Mac Pro - no expandable memory per Gurman

[Cmaier]

Future Mac Pro may use Apple Silicon & PCI-E GPUs in parallel | AppleInsider

Despite Apple Silicon currently working solely with its own on-board GPU cores, Apple is researching how to support more options, like PCI-E GPUs, all working in tandem.

appleinsider.com

 

[mr_roboto]

Future Mac Pro may use Apple Silicon & PCI-E GPUs in parallel | AppleInsider

Despite Apple Silicon currently working solely with its own on-board GPU cores, Apple is researching how to support more options, like PCI-E GPUs, all working in tandem.

appleinsider.com

My interpretation of the fragments of the patent copied into that article: it probably doesn't cover using arbitrary third party GPUs. Instead, it's about gluing multiple Apple architecture GPUs into a single logical GPU.

 

[exoticspice1]

multiple

My interpretation of the fragments of the patent copied into that article: it probably doesn't cover using arbitrary third party GPUs. Instead, it's about gluing multiple Apple architecture GPUs into a single logical GPU.

So an Apple GPU PCIe card??

 

[Cmaier]

So an Apple GPU PCIe card??

I don’t think the patent application says anything about pcie. It just says slots.

 

[theorist9]

Does the patent explain how UMA is going to work with these extra GPUs? The article offered this quote from the patent, which is possibly related, but I don't know how to interpret it:

"This may improve cache efficiency, in some embodiments, by allowing graphics work that accesses the same memory areas to be assigned to the same group of sub-units that share a cache."

I then converted the patent to PDF, used Acrobat's OCR, and did a search for "memory". This is all I could recognize that might be relevant. There weere no instances of "unified memory", but I did get a hit for "shared memory":

 

[exoticspice1]

I don’t think the patent application says anything about pcie. It just says slots.

Yes but isn't the most common slot PCIe?

 

[exoticspice1]

Guys, the Mac Pro is gonna be whatever Mark German said it won't be. So it will have things that no other AS Mac had before.

The 2019 Mac Pro was so unique compared to the other Macs at the time excluding expandability. It has 2 GPU chips on one GPU card and a custom accelerator and Apple did that on a x86 platform. When Apple owns the whole stack, you would think they would try something different again, right??

 

[Cmaier]

Yes but isn't the most common slot PCIe?

Sure. But there have been other slots before and there will be other slots in the future. No idea what Apple has in mind for the future, and the patent application is not limited to PCIe.

 

[exoticspice1]

m

Sure. But there have been other slots before and there will be other slots in the future. No idea what Apple has in mind for the future, and the patent application is not limited to PCIe.

might as well be custom slots then since Apple will not write drivers for Windows or Linux so you can't use them in desktop PCs.

 

[Cmaier]

m

might as well be custom slots then since Apple will not write drivers for Windows or Linux so you can't use them in desktop PCs.

Exactly. At least for graphics. There are of course pcie cards for other purposes that don’t need special drivers.

 

[B01L]

Future Mac Pro may use Apple Silicon & PCI-E GPUs in parallel | AppleInsider

Despite Apple Silicon currently working solely with its own on-board GPU cores, Apple is researching how to support more options, like PCI-E GPUs, all working in tandem.

appleinsider.com

From "The Other Place"...

Me, I am still hoping for add-in ASi GPGPUs...

 

[Joelist]

Gurman has been wrong a lot of late. I have a strange feeling there won't be a Mac Pro because the Mac Studio essentially occupies the same slot especially with the Ultra iteration of the M Series.

 

[Colstan]

From "The Other Place"...

deconstruct's essays are always interesting, I'll give him that.

Me, I am still hoping for add-in ASi GPGPUs...

Okay, I'll bite. You've been a big proponent for Apple designing, manufacturing and releasing GPGPUs, @B01L, yet you've never explained why. That's a rather specific niche, one which I don't foresee Apple tackling. They're having a hard enough time justifying the existence of a Mac Pro at all, let alone accelerators that would likely ship in even lower volume. As of yet, the rumor mongers have been bone dry on the existence of such a creature.

