I keep asking myself "why a Mac Pro in the Apple Silicon age"?

[throAU]

Networking is already built in. So are Audio and Video (in fact AS has specific hardware video accelerators - basically every M series except for the original M1 has built in Afterburner). Raid cards are interesting but again the storage you can get on card is already pretty huge.

What about the next type of card that doesn’t yet exist?

Without a high speed bus to the cpu, it can’t be invented.

And thunderbolt is great but it isn’t anywhere close to the full pcie5 x16 spec.


Also. Built in 10 gig is ok but the processing power in a new pro will be enough that we may want 100 gig networking or faster to access massive data sets on enterprise SAN storage - to get data in and out of it fast enough for the compute hardware to not be sitting idle waiting. Just as one trivial example.

10 gig networking isn’t fast in 2023 and even 1-2 consumer SSDs will outrun it by an order of magnitude.

Give it two years and single consumer SSDs will keep up with 100 gig Ethernet.

We need slots at this end of the market.

If the new pro is going to be aimed at AR/VR content creation I can easily see the bandwidth requirement. I know carmack reckons something like 16k per eye at high frame rate is required as the end game for actually good VR and we aren’t there yet.

The bandwidth needs are huge.

 

[Nycturne]

What about the next type of card that doesn’t yet exist?

Without a high speed bus to the cpu, it can’t be invented.

And thunderbolt is great but it isn’t anywhere close to the full pcie5 x16 spec.


Also. Built in 10 gig is ok but the processing power in a new pro will be enough that we may want 100 gig networking or faster to access massive data sets on enterprise SAN storage - to get data in and out of it fast enough for the compute hardware to not be sitting idle waiting. Just as one trivial example.

10 gig networking isn’t fast in 2023 and even 1-2 consumer SSDs will outrun it by an order of magnitude.

Give it two years and single consumer SSDs will keep up with 100 gig Ethernet.

We need slots at this end of the market.

If the new pro is going to be aimed at AR/VR content creation I can easily see the bandwidth requirement. I know carmack reckons something like 16k per eye at high frame rate is required as the end game for actually good VR and we aren’t there yet.

The bandwidth needs are huge.

I would probably say what while the top end is still pushing into new frontiers, the % of computer users that need it is shrinking. Different groups of users are taking offramps as certain things stop being the bottlenecks in their process. This creates some issues for striking good balance for a company like Apple where they aren’t a staple in things that are I/O intensive and don’t seem terribly interested in claiming or retaking those markets.

The engineering teams I’ve worked with have gone from grabbing the beefiest Mac Pros we could budget to the Studio. So it’s not like you can count on them to help feed the Mac Pro beast these days unless they have specific needs. Same with the audio crowd as there’s a good selection of USB and Thunderbolt audio interfaces that don’t suck. Our work environment can’t even take advantage of 10Gbit networking at the moment as their isn’t enough I/O demand when you are just pushing code around to justify upgrading the switches and backbone to support it.

Heck, just give me a M2 Max MacBook Pro and a good dock and I am still more productive than I would be on an Intel Mac Pro at double the cost. And now I don’t need a second machine for meetings/demos.

Not saying that such a thing shouldn’t exist, but rather the justification to bean counters gets harder the more other devices in the lineup can cover the use cases for less money. And it sucks, because I absolutely think that a Mac Pro is a key piece of keeping a small but vital segment of the Mac user base around.

 

[throAU]

I would probably say what while the top end is still pushing into new frontiers, the % of computer users that need it is shrinking.

Definitely.

However, for those in those fields that require compute, the performance demand is there.

Additionally, for anyone waiting for the machine to do anything the demand is there, it's just a sliding scale of need vs. cost.

 

[Joelist]

I would probably say what while the top end is still pushing into new frontiers, the % of computer users that need it is shrinking. Different groups of users are taking offramps as certain things stop being the bottlenecks in their process. This creates some issues for striking good balance for a company like Apple where they aren’t a staple in things that are I/O intensive and don’t seem terribly interested in claiming or retaking those markets.

