. Most heavy processing (on workstations) is done with GPUs anyway, although it's a shame Apple doesn't offer Nvidia cards. People doing ML stuff or GPU rendering seem to have their boxes stuffed with GeForce cards.

 

[Moonjumper]

Yes, but Intel are stuck on 14nm whereas GPUs are either already on 7nm or soon will be. So Apple won‘t get the easy advantage of simply manufacturing on a smaller process node. Also, x86 desktop chips are at 4-5GHz these days. If Apple Silicon needs to use higher clock speeds than in the iPhone to be competitive in a workstation context, some of that power efficiency will Inevitably diminish. ARM is efficient, but it isn’t magic. Apple has business reasons for making the transition, alongside performance ones.

I also doubt that tile based rendering or a graphics API are suddenly going to give Apple a huge advantage over Nvidia (or even AMD) and Dx12 / Vulkan, somehow making chips that are as fast as the industry leader’s whilst using a fraction of the power. Again, Apple’s A-series advantage is in performance per watt, not absolute performance.

I agree there is not the advantage of running on a smaller process node, which is why I did not challenge that assertion. It was your assertion SoC is not practical because GPUs consume hundreds of watts that I was commenting on, and your statement about about an advantage in performance per watt helps backs that up.

Apple have commented on the advantages of Tile-Based Deferred Rendering, and I believe it will be a major differentiator. But that was given as an example, not the total solution.

I think I did not properly include the gist of what I was trying to get at. Apple's approach is very different in so many areas that using examples of what works or doesn't work in the Intel + various GPU options cannot be relied upon to say what will happen with Apple Silicon.

 

[mode11]

I agree that Apple may be able to try some different approaches, given they control the OS that has to support the hardware. But Apple are still bound by the laws of physics, and certain computing paradigms. For example, TBDR has been around since the mid-nineties (the Dreamcast used it), and has both advantages and disadvantages. If it was a pure slam dunk, everyone would be using it.

The general point about performance per watt is that all architectures use disproportionately more power at higher clock speeds, partly because they tend to need more voltage to achieve these speeds stably. Processors designed for higher frequencies also need additional features. A Raspberry Pi is probably incredibly efficient for the performance it gives, but it can't just be given more power and cranked up to desktop level performance. The CPU would likely need a longer pipeline, with better branch prediction to keep that filled etc, in order to scale in frequency. By that point you've got a desktop CPU. Intel aren't idiots, despite their recent troubles.

 

[Unregistered 4U]

One slide on a WWDC presentation is not evidence to the contrary.

It’s not “one slide on a WWDC presentation“. It’s referenced being connected to several of the enabling technologies of Apple Silicon. However, there’s no way to see it for yourself other than “seeing it for yourself” (watching a bunch of the presentations). They have a transcript of all of them, but it’s probably better to listen to it in the background while doing something else, then you can flip to the video if you need the visual.

For example, TBDR has been around since the mid-nineties (the Dreamcast used it), and has both advantages and disadvantages. If it was a pure slam dunk, everyone would be using it.

When I was looking up the technical details of the Dreamcast from awhile back, I was always curious why TBDR had never been more widely adopted. It appears that some of the earlier implementations weren’t up to the task, performing some actions in software instead of hardware. Anyway, around that time I came across this article/video. NVIDIA wasn’t performing TBDR in the strictest sense, but certainly some lessons learned filtered their way into how they do IMR.

Just a few days ago, there were folks saying that An ARM processor cannot run macOS. I’m going to be holding off any of my “cannot’s“ regarding their GPU setup until I’m able to make a purchase and check it out myself. They obviously THINK they can match the current Intel/AMD performance or this discussion wouldn’t even exist!

 

[Pressure]

A long time ago there were the PowerVR Kyro graphic cards.

 

[mode11]

Just a few days ago, there were folks saying that An ARM processor cannot run macOS.

On the contrary, everyone has been predicting the Mac transition to ARM.

They obviously THINK they can match the current Intel/AMD performance or this discussion wouldn’t even exist!

Matching the performance of Intel UHD graphics? No problem. Matching AMD Big Navi? Not so much.

 

[macguru9999]

What does it mean ? It means that the arm mac mini might be a useful computer, with a powerful processor and useful integrated graphics. A viable desktop computer if it can be configured with enough ram and storage. but being apple, there will probably be a catch. And an arm mac pro ? thats anyone's guess at the moment, and for a couple of years ill bet. if ever.

 

[StuAff]

Matching the performance of Intel UHD graphics? No problem. Matching AMD Big Navi? Not so much.

I agree. Apple obviously wants/expects/demands developers target the GPU on the SoC. So, the WWDC presentations are concentrating on that. Quite right. Inferring that Apple is going to have that, or some internally developed GPU standalone, as its sole graphics solution…no. Just, no. Having spent years building support for all the GPU you can fit/attach to a system, they're not going to throw all that away. There are references to upcoming AMD products in the Big Sur betas. They're not going to do that just for the x86 machines they haven't launched yet. Even if the next Mac Pro was quad SoC, those integrated GPUs aren't going to be a match for a couple of Vega II Duos & 128GB of dedicated memory, let alone whatever AMD and Apple are actually working on.

 

[mode11]

Apple will certainly use their integrated graphics on all their low end machines / ultra light laptops, and perhaps all the way up to their top end MacBook Pro’s. Smaller iMacs too. Larger iMacs could use a ‘Pro’ 8-core ARM chip featuring only CPU cores (+ related logic), plus a midrange AMD GPU. The Mac Pro could then use multiple ‘Pro’ 8-core chiplets in one package; one at the low end, two at the mid range and four at the high end. Then combine that with MPX modules over PCIe 4.0 - whatever the equivalent of Vega II (Duo) is at the time of launch.

Where would this leave the iMac Pro? Discontinued most likely. It was conceived as the replacement for the Mac Pro, but once they changed their mind it‘s arguably had a question mark over it. The ARM iMacs will surely abandon HDDs (even Apple isn’t brazen enough to treat SSDs as a premium option in 2021), so would then have space for a version of the Pro’s cooling system, currently one of it’s major advantages.