Apple Silicon Mac Pro predictions

[Jorbanead]

If designing an ARM CPU for the Mac Pro alone made no economical sense to Apple, I guess they could perhaps source a high-performance ARM server CPU from someone else. Marvell’s Thunder X2 for instance has up to 32 cores/128 threads and supports up to 2TB of RAM.

Apple wouldn’t do this. Their custom chips have so many differences from the competition that their flagship Mac would lack many key features their chips are known for that all their other macs would have. Their flagship Mac needs to have the best features. The Mac Pro is the prime target for some insane machine learning cores and that will be a huge selling point.

 

[Krevnik]

If designing an ARM CPU for the Mac Pro alone made no economical sense to Apple, I guess they could perhaps source a high-performance ARM server CPU from someone else. Marvell’s Thunder X2 for instance has up to 32 cores/128 threads and supports up to 2TB of RAM.

I don’t even think they have to design something for the Mac Pro alone. They could probably get some decent overlap with the 27” iMac. Apple can even merge the 27” iMac and iMac Pro lines into a single product and get even more overlap with the Mac Pro.

Something Apple could do here to make things easier on themselves would be to follow Threadripper’s lead and support ECC RAM without it being required.

Apple doesn’t have to slice and dice their chip lineup the same way Intel or AMD does.

 

[yyc_engineer]

Deleted. Wrong thread

 

[Jorbanead]

With the news of a possible "mini" Mac Pro (We don't know if this is replacing the existing Mac Pro, or if this is a new product) I have fantasized about what the internals may look like. Here is my prediction:

The Apple Silicon Mac Pro will cut down on PCIe slots. There will be one MPX slot, and 5 total PCIe slots. There will only be 2 large fans instead of 3. The rest of the machine will be very similar to what we have now.

The ASi Mac Pro MPX modules will be slightly shorter than the intel Mac Pro. They still include the same two connectors in the same place, but I have cut out the extra room that was added after the thunderbolt connector. The new MPX modules will work on both new and old Mac Pro, but the current MPX modules will only work on intel (since they will be slightly longer).

 

[Joe The Dragon]

I would say current apple* video MPX cards will not work in new system.

*non apple rom updates may make them work somewhat in new system

Some new MPX video cards may not work in old due to ARM only roms.

 

[Apple Knowledge Navigator]

I seem to recall that Adobe ported over apps in a few days.

Source?

 

[ArPe]

With the news of a possible "mini" Mac Pro (We don't know if this is replacing the existing Mac Pro, or if this is a new product) I have fantasized about what the internals may look like. Here is my prediction:

The Apple Silicon Mac Pro will cut down on PCIe slots. There will be one MPX slot, and 5 total PCIe slots. There will only be 2 large fans instead of 3. The rest of the machine will be very similar to what we have now.

The ASi Mac Pro MPX modules will be slightly shorter than the intel Mac Pro. They still include the same two connectors in the same place, but I have cut out the extra room that was added after the thunderbolt connector. The new MPX modules will work on both new and old Mac Pro, but the current MPX modules will only work on intel (since they will be slightly longer).

View attachment 1117840

Lol that’s hilarious. PCIE slots have to be double wide otherwise you’ll have an issue fitting graphics cards in there. 😝

Ideally two double wide slots and two single width for peripheral cards.

 

[Jorbanead]

Lol that’s hilarious. PCIE slots have to be double wide otherwise you’ll have an issue fitting graphics cards in there. 😝

Ideally two double wide slots and two single width for peripheral cards.

They don't have to be - you'll just cover some slots up. I actually meant to only put in 4 slots, with one double wide, but I finished the photoshop export before I realized this. Still the basic concept is more what I was going for.

 

[Apple Desktop Bus (ADB)]

It'll certainly be the last Mac updated to the new processors.

 

[Boil]

They don't have to be - you'll just cover some slots up. I actually meant to only put in 4 slots, with one double wide, but I finished the photoshop export before I realized this. Still the basic concept is more what I was going for.