The engineering teams I’ve worked with have gone from grabbing the beefiest Mac Pros we could budget to the Studio. So it’s not like you can count on them to help feed the Mac Pro beast these days unless they have specific needs. Same with the audio crowd as there’s a good selection of USB and Thunderbolt audio interfaces that don’t suck. Our work environment can’t even take advantage of 10Gbit networking at the moment as their isn’t enough I/O demand when you are just pushing code around to justify upgrading the switches and backbone to support it.

Heck, just give me a M2 Max MacBook Pro and a good dock and I am still more productive than I would be on an Intel Mac Pro at double the cost. And now I don’t need a second machine for meetings/demos.

Not saying that such a thing shouldn’t exist, but rather the justification to bean counters gets harder the more other devices in the lineup can cover the use cases for less money. And it sucks, because I absolutely think that a Mac Pro is a key piece of keeping a small but vital segment of the Mac user base around.

This is what I was thinking too. The Studio especially directly addresses a large part of the former Pro audience. And with Apple Silicon unless there is enough of an audience it becomes unprofitable to set up such small fab runs of silicon except at really high price points.

 

[Yoused]

Seems to me that if the Studio covers the needs of a large fraction of the old Mac Pro market (and ISTR hearing that it is extremely difficult to saturate an Ultra GPU), the sensible approach would be for Apple to establish the Honeycrisp Agency, devoted to producing bespoke Apple Silicon products. Honeycrisp can build you a one-board computer with all that stuff that you would have in cards, in a handy 1 or 2U form factor, which will be faster and more reliable than plugging in the shit yourself. With AppleCare, on the house. Kind of along the lines of how IBM sells POWER systems, but on a smaller scale. Because, manufacturing a pico-niche product just no longer makes sense.

 

[throAU]

I think that if you need significantly more than a Mac Studio the next step may end up being a rack in a different room. If you’re in that ultra high tier market segment why piss about with desktop hardware at that point. At that stage you’re in the “I will buy whatever you can make” sort of category. Maybe use a studio as the remote terminal for something properly powerful.

Rack mount it, hook it up directly to a flash storage array, run big noisy fans on it and run it as hard as you can somewhere away from your desk in an air conditioned room.

I’m taking about something say 4x RU with SOC slots into a common backplane. You just add m2 or m3 ultras to it as cards which contain their local onboard 512GB of unified memory. Connected to a common bus so they can work as a single multi processor system with TBs of memory and thousands of GPU cores.


Edit : or even more likely - like the honey crisp suggestion above. Apple just cloud host it for you. You just supply a big enough internet pipe and apple look after it. Storage management, backups, etc. the lot.

You just interact with it via the studio on your desk.

 

[Citysnaps]

For people needing slots, a fast bus, and/or rack mounting, MacPro, especially one that's AS-fueled, is still the answer.

 

[throAU]

For people needing slots, a fast bus, and/or rack mounting, MacPro, especially one that's AS-fueled, is still the answer.

For now, but that platform is old and Apple could/should build something much better.

 

[Citysnaps]

For now, but that platform is old and Apple could/should build something much better.

I didn't mean to suggest the current rack mount MacPro couldn't evolve.

But it would still need slots, a fast bus that current cards could use (PCIe), and be in a rack mount chassis to accommodate people who need that capability.

At the moment I personally don't see the current rack mount MacPro being deficient in any area (other than being Intel fueled). If anything, the 1.4 kilowatt power supply could be trimmed back now that power-efficient Apple Silicon processors are available.

Maybe there'll be an announcement at the upcoming WWDC.

 

[Yoused]

But it would still need slots, a fast bus that current cards could use (PCIe), and be in a rack mount chassis to accommodate people who need that capability.

I think they will go with "you tell us what you need and we will make it." The SoC/SSD will sit at the heart of a "cakeboard" (like a breadboard, only sweeter) and the Pro division will assemble the functionality that you used to add with cards right on the board, with a DMA protocol controller (not PCIe). It will all fit in a 1U, because there will be no need for card clearance. And, of course, it will be spendy (like three Studio Ultra base pricing, but still cheaper than big iron).

