Apple Silicon Mac Pro predictions

[trevpimp]

I feel like an Chipset produced by Apple's Silicon that would be equivalent to the processing power needed in comparison to the Mac Pros current configuration I would be highly amazed at how Apple has transitioned such processor power with such little power consumption and processing speeds

I do expect this to be long awaited and anticipated with their current transition

 

[playtech1]

I wonder whether Apple will create a CPU specifically for the Pro, given that it's a fairly niche product (at least at the current price point). Same for a lot of the other elements, like a Pro level GPU or something equivalent to SLI/Infinity Fabric to connect multiple GPUs.

I hope I'm wrong, but I would bet that whatever CPU goes into the Mini / iMac equivalents will also go into the Pro, albeit there will be more of them, perhaps clocked a bit higher with beefier cooling. First Apple quad CPU perhaps?

 

[Kostask]

Could go ether way.

I think there will be 3 SoCs: the "base" SoC, with on board 16GB RAM, all the basic functions, at least a 4 HP core CPU (I'm thinking 8), the 4 USB4/TB4 ports, whatever hardware accelerators that Apple deems necessary (Touch ID, Facetime Camera, etx.). This goes into the base Mac Mini, the entry level Laptop, the lowwest end small MacBook Pro (I don't want to get into what screen size or case will be used), and probably the iMac 24". The "midrange" SoC, with more CPU and CPU cores, more or beefed up iGPU cores, and the 4 USB/TB4 ports, still with the 16GB of SoC RAM. Clocks will be faster across the board. This goes into the higher end small MacBook Pro, the 16" MacBook Pro, the higher end 24" iMac, the Mac Mini Server, and the base iMac 30".

Within each of the SoC types, you will see variations in maximum RAM size (or even, in the case of the entr7 level laptop, whether there are memory upgrade options at all, and only two USB5/TB4 ports; the smaller MacBook Pro would have memory upgrades available and 4 USB/TB4 ports), and clock speeds. You would have the same in the midrange SoC lineup.

i peronsally believe that there will be a "performance" SoC, with maximized HP core count, maximized ML/AI core count and power, highest performance iGPU, and added on logic to permit a lot of external (=off SoC) RAM, dedicated GPUs and MPX modules, and logic to allow multiple SoCs to work together. Then you would get the top line 30" iMac using one or two of these "perfromance SoCs", and the new AS Mac Pro using 1-4 (or possibly even more) of them. Again with the various clock speed options.

 

[jinnyman]

I pretty much agree that SoC approach is not good for Mac Pro level of workload.
It’s not good for thermal atleast.

If Apple is planning to continue MP, then they have to come up with some way to incorporate dGPU or Apple’s version of dGPU. Maxing out GPU with huge rendering tasks while sharing memory bandwidth with CPU does seem no go.

 

[Boil]

Could go ether way.

I think there will be 3 SoCs: the "base" SoC, with on board 16GB RAM, all the basic functions, at least a 4 HP core CPU (I'm thinking 8), the 4 USB4/TB4 ports, whatever hardware accelerators that Apple deems necessary (Touch ID, Facetime Camera, etx.). This goes into the base Mac Mini, the entry level Laptop, the lowwest end small MacBook Pro (I don't want to get into what screen size or case will be used), and probably the iMac 24". The "midrange" SoC, with more CPU and CPU cores, more or beefed up iGPU cores, and the 4 USB/TB4 ports, still with the 16GB of SoC RAM. Clocks will be faster across the board. This goes into the higher end small MacBook Pro, the 16" MacBook Pro, the higher end 24" iMac, the Mac Mini Server, and the base iMac 30".