MPX is a 4-slot card (mostly the ginormous heat sink), your PCIe setup has room for one MPX card & one half-length I/O card, but no Afterburner card...?

I would think the main reason for the MPX slot in the first place was the dual GPU cards needing the extra power...?

Since Apple is moving to Apple silicon GPUs (APUs, whatever) & has stated they will be using Apple GPUs (whatever that really means is to be revealed), I would think the power draw for any Apple GPU might be a good bit lower than a card with dual Vega 20 GPUs & 64GB of HBM2...?

So that means Apple could probably get away with using shorter PCBs, so a shorter overall chassis (even more than what you have shopped)...?

 

[Krevnik]

MPX is a 4-slot card (mostly the ginormous heat sink), your PCIe setup has room for one MPX card & one half-length I/O card, but no Afterburner card...?

I would think the main reason for the MPX slot in the first place was the dual GPU cards needing the extra power...?

Since Apple is moving to Apple silicon GPUs (APUs, whatever) & has stated they will be using Apple GPUs (whatever that really means is to be revealed), I would think the power draw for any Apple GPU might be a good bit lower than a card with dual Vega 20 GPUs & 64GB of HBM2...?

So that means Apple could probably get away with using shorter PCBs, so a shorter overall chassis (even more than what you have shopped)...?

The MPX connector provides 3 things: power, PCIe lanes, and Displayport lanes. Without the DisplayPort lanes, you can’t feed the built-in Thunderbolt 3 ports with video. Without the PCIe lanes, you can’t feed on-GPU Thunderbolt 3 controllers with data connectivity. PCIe also doesn’t provide a ton of power, so even assuming that Apple makes their own MPX GPUs, and that they can cut power consumption in half for the same performance, they’ll still very likely need the extra power from the MPX connector.

The other hurdle is that MPX GPUs and the MPX drive bay all assume a certain depth of the chassis (front to back) for the cooling to work properly. So I think your answer of ditching the MPX connector is more likely than shortening that front-to-back length, unless the larger Mac Pro also adopts the new front-to-back length as well. But I don’t think it’s terribly likely they’ll ditch the MPX connector this quickly, and shortening the MPX bay is a great way to piss folks off.

I think a smaller Mac Pro is more likely to ditch the space in the top used for the 2-drive bay accessory that Jorbanead left in. If the SoC is moved to the back side of the board with the RAM using something like the iMac Pro’s cooling setup, then you get more options for the layout of the expansion slots, and more space for a couple slots beyond the single MPX bay.

 

[Jorbanead]

MPX is a 4-slot card (mostly the ginormous heat sink), your PCIe setup has room for one MPX card & one half-length I/O card, but no Afterburner card...?

The current MPX modules take up 4 slots, but that doesn’t mean the new ones will.


Apple Silicon will likely have video encoders and decoders that could possibly handle a lot of what the afterburner card does - but there’s still room Assuming Apple makes the new mpx modules 2-slots and not 4.

I think a smaller Mac Pro is more likely to ditch the space in the top used for the 2-drive bay accessory that Jorbanead left in. If the SoC is moved to the back side of the board with the RAM using something like the iMac Pro’s cooling setup, then you get more options for the layout of the expansion slots, and more space for a couple slots beyond the single MPX bay.

I think you’re right actually but I did this photoshop really quick and it was more just a quick mock-up of the size of machine we could see while still having plenty of room for modules. I would personally like to see that space used in some way.

 

[Krevnik]

Apple Silicon will likely have video encoders and decoders that could possibly handle a lot of what the afterburner card does - but there’s still room Assuming Apple makes the new mpx modules 2-slots and not 4.

Depends. At least right now the A series doesn’t have the logic blocks that the Afterburner provides. But Apple could add it, for sure.

The Afterburner is also more flexible by being an FPGA rather than an ASIC. This means it can morph into different accelerators, and can be used to allow new accelerators to be brought to older SoCs. So if Apple’s willing to invest, there’s quite a bit of value there.

As for MPX modules going 2 slot, I think we will have to wait and see.