 

[Citysnaps]

I think they will go with "you tell us what you need and we will make it." The SoC/SSD will sit at the heart of a "cakeboard" (like a breadboard, only sweeter) and the Pro division will assemble the functionality that you used to add with cards right on the board, with a DMA protocol controller (not PCIe). It will all fit in a 1U, because there will be no need for card clearance. And, of course, it will be spendy (like three Studio Ultra base pricing, but still cheaper than big iron).

I'm scratching my head on that, not following what you're saying. And at the same time I'm intrigued.

Who is the "us" who will be doing the systems engineering, designing, and "assembling the functionality?" Currently there are a large array of PCIe cards that offer a wide range of functionality to address different needs over a wide range commercial/industrial/research/aerospace/etc disciplines, at different levels of performance, number of channels (analog and digital), etc. The flexibility offered, while not limitless, is great.

As an example, let's say I want to design a system in a rack mount chassis that would sample a 50 MHz wide bandwidth RF/IF analog spectrum coming from a site output...say, with a shielded (to suppress digital noise) 14-16 bit 150 MSPS A/D converter, which then drives a number of independently digitally tunable channels (say 8) of digital tuning (ie downconversion to baseband producing I and Q digital samples for each of the 8 channels), followed by digital filtering and decimation, and finally demodulation and distribution (storage, to other systems, etc) for each of the 8 channels.

The above is an example of what can be done with a handful of off-the-shelf PCIe cards for one particular (aerospace) discipline. There are many others.

 

[throAU]

What I envision or would like to see is a backplane (like used in high end chassis switches) with slots for SOC cards (i.e., add as many Mx-Ultra boards as you like) and other slots for general PCIe IO. e.g., maybe up to 4 or 8 SOC cards each with an Mx-Ultra and 512 GB of RAM on each, that all act as a single unified platform via the high speed backplane board.

Rack/stack them as you see fit

 

[leman]

I think they will go with "you tell us what you need and we will make it." The SoC/SSD will sit at the heart of a "cakeboard" (like a breadboard, only sweeter) and the Pro division will assemble the functionality that you used to add with cards right on the board, with a DMA protocol controller (not PCIe). It will all fit in a 1U, because there will be no need for card clearance. And, of course, it will be spendy (like three Studio Ultra base pricing, but still cheaper than big iron).

I have difficulty envisioning how something like this would work. For example, what advantage is there is developing a custom protocol instead of PCIe? There are a lot of add-on cards on the market with all kinds of functionality, it’s not feasible or reasonable for Apple to replicate even a popular portion of that functionality.

 

[Yoused]

A card is a copse of chips that do some stuff and a signal connection or two. They would be able to do the same stuff whether on the main board or on the card – the PCIe connection is just an extra interface for the modularity part.

And as much as I like modularity, Apple is moving away from it. I can choose to embrace one or the other – you might be able to guess which one I will give up.

A Mac Pro main board would consist of an SoC or two, an SSD and those other little signal balancing things. On a standard rack board, that leaves around two thirds of the board blank, after you put in juice and cooling.

The Mac Pro customer service division would take an order for "I need a Mac that does (this and this and this)", acquire the chips that would be on the boards, use some sort of tool to map them out, scribe it with a leads plotter and finish it with a chip setter, attaching the signal connections at the edge.

The chips that would have been on cards are now right on the main board. There is a controller chip over near the SoC that manages signaling and DMA, probably to its own RAM buffer that the SoC can access directly. All the intervening PCI overhead is gone, other than how the controller can make driver software see a card there, if necessary.

If the customer can get all the functionality they need without cards, and it is within their budget, why would they yearn for cards? A tidy single board that fills their needs sounds appealing. The idea may seem fanciful, but it is consonant with Apple's current MO.