Unless you are in a specific field which can take advantage of such a thing, the average user would have no use for a shiny gewgaw like that. Apple Silicon has already integrated everything that the vast majority of Mac users need, and the fruit company can continue to integrate with impunity given their access to the latest nodes and near endless coffers. (Example: rumored ray tracing support.) The one limiting factor is engineering resources. Talented engineers don't grow on fig trees, and they sometimes leave to go into other fields, like patent law. I think those talents would be better spent on designing the next generation of Apple Silicon, which can be used in the entire Mac line.

So, regarding my question, is there a specific workload that you personally need an Apple GPGPU for?

 

[B01L]

Okay, I'll bite. You've been a big proponent for Apple designing, manufacturing and releasing GPGPUs, @B01L, yet you've never explained why.

I have, but I will cover it again, from a 3D/DCC perspective...

The current state of raw GPU power in ASi cannot match that provided on the PC (Nvidia) side of things, but it is my understanding that the actual "day-to-day" workflow in the (properly optimized for ASi/Metal) DCC software of choice is pretty darned good...

It seems to be the rendering which falls short, everyone wants the blazingly fast power of the almighty 4090 to crush their render speeds...

Even if a move to the assorted 3nm processes suddenly gives a larger transistor budget to supercharge the ASi GPU performance to 4090 levels, this would probably be with an "Extreme" SoC configuration, so no GPU grunt beyond the SoC...

Then the 5090 is released and the PC gang berates Apple again for poor GPU performance...

All this clamoring for GPU performance is basically centered around render speed tests, not real world workflow within the DCC software of choice, so why not a division of labor...?

iGPU (included with ASi Mac Pro purchase) for display output and a focus on viewport response...

GPGPU (an optional BTO add-in) for sending render jobs for processing...

That's a rather specific niche, one which I don't foresee Apple tackling. They're having a hard enough time justifying the existence of a Mac Pro at all, let alone accelerators that would likely ship in even lower volume. As of yet, the rumor mongers have been bone dry on the existence of such a creature.

Anyone needing PCIe slots for things other than discrete GPUs, and needing said slots for things that TB4 bandwidth cannot handle, these are the folks needing a Mac Pro...

Networking (25GbE & up), high-speed RAID storage (up to eight M.2 NVMe SSDs on a card), 8k video I/O cards; these are all things that want more bandwidth than TB4 provides...?

There are more uses for PCIe than "muh 4090"...

Unless you are in a specific field which can take advantage of such a thing, the average user would have no use for a shiny gewgaw like that.

Hence these GPGPU cards being a BTO option...

Apple Silicon has already integrated everything that the vast majority of Mac users need, and the fruit company can continue to integrate with impunity given their access to the latest nodes and near endless coffers. (Example: rumored ray tracing support.) The one limiting factor is engineering resources. Talented engineers don't grow on fig trees, and they sometimes leave to go into other fields, like patent law. I think those talents would be better spent on designing the next generation of Apple Silicon, which can be used in the entire Mac line.

All those folks who put high-power GPUs in their Intel Mac Pros, or used a high-power GPU in an eGPU enclosure connected to their Intel MacBook Pros, for the express purpose of accelerating things like 3D software performance or DaVinci Resolve performance; these are the folks who would benefit from an ASi GPGPU...

Or even in a networked rack that contains numerous ASi GPGPUs, for the renderfarm crowd...!

I envision a future Pixar, a frontend with a Mn Extreme Mac Pro Cube on the desk of every 3D artist and a massive ASi GPGPU renderfarm on the backend...! ;^p

So, regarding my question, is there a specific workload that you personally need an Apple GPGPU for?

No personal need, last time I was messing with 3D on the Mac was EIAS on a Mac clone; I am just a simple layman who enjoys thinking about 3D stuff from time to time...

 

[Colstan]

I have, but I will cover it again, from a 3D/DCC perspective...

Ah, I must have missed it, sorry about that.

It seems to be the rendering which falls short, everyone wants the blazingly fast power of the almighty 4090 to crush their render speeds...

Which is why most of those folks probably already have a 4090. Or are looking into a Titan class card.

Then the 5090 is released and the PC gang berates Apple again for poor GPU performance...