Within each of the SoC types, you will see variations in maximum RAM size (or even, in the case of the entr7 level laptop, whether there are memory upgrade options at all, and only two USB5/TB4 ports; the smaller MacBook Pro would have memory upgrades available and 4 USB/TB4 ports), and clock speeds. You would have the same in the midrange SoC lineup.

i peronsally believe that there will be a "performance" SoC, with maximized HP core count, maximized ML/AI core count and power, highest performance iGPU, and added on logic to permit a lot of external (=off SoC) RAM, dedicated GPUs and MPX modules, and logic to allow multiple SoCs to work together. Then you would get the top line 30" iMac using one or two of these "perfromance SoCs", and the new AS Mac Pro using 1-4 (or possibly even more) of them. Again with the various clock speed options.

USB4 has TB3 as part of the spec, TB4 is its own thing, and there is no USB5 yet...

 

[glindon]

I don't think it can. Not when we're talking potential terabytes of RAM anyway. I think the emphasis is on the unified memory for the 1st gen silicon, but the "Pro" machines will have to address more RAM that can be reasonably fit onto a CPU die. Also the whole point of GDDR is that it's better for graphics workloads, so why would they opt for just using the system RAM? I think to at least achieve feature-parity with the current Mac Pro, Apple would need to have at least a dedicated GPU and VRAM. Maybe as an add-in card? They have special MPX slots in the 7,1, maybe they'll be updated with a faster proprietary interconnect.

Unified memory doesn't mean it has to be on-die. iPads have had the memory chip next to the cpu instead of on top of the cpu like iPhone where space is limited. Obviously they are going to have some design something similar to AMD infinity fabric to link all the cpus, gpus, and shared memory. Apple gpus are tile rendered so they can do more with less memory so less will be needed there.

 

[Kostask]

TB3 and TB4 are pretty much the same. The Data Transfer speed is the same, and the connectors are the same. TB4 is more about power delivery, video output capability, and the use of PCIe 4.0 behind the port. Both USB 4 and TB4 use the USB "C" connector. AS SoCs will use TB4, as it is TB3 compatible, and if you are developing new SoC port controllers anyway, why would you develop for an older specification? It doesn't impact existing TB3 devices, and will remain useful going forward for longer.

USB4 does NOT preclude the use of TB4.

 

[Sarajiel]

I don't think it can. Not when we're talking potential terabytes of RAM anyway. I think the emphasis is on the unified memory for the 1st gen silicon, but the "Pro" machines will have to address more RAM that can be reasonably fit onto a CPU die. Also the whole point of GDDR is that it's better for graphics workloads, so why would they opt for just using the system RAM? I think to at least achieve feature-parity with the current Mac Pro, Apple would need to have at least a dedicated GPU and VRAM. Maybe as an add-in card? They have special MPX slots in the 7,1, maybe they'll be updated with a faster proprietary interconnect.

Maybe I'm thinking a bit old school here, but Apple could try to use a crossbar switch between system memory, storage controllers, I/O devices, CPU and GPUs. The old Silicon Graphics Inc. machines like the sgi Octane workstation and a bunch of their HPC servers used such a system for the compute nodes within the ccNUMA fabric.
As a reminder, those systems dominated the world of supercomputers in the early and mid 1990s and were also the kind of machines used for high end 3D CGI like in Jurassic Park or Pixar movies of that time, or professional virtual reality solutions for the aeronautics and the automotive industry.

Since a decent high performance CPU will most likely have between 40 and 120 P-cores, it's somewhat unlikely that Apple will bother with a large amount of graphics cores on a monolithic die anyway. They could try to create a chiplet design similar to what AMD uses for many of their current CPUs, but that would potentially require huge amounts of HBM2E stacks to reach enough memory capacity to match the current Mac Pros.

With a crossbar switch Apple could however create a design with multiple ports per system device that connect to the Xbar and are switched on demand between the various devices and the system memory. Such a design would probably scale quite nicely with different memory and GPU configurations for a wider variety of use cases if desired. It would also allow for GPU processing with lesser amounts of data being copied over the PCIe bus since the GPU would have direct access to the same memory as the CPU.
While I have to admit that this is a rather simple description of the technology, it's pretty much what SGI of the old days designed for their graphics subsystems in machines that had up to 256 CPUs and multiple graphic pipelines.