I think you’re right actually but I did this photoshop really quick and it was more just a quick mock-up of the size of machine we could see while still having plenty of room for modules. I would personally like to see that space used in some way.

No worries, I was trying to use your mock up as a point of reference more than criticize it.

 

[Boil]

The heat sink on a MPX GPU has thicker fins than normal, with a wider than normal spacing of said fins as well; this is needed for proper airflow & heat removal since the only air is coming from the front fans... Now, if the power draw of an Apple GPU is significantly less than that of 'normal' mid to high end GPUs (250 to 400 watts these days), they might be able to get by with a 2 or 3 slot heat sink, rather than the four slot monster that is the current norm...

Just because they are switching to their own Apple silicon APU / CPU / GPU / whatever does not mean they are suddenly going to shrink the goodly-sized single-slot Afterburner card into a sliver of a speck of silicon in aforementioned computing chip...

Now, being that the main compute chip n a reduced-size Mac Pro is supposed to include a GPU, an add-in GPU is not required for functionality, meaning the MPX slot is open, as are the ones it would normally block...

So audio dudes could have four full-length PCIe slots for their audio add-in cards, and a half-length slot up top for the I/O...

Video editors could add a GPU in if they want, but they would still have the slot open for an Afterburner card... Video dudes may prefer some sort of A/V I/O card as well, or a few M.2 NVMe SSD RAID PCIe cards...?

Seems like the (possible / potential / whatever) reduced power draw from the transition to Apple silicon will make for reduced cooling needs, making for reduced mass needed in semi-passive heat sinks... Smaller chassis...!

But come on Apple, put that same reduced package Mac Pro mobo into a less 'fussy' chassis, give the masses a lower cost Mac tower, give us the mythical xMac...!!!

EDIT TO ADD:

I would hope, reduced-size Mac Pro or xMac, for four full-length PCIe slots (bottom one a MPX slot?) & one half-length PCIe slot (for the I/O card)...

Then I would want an Apple GPU (2-slot, but still the semi-passive heat sink), a single-slot M.2 NVMe Gen4 SSD RAID card, a single-slot Afterburner card, & the single-slot half length I/O card could all fit in there...!

Use the a few of the TB4 ports for A/V I/O & a control surface or two, a few more for a couple of monitors & you would have a complete personal Digital Content Creation workstation...!

The modern day SGI O2...!

I would just rather forgo the fancy post modern cheesegrater grill & space frame chassis (with useless handles & feet or wheels) for a cost reducing (yet still a quality Apple design) chassis; and not needing a 1.4kW PSU has to save a few bucks as well...!

Come on Apple, xMac...!!!

 

[DPUser]

Come on Apple, xMac...!!!

Yes!

 

[wmy5]

While I am excited about the performance and energy efficiently of Apple GPU, I still think the likelihood of Apple producing its own high-end, discrete GPU is pretty low. Not that Apple cannot, but the sale volume of this kind chips can't justify the development/manufacturing cost. How many Mac Pro can Apple sell? I think it's only a few % of MacBook Pro 13.

I think Apple will continue to use AMD discrete GPU, along with Apple-designed CPU.

 

[Krevnik]

While I am excited about the performance and energy efficiently of Apple GPU, I still think the likelihood of Apple producing its own high-end, discrete GPU is pretty low. Not that Apple cannot, but the sale volume of this kind chips can't justify the development/manufacturing cost. How many Mac Pro can Apple sell? I think it's only a few % of MacBook Pro 13.

I think Apple will continue to use AMD discrete GPU, along with Apple-designed CPU.

I tend to agree, but one thing Apple could do is use their own GPU to make a discrete GPU optional on the Mac Pro. They aren’t completely out of reach of being able to replace the 580X or W5500X with their own GPU. So roll that into the SoC, and make a discrete GPU an add-on for those that want/need it.

With that sort of setup, a smaller Mac Pro with a single MPX bay starts making more sense.

 

[wmy5]

I tend to agree, but one thing Apple could do is use their own GPU to make a discrete GPU optional on the Mac Pro. They aren’t completely out of reach of being able to replace the 580X or W5500X with their own GPU. So roll that into the SoC, and make a discrete GPU an add-on for those that want/need it.