Which I think Apple is perfectly fine with. They haven't shown much interest in beating Nvidia's best. They haven't used Nvidia since the Mesozoic epoch of the Mac. They could have kissed and made nice with Nvidia if they considered it a "must have" but they didn't. Nerds like us care about this sort of thing, but I think Apple ceded that market long ago when they went AMD exclusive. They've spent the last decade trying to figure out what to do with the Mac Pro, and we're likely getting a version that's going to make the enthusiast crowd (all dozen of them) go ballistic.

Pat Gelsinger derided Apple as a "lifestyle company", but I think Steve Jobs would take that as a compliment. Apple has been focused on consumers for a long time, that's not likely to change.

Here's my thinking: the Mac is a small percentage of Apple's overall revenue, around 10%. Yes, that's still a fortune 100 company by itself, but it's still small in Apple corporate terms. Desktops make up a minority of that total. I'd say laptops are at least 75%, if not more of Macs sold. I've seen professional analysts who think it's closer to 85%.

Looking at the desktop line, the iMac, Mac mini, and Mac Studio are the lion's share of desktops sold. Over at the other place, they act like the Mac Pro sells volume in the millions. I'd think it's more like 0.1 million, if that many, and it's reducing further. Mac Pro owners are going to be the primary market for a theoretical GPGPU, perhaps with some around the edges with MacBook Pros and Mac Studios, but I seriously doubt those GPGPU customers would even match the volume of the Mac Pro. I think Apple would rather spend those engineering resources on the M3, M4, and future generations of Apple Silicon, not a side project with a limited audience. We've all heard about the brain drain that may or may not have happened within the fruit company, but regardless, smart engineers aren't limitless.

My perspective is that of the bean counters. They're cold, hard, calculating. Apple likes to "surprise and delight" their customers at their fancy events, but they're still a corporation, like any other, and have to prioritize. So, I understand your argument, I just don't see it happening, given the current market which Apple serves. It's an interesting thought experiment, I'll certainly give you that.

All those folks who put high-power GPUs in their Intel Mac Pros, or used a high-power GPU in an eGPU enclosure connected to their Intel MacBook Pros, for the express purpose of accelerating things like 3D software performance or DaVinci Resolve performance; these are the folks who would benefit from an ASi GPGPU...

To quote @Cmaier, "I think the Mac Pro is one bad meal away from being canceled". I wouldn't be surprised if the Apple Silicon version is the last hurrah of the beast, created out of obligation more than anything else. It's kinda like how Apple switched the Xserve over to Intel, even though nobody was buying them, before Steve Jobs lead it to the gallows. It's been a slow process, but the "lifestyle company" is becoming more consumer focused, as time marches on. I think GPGPUs are the inverse of that trend.

No personal need, last time I was messing with 3D on the Mac was EIAS on a Mac clone; I am just a simple layman who enjoys thinking about 3D stuff from time to time...

Certainly nothing wrong with that. I'm not in the market for an Apple Silicon Mac Pro, but find it infinitely interesting, because of the possibilities. As we approach a release date, I'm more convinced that we'll collectively respond with "that's it?". My interest is more in how much it will inform us of Apple's plans for the Mac moving forward, since high-end tech typically trickles down the lineup. Part of that includes whether GPGPUs will be part of that future, but I have yet to see any indication of that, and I don't really see a sizable market for a custom Apple solution.

Regardless, I appreciate hearing your reasoning and the resultant discussion.

 

[B01L]

@Colstan

Personally, I would pony up for a Mn Extreme Mac Cube; plenty of processing power in an elegant Cube chassis to see me thru the NeXT decade or so...! ;^p

 

[Colstan]

@Colstan

Personally, I would pony up for a Mn Extreme Mac Cube; plenty of processing power in an elegant Cube chassis to see me thru the NeXT decade or so...! ;^p

The cube was an awesome idea, if flawed in execution. I think the Mac Studio is the closest we're going to get these days. There's been some speculation that the Mac Studio is a "one and done" project, while waiting for the Mac Pro, but I believe that thinking is incorrect. It fits a hole in the headless desktop line, particularly in the price gap between the M2 Pro Mac mini and the Mac Pro.

My guess, and this is purely a guess, is that the reason that the Mac Studio hasn't received an upgrade to the M2-series is because Apple wants the Mac Pro to have every advantage it can get, particularly if it's going to use an M2 Ultra, and not an M2 "Extreme", assuming such a thing ever existed outside of Gurman's articles.