The huge drawbacks with such a system are probably various patents which are most likely all over the place now at the likes of nVidia and AMD after the MIPS spin-off from the old SGI and the later bankruptcy, and of course the quite expensive design in terms of PCB space and custom interconnects which most likely would translate into an even more expensive Mac Pro than the one Apple sells today.

 

[leman]

Maybe I'm thinking a bit old school here, but Apple could try to use a crossbar switch between system memory, storage controllers, I/O devices, CPU and GPUs.

This is also something I imagine happening. Metal already has support for device “locality” - it organizes GPUs into peer groups where each GPU in the same group has fast connection to each other. This is essentially NUMA hierarchy detection.

Apple could build on these APIs to build a NUMA system with a unified memory hierarchy. Such system could use a single RAM backing store, with multiple CPUs and GPUs on extension boards, connected asymmetrically. Each extension board would come with its fast HBM cache. APIs would then allow you to schedule work on CPU and GPU cores that are memory-local to each other, while still being able to communicate with the rest of the system (albeit at a performance cost). I think such design could be successful, if Apple can deliver a NUMA API that is easy enough to use.

 

[rezwits]

TL;DR real quick, if they went to Ryzen, and dropped Intel like a bad habit, for just one Rev would be pretty slick. And then made ARM chips for the MP in say 5 years from now? They won't tho

 

[Hunglk]

But MPX is really just added power and TB and some added pci-e lanes.

now the AMD EPYC dual systems use the pci-e bus for CPU TO CPU links.

Amount of RAM aren't dependent on CPU die!
All Apple silicon chips are 64-bit!

 

[JMacHack]

Unified memory doesn't mean it has to be on-die. iPads have had the memory chip next to the cpu instead of on top of the cpu like iPhone where space is limited. Obviously they are going to have some design something similar to AMD infinity fabric to link all the cpus, gpus, and shared memory. Apple gpus are tile rendered so they can do more with less memory so less will be needed there.

Oh, I was under the assumption that unified memory had to be on the SoC, I'm not very technical.

The huge drawbacks with such a system are probably various patents which are most likely all over the place now at the likes of nVidia and AMD after the MIPS spin-off from the old SGI and the later bankruptcy, and of course the quite expensive design in terms of PCB space and custom interconnects which most likely would translate into an even more expensive Mac Pro than the one Apple sells today.

I was also curious how Apple planned on creating a desktop-level GPU and navigate the minefield of patents. I've heard (let me reiterate, I'm not very technical) that it's extremely difficult due to that.

TL;DR real quick, if they went to Ryzen, and dropped Intel like a bad habit, for just one Rev would be pretty slick. And then made ARM chips for the MP in say 5 years from now? They won't tho

I'm sure it was considered when Zen came out, but if I had to guess Apple sees more potential in their own processors than switching to AMD.

 

[Kostask]

Apple can either come up with their own solution, or license a solution from any number of sources. This is not the stumbling block. They could license Infinity Fabric from AMD in return for dGPU business, or the crossbar switch idea from whoever owns the SGI patents, or some thing that was used on Cray supercomputers. Lots of solutions are possible. Even what they did with the Dual CPU Cheesegrater MacPros may be applicable.

Unified memory means that memory is equally accessible to CPUs and GPUs. It has no implications for on SoC or off SoC, either are valid.

 

[leman]

Oh, I was under the assumption that unified memory had to be on the SoC, I'm not very technical.

Unified memory simply means that the CPU and the GPU share the memory. In other words, in a unified memory architecture the CPU abdrehen GPU are able to access the same data without incurring any copies. A straightforward way to implement this is to route both the CPU and the GPU through the same memory controller. In current Apple SoC, CPU, GPU, memory controllers are shared cache are part of the SoC, but the system memory itself resides on external chips.