With that sort of setup, a smaller Mac Pro with a single MPX bay starts making more sense.

Another possibility is that Apple will design a discrete GPU chip suitable for MacBook Pro 16 and iMac where the sale volume can amortize the cost of developing a new chip. For Mac Pro Apple may adopt a multi-chip design. For example, a MPX module can fit 2-8 mobile discrete GPU chips.

 

[deconstruct60]

Another possibility is that Apple will design a discrete GPU chip suitable for MacBook Pro 16 and iMac where the sale volume can amortize the cost of developing a new chip. For Mac Pro Apple may adopt a multi-chip design. For example, a MPX module can fit 2-8 mobile discrete GPU chips.

There are in the same zipcode of covering the MBP 16" current dGPU with what they have now. If the display output system was much better ( e.g, could do 4 independent DPv1.4 streams ) , the GPU cores doubled up , and increase the system memory bandwidth they'd cover it. And that could cover the iMac 21.5" dGPU usage also.

Likely the only thing left after Apple's "assault" on dGPUs will be the 27" iMac variants and the Mac Pro. Apple probably wants to nuke all dGPUs in their laptops. And then whatever else falls out of that as they reuse the MBP 16" SoC in the desktops. (Mini , iMac 21-24" ).

A discrete Apple GPU would mean they'd have to create another Metal 'family' that had different characteristics. I think that they can just avoid that if they just detach the smaller screen iMacs and leave "the rest" to AMD. If they make "the rest" pool too small they run the risk of the 3rd party vendor just walking away. And then Apple would have to do the low volume work themselves. I think that is going to be a large enough task for Apple on the Mac Pro CPU SoC without doubling down on it.

2-8 mobile GPUs isn't going to do much. The GPU memory is going to be so fragmented that there would only be more niche applications that will be conducive to that. Intel is going to try with the DG1 ( Xe-LP )

https://videocardz.com/press-release/intel-announces-h3c-xg310-server-gpu-with-four-xe-lp-gpus

and and the Xe-HP

https://www.anandtech.com/show/16238/intels-xehp-now-available-to-select-customers

There is not a super long line of folks lining up to get those. ( I think there is some synergies for Mac Pros focused on video transcoding on the first but wouldn't be a broad usage card. )


P.S. I suspect that what some folks are calling a "discrete" Apple GPU is really a major tweak to their iGPU design that perhaps has a larger internal-to-GPU cache (similar to AMD's Infinity Cache) or some kind of small (2GB) HBM "backside" VRAM/Cache attached to it for framebuffers and bulky texttures they don't want to clog up the unified system cache with. It still would be Unified Memory ... just much less thrashing of the shared System Cacche and Unified RAM. From the CPU side it could still present a Unified Memory Programming model and be more an adjustment to what is cached where. If it had the HBM as VRAM-cache then it would somewhat look like a dGPU on the same package but really still be an iGPU.

 

[Krevnik]

Another possibility is that Apple will design a discrete GPU chip suitable for MacBook Pro 16 and iMac where the sale volume can amortize the cost of developing a new chip. For Mac Pro Apple may adopt a multi-chip design. For example, a MPX module can fit 2-8 mobile discrete GPU chips.

The thing is, the M1 is already in striking distance of the GPUs used in the 16” MBP, and the base GPUs in the Mac Pro. I’d give good odds of Apple being able to ditch discrete GPUs on the 16” and a good chunk of the iMac line. What sort of dGPU Apple uses where they need one is a bit academic at this point until get more information. I tend to think that Apple is as likely to produce their own dGPU as not.

Right now though, when buying a Mac Pro, you need one MPX bay just for a GPU. The second bay can be used for RAID or GPU. So the rumor of a smaller Mac Pro makes more sense if the smaller Mac Pro didn’t need to spend an MPX bay on a GPU unless you actually had need of it.