As I said, a lot of unanswered questions. If Ross Young is right about a 15-inch MacBook Air going into production, which means it should be released reasonably soon, then there's a good chance that the Mac Pro will be announced alongside it. Hopefully, that means that we won't have to wait much later than March or April to get the skinny. At the latest, June, but I'm hoping sooner.

 

[Yoused]

Hence these GPGPU cards being a BTO option...

I believe you are failing to grasp what GPGPU means. It is not different hapdware, it is leveraging the immensely wide power of the GPU onto embarrassingly parallel workloads. The GPU is already a GPGPU. "CUDA" is mostly just an api, and perhaps a small amount of support logic, for feeding large worksets into the GPU, which is already built for that. All Apple really needs is a good upgrade to Metal 4.

 

[B01L]

The cube was an awesome idea, if flawed in execution. I think the Mac Studio is the closest we're going to get these days. There's been some speculation that the Mac Studio is a "one and done" project, while waiting for the Mac Pro, but I believe that thinking is incorrect. It fits a hole in the headless desktop line, particularly in the price gap between the M2 Pro Mac mini and the Mac Pro.

LOL, I argued for quite some time about how a Mn Pro Mac mini was needed to "fill the gap" between a Mn Mac mini and a Mn Max Mac Studio; MANY disagreed, but here we are with the M2 Pro Mac mini...

But once we get Mn Ultra & Extreme Mac Pros, there will be a gap in the headless desktop line-up; being a machine wanted by those with a need for the most compute power available, but without a need for PCIe slots; this is where the Mn Extreme Mac Cube would come into play...

	Mac mini​	Mac Studio​	Mac Cube​	Mac Pro​
Mn	*​
Mn Pro	*​
Mn Max		*​
Mn Ultra		*​		*​
Mn Extreme			*​	*​

My guess, and this is purely a guess, is that the reason that the Mac Studio hasn't received an upgrade to the M2-series is because Apple wants the Mac Pro to have every advantage it can get, particularly if it's going to use an M2 Ultra, and not an M2 "Extreme", assuming such a thing ever existed outside of Gurman's articles.

I feel that the main reason to "hold back" the M2 Max/Ultra ?Mac Studios would be to "force" sales of the M2 Ultra-based Mac Pro...

I believe there will be a Mn Extreme at some point, and I feel Apple was really counting on the 3nm processes being available in their initial planning...

As I said, a lot of unanswered questions. If Ross Young is right about a 15-inch MacBook Air going into production, which means it should be released reasonably soon, then there's a good chance that the Mac Pro will be announced alongside it. Hopefully, that means that we won't have to wait much later than March or April to get the skinny. At the latest, June, but I'm hoping sooner.

If no M2 Ultra Mac Pro in March or April, then maybe Apple debuts the M3 Ultra/Extreme Mac Pro at WWDC this year...?!?

If the latter, then a press release launch for the M2 Max/Ultra Mac Studios right before WWDC...

I believe you are failing to grasp what GPGPU means. It is not different hapdware, it is leveraging the immensely wide power of the GPU onto embarrassingly parallel workloads. The GPU is already a GPGPU. "CUDA" is mostly just an api, and perhaps a small amount of support logic, for feeding large worksets into the GPU, which is already built for that. All Apple really needs is a good upgrade to Metal 4.

I am seeing a GPGPU as a GPU with no display output, more a General Purpose device for handling "blind" compute/render tasks...

 

[Yoused]

I am seeing a GPGPU as a GPU with no display output, more a General Purpose device for handling "blind" compute/render tasks...

The PowerVR architecture used in the M-series consists of a cluster of compute units that handle SIMD instructions, with predication, in a massively parallel way. In essence, it is a bunch of "GPGPU" compute cores, with a separate display output generator which is not really directly connected to the compute cores. If you want the GPGPU power, you just add more GPU compute cores, because that is what they are. For example, if you wanted to add realtime gravity to Celestia, you would do it in the GPU. If Metal lacks the proper tools for implementing GPGPU compute on AS, that is a huge deficiency in the API that Apple needs to work on (I was given to understand that Metal was meant to converge OpenGL and OpenCL capabilities in an improved design).