 

[leman]

TL;DR real quick, if they went to Ryzen, and dropped Intel like a bad habit, for just one Rev would be pretty slick. And then made ARM chips for the MP in say 5 years from now? They won't tho

Not good enough for Apple. Ryzen current advantage over Intel is sane power consumption, which allows them to get better multi-core performance. But they have slightly lower single core performance.

Apple of the other hand has an architecture with the potential to outperform both the Ryzen and the upcoming Intel CPUs.

 

[theorist9]

Depends on what you use for system RAM. Something like HBM would work. GDDR has a lot of bandwidth (can transport a lot of data at once) but very high latency (takes a lot of time to access data), so it's not suitable to feed a CPU. Low latency of GDDR might also be an issue for some professional workloads, now that the GPUs are more flexible and memory access is less predictable. There is a reason why HBM is widely used in HPC.

As I understand it, with GDDR you have higher bandwidth but higher latency (making it more suitable for GPUs), while with DDR you have lower latency but lower bandwidth (making it more suitable for CPUs). Thus, for unified memory, you'd need something that has both higher bandwith and lower latency, which is where HBM comes in (indeed, from what I've read, it exceeds GDDR in bandwidth and DDR in low latency, and is more efficient than both).

But how much more expensive would HBM be? From OWC you can get 384 GB of Mac Pro RAM (32GB x 12, 2933MHz DDR4 PC4-23400 RDIMM) for $2,250, 768 GB (64GB x 12, 2933MHz DDR4 PC4-23400 RDIMM) for $4,950, or 1.5 TB (128GB x 12, 2933MHz DDR4 PC4-23400 LRDIMM) for $12,500. What might the equivalent sizes cost for HBM?

And if Apple went with HBR, could it be user-upgradable (like DDR RAM) or not (like GDDR RAM)?

[Also, I think you've stlll got a typo in your post, in your 2nd reference to GDDR latency.]

 

[leman]

But how much more expensive would HBM be? From OWC you can get 384 GB of Mac Pro RAM (32GB x 12, 2933MHz DDR4 PC4-23400 RDIMM) for $2,250, 768 GB (64GB x 12, 2933MHz DDR4 PC4-23400 RDIMM) for $4,950, or 1.5 TB (128GB x 12, 2933MHz DDR4 PC4-23400 LRDIMM) for $12,500. What might the equivalent sizes cost for HBM?

And if Apple went with HBR, could it be user-upgradable (like DDR RAM) or not (like GDDR RAM)?

Things like HBM and LPDDR are fundamentally not user upgradeable, since they don’t come as slotted RAM. For the same reason I don’t think it makes much sense to talk about cost, it’s not something you can buy as the end customer anyway. The a RAM itself is more expensive, but there are additional factors as well that increase cost.

I believe that RAM cost is not a factor for Apple, since their computers are already expensive. Apple could afford using HBM at the current price point, especially since Apple Silicon would save them some money. Then, there is the factor of product differentiation. Apple offering HBM in their computers would cement their premium status and offer new performance heights in the customer segment. And they can absorb the costs by making their own silicon plus streamlining the model range. It’s simply not possible in the PC world where components are sold for profit.

And finally, using high speed RAM will “finally” provide the definitive justification for soldered on memory. If you are using regular old DDR, users will reasonably complain. If you are using something that’s 5-6 times faster... then it suddenly sounds like a much better tradeoff.

[Also, I think you've stlll got a typo in your post, in your 2nd reference to GDDR latency.]

Thanks!

 

[Yebubbleman]

I agree. Apple has been working on Silicon for many years, and the redesign of the Mac Pro makes no sense if it is to be used only for one generation. I expect the first AS based Mac Pros to look exactly like the current cheese grater. As for supporting third party graphics cards? I have no idea, could see it going either way. I hope they will leave it open, though.