 

[iPadified]

Power draw scales linearity with core count as i performance. The MP can cool a >700mm^2 chip at 300W. Seem to be that the 120^2 of M1 can easily be scaled a factor of six on everything (as metal score is around 20.000 for M1, a scaled chip would have 120.000). How much more graphics power do you need? It is quite different with compute/ML and there we may see MPX modules the are dedicated to that.

Will we see these behemoth chip? Perhaps not but then Apple need to solve modularity to get the correct GPU/CPU mixture for every high end user.

I do not think we will see AMD GPUs in the small MP. Another platform to optimise code for and it has been a mess for decades...

 

[wmy5]

Likely the only thing left after Apple's "assault" on dGPUs will be the 27" iMac variants and the Mac Pro. Apple probably wants to nuke all dGPUs in their laptops. And then whatever else falls out of that as they reuse the MBP 16" SoC in the desktops. (Mini , iMac 21-24" ).

A discrete Apple GPU would mean they'd have to create another Metal 'family' that had different characteristics. I think that they can just avoid that if they just detach the smaller screen iMacs and leave "the rest" to AMD. If they make "the rest" pool too small they run the risk of the 3rd party vendor just walking away. And then Apple would have to do the low volume work themselves. I think that is going to be a large enough task for Apple on the Mac Pro CPU SoC without doubling down on it.

Yeah, good points. I tend to agree. A "M1X' class SoC certainly cover all MacBook Pro 16 line up and iMac 21 (or future 24 as rumors suggest). I am not certain AMD graphics can catch up with Apple's design in terms of performance per watt. Their recent RX 6000 series is a big step forward but are they Apple-level good?

 

[deconstruct60]

I am not certain AMD graphics can catch up with Apple's design in terms of performance per watt. Their recent RX 6000 series is a big step forward but are they Apple-level good?

https://www.anandtech.com/show/14492/samsung-amds-gpu-licensing-an-interesting-collaboration

https://www.notebookcheck.net/AMD-a...roys-the-Adreno-650-in-GFXBench.463359.0.html

AMD can catch up if they pot money and resources into it and they don't get entangled in some Samsung process variant that goes sideways. They need a slightly different design constraint criteria. But they have already been working toward that for over a year. It probably won't pop out of the pipeline until next Fall '21 earliest ( if everything goes right in the joint venture).

Then AMD would have to weave it back into an APU (which is also likely somewhat coupled to Samsung process).

The advantage that AMD is on track to development a firm foundation on is that they sell lots more GPUs than Apple does. Which means as some put they should be able to have three different sets of teams focused off on three different areas. ( Gaming (+ console) , DataCenter compute , and mobile). If they fumble the Data Center compute then perhaps that messes up the balance. But the MI100 seems on track to get some solid wins. ( It is also a bit of dual edge sword in that they have other irons in the fire too. )

AMD lending hard on "infinity cache" in a manner similar to how hard Apple is leaning on their system cache is demonstrative that they are on the some of the same paths. Just iterating on process fab slower because substantively bigger dies to deal with. When AMD moves RNDA2 down to 5nm with some tweaks they will get another "pop" in pref/watt. ( all of the increment from RDBA1 to RDNA2 was without a huge fab node improvement. Can't really say they don't know how to optimize a specific targeted fab node. )

But AMD isn't likely to close the gap rapidly in the next 24 months. They are getting ramped up to get in the race though. However, they are making serious money doing things with a slightly different focus so not a "throw everybody at this" development area. (and how Nvidia digests ARM , if allowed, could throw a curve ball at this also. ) It is defiantly possible they will keep the edge on the higher end discrete GPU performance value add. It gets more murky whether they are going to help Windows 10 in the medium range future in the > $1.4K laptop space.

 

[Serban55]

Amd can catch up, but under that old windows 10 as OS
Its clear Satya Nadella doesnt care any more about win10....microsoft is focus on services.Im sad to see windows10 let go, for 3 years in a row no big improvements were made ...but on the apple side..im pretty impress even with something "small" like iphone ProRaw
So i hope amd , intel and everyone can still use an OS that is "worthy" for their hardware in 2030
Because linux is not there for now