Also, I would really love to see something between the Mac Pro and the Mac mini, something like the PM Cube. A Mac mini Pro.

They could've easily kept around the cylinder as a Mac Pro Cube type option. Just update the CPU, GPU and Thunderbolt ports from 2 to 3 while keeping the same price point and that would've been fine for the subset of Mac Pro crowd that didn't need either drives or PCIe cards. The cylinder's only flaw as a "Mac Pro" product was that it was the only option offered (and that you had people that legitimately needed drives and PCIe slots). It is affordable enough compared to the current Mac Pro that there would've still been buyers for it.

ASi Mac Pro will be the very last Mac to make the transition/update. Remember these are machines used to make money.

I don't think anyone is denying or disagreeing with that. Also think that this pretty much goes without saying at this point.

I hear you. One thing to keep in mind that Apple actually writes the drivers for all devices in the Mac. It's part of the reason Apple went away from Nvidia. Nvidia won't let Apple write up to the stack the way they want to.

Do you have any sources on this? Not that I'm doubting you; I've just always found the Apple/NVIDIA drama to be interesting. Plus, I'm one of the many that were sad to see them pretty much permanently banned from the Intel Mac parade.

I 100% agree. What I suspect, is we are going to see a much richer fabric design to allow enhanced access to system memory, and any graphics cards would need to have their own dedicated bank of memory... unless they are able to provide high enough bandwidth through the system for it to get access to memory.... I suspect we'll see a T3 at some point that will be dedicated to provisioning of resources at a far more granular level.

The successor to the T2 is Apple Silicon. Unless they brand the SoC as T3, you're probably not seeing that. You might see a different co-processor designed for the Mac Pro, but even then, I'm pretty skeptical.

nVidia won't let anybody write to its hardware, I think they consider it a competitive advantage. Apple won't let anybody write kernel level drivers for MacOS. End result is no nVidia cards on Macs. Not likely to change.

I have heard that the Boot Camp drivers for Mac hardware are written by software houses outside of Apple. That would make sense, as there are a few software houses who can write Windows drivers (as opposed to no software houses who can write MacOS kernel drivers). Apple just licenses those Windows drivers for BootCamp.

Do you have sources on any of that? Again, not doubting; just would love to read more about it!

Also, yeah, with the move to Apple Silicon, no future NVIDIA presence on the Mac basically became a guaranteed certainty.

I wonder whether Apple will create a CPU specifically for the Pro, given that it's a fairly niche product (at least at the current price point). Same for a lot of the other elements, like a Pro level GPU or something equivalent to SLI/Infinity Fabric to connect multiple GPUs.

I hope I'm wrong, but I would bet that whatever CPU goes into the Mini / iMac equivalents will also go into the Pro, albeit there will be more of them, perhaps clocked a bit higher with beefier cooling. First Apple quad CPU perhaps?

I'll bet they'll use a special CPU (even if it isn't marketed as such). I also bet that we'll see the return of multiple CPUs/SoCs to the Mac Pro for the first time since MacPro5,1.

TB3 and TB4 are pretty much the same. The Data Transfer speed is the same, and the connectors are the same. TB4 is more about power delivery, video output capability, and the use of PCIe 4.0 behind the port. Both USB 4 and TB4 use the USB "C" connector. AS SoCs will use TB4, as it is TB3 compatible, and if you are developing new SoC port controllers anyway, why would you develop for an older specification? It doesn't impact existing TB3 devices, and will remain useful going forward for longer.

USB4 does NOT preclude the use of TB4.

What makes you so sure that they'll adopt Thunderbolt 4 out of the gate? Let alone have support built into the SoC? I do think that they'll have to use an Intel controller for Thunderbolt support. But even so, I'd be shocked if they had it ready out of the gate. It took Apple a good while to get Thunderbolt 3 on something (and that was while using Intel Macs still!).

 

[leman]

What makes you so sure that they'll adopt Thunderbolt 4 out of the gate? Let alone have support built into the SoC? I do think that they'll have to use an Intel controller for Thunderbolt support. But even so, I'd be shocked if they had it ready out of the gate. It took Apple a good while to get Thunderbolt 3 on something (and that was while using Intel Macs still!).

At the WWDC they have stated that Apple Silicon Macs will provide isolated MMIO for each connected device for better security. I think this is a strong indication that they have their own controller hardware ready to go. Besides, as they use their own CPUs, they should be able to control the properties of the data bus much better instead of being limited by Intel.I expect Apple Silicon Macs to support USB4 with Thunderbolt at launch.

Also, wasn’t Apple the first company to get Thunderbolt 3 support?

 

[thenewperson]

Also, wasn’t Apple the first company to get Thunderbolt 3 support?

Surprisingly not. According to Wikipedia they were beaten by almost a year by other OEMs.

 

[Yebubbleman]

At the WWDC they have stated that Apple Silicon Macs will provide isolated MMIO for each connected device for better security. I think this is a strong indication that they have their own controller hardware ready to go. Besides, as they use their own CPUs, they should be able to control the properties of the data bus much better instead of being limited by Intel.I expect Apple Silicon Macs to support USB4 with Thunderbolt at launch.

Also, wasn’t Apple the first company to get Thunderbolt 3 support?

Surprisingly not. According to Wikipedia they were beaten by almost a year by other OEMs.

Yeah, I remember seeing a co-worker that had an Alienware that had a single Thunderbolt 3 port and remarking that, at the time, Apple was late to the party (this was maybe a month before the 2016 MacBook Pros came out).


I do believe that Apple was first to both Thunderbolt 1 and 2, though.

 

[leman]

Surprisingly not. According to Wikipedia they were beaten by almost a year by other OEMs.

Yeah, I remember seeing a co-worker that had an Alienware that had a single Thunderbolt 3 port and remarking that, at the time, Apple was late to the party (this was maybe a month before the 2016 MacBook Pros came out).

You are both right. I had another look and Apple took about 10 months more to adopt USB-C and Thunderbolt 3. But when they did, they went in 100%, that is probably what confused me.

 

[Woochoo]

I expect 6 more years of no Mac Pro updates like with the trash can.

 

[Boil]

I expect 6 more years of no Mac Pro updates like with the trash can.

From the Apple press release:

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

Now come the end of that two year tmeframe (we will just say Dec. 31, 2022) I would expect Apple to announce the Apple silicon evolution of the Mac Pro (whether they keep the Cheesegrater v2.0 or not), or an announcement that they are going to cut the "Pro" line-up back to just MacBook Pros & iMac Pros...

I am hoping for a Trashcan V2.0 in a return of the Cube:

Mac Pro Cube - starting at US$5,999.00

48 P cores / 4 E cores / 96 GPU cores - CPU / GPU Chiplets & RAM on interposer / System in Package (SiP) design
HBM3 Unified Memory Architecture - 128GB / 256GB / 512GB
NVMe RAID 0 (dual NAND blades) 4TB / 8TB / 16TB
Eight USB4 / TB4 ports
Two 10Gb Ethernet ports
One HDMI 2.1 port
Three MPX-C slots (for use with asst. MPX-C expansion modules)

But a Cheesegrater V2.0-themed xMac (smaller chassis / less PCIe slotsthan MP7.1, but larger than a modern Cube) would be more than acceptable...!

 

[vigilant]

Do you have any sources on this? Not that I'm doubting you; I've just always found the Apple/NVIDIA drama to be interesting. Plus, I'm one of the many that were sad to see them pretty much permanently banned from the Intel Mac parade.

I mean I’m not going to give you a name and title, but I’ll leave it at I have a couple of friends at NVIDIA. I also have some friends at Apple, but I don’t even know what the Apple friends do. I just won’t ask.