 

[Nycturne]

I have to agree with Leman on this one, I’m not sure why you would replace DMA-based PCIe with a DMA-based custom protocol, at least for peripheral chips. Maybe you could beat PCIe on overhead with enough work, but it’s not a guarantee, plus its now one more thing you have to own rather than letting a SIG work on it for you. And it also means you can’t use off-the-shelf peripheral controller chips that use PCIe unless you bridge it to the custom protocol, which just adds extra overhead. PCIe 5.0 is overkill for many peripherals as it is outside of SSDs, which is also true of PCIe 4.0 today. So it’s not like PCIe has a bandwidth issue at the moment or is a bottleneck.

Just because a peripheral shares the same board doesn’t mean you can’t or don’t want to use PCIe for peripheral interaction. Apple’s interaction with peripheral chips on Apple Silicon relies on PCIe today. And if the goal is to allow folks that need a Mac Pro to have their peripherals baked onto the PCB with the SoC, all the more reason to not require custom FPGA/ASIC work and just use existing PCIe-based peripheral chips like Apple does for the Mini/Studio/etc.

 

[Citysnaps]

The Mac Pro customer service division would take an order for "I need a Mac that does (this and this and this)", acquire the chips that would be on the boards, use some sort of tool to map them out, scribe it with a leads plotter and finish it with a chip setter, attaching the signal connections at the edge.

I just don't see that happening as the range of possibilities of problems a MacPro could solve with slots is immense. That would require Apple to employ a cadre of systems engineers with the expertise to understand the requirements and hardware engineers to craft the solution, all well-versed in a wide array of applications and technologies in analog/digital and signal processing domains.

The digital tuner example I posted above (having system engineered and designed them and other related signal processing systems in the past when I worked in defense/aerospace) is just one application of likely hundreds/thousands of potential uses. A very tiny sliver of the whole.

Would Apple have the resources to handle a few hundred or more of these semi-custom designs? In a timely manner? Would it be a good business case for Apple? That's very questionable. Much easier to create a rack mount MacPro with slots - it's what they're good at, involves no risk, and is easy to service.

At the time, years ago, I designed my own signal acquisition and processing cards. Today I'd probably use off-the-shelf cards, but still might design my own (if I were still in the business). In either case I'd need a high performance bus and a rack mount MacPro.

If Apple decided to go in the direction of creating semi-custom solutions, would they have the expertise to offer rapid tech support when needed? Or upgrade the software/hardware when needed as future better performing options became available? That's pretty easy with cards.

I would also be very reluctant (more likely never happen) essentially subcontracting and trusting another company taking my top-level block diagram and implementing a solution that meets requirements. Would they know about certain tricks in implementing certain signal processing functions that would translate to faster/better performance (thus providing a competitive edge)? Would I want to share that with another party? Not likely.

Not wanting to be reflexively negative to the idea, I still see a lot of issues. And not many pluses going in that direction.

 

[Cmaier]

I just don't see that happening as the range of possibilities of problems a MacPro could solve with slots is immense. That would require Apple to employ a cadre of systems engineers with the expertise to understand the requirements and hardware engineers to craft the solution, all well-versed in a wide array of applications and technologies in analog/digital and signal processing domains.

The digital tuner example I posted above (having system engineered and designed them and other related signal processing systems in the past when I worked in defense/aerospace) is just one application of likely hundreds/thousands of potential uses. A very tiny sliver of the whole.

Would Apple have the resources to handle a few hundred or more of these semi-custom designs? In a timely manner? Would it be a good business case for Apple? That's very questionable. Much easier to create a rack mount MacPro with slots - it's what they're good at, involves no risk, and is easy to service.

At the time, years ago, I designed my own signal acquisition and processing cards. Today I'd probably use off-the-shelf cards, but still might design my own (if I were still in the business). In either case I'd need a high performance bus and a rack mount MacPro.

If Apple decided to go in the direction of creating semi-custom solutions, would they have the expertise to offer rapid tech support when needed? Or upgrade the software/hardware when needed as future better performing options became available? That's pretty easy with cards.

I would also be very reluctant (more likely never happen) essentially subcontracting and trusting another company taking my top-level block diagram and implementing a solution that meets requirements. Would they know about certain tricks in implementing certain signal processing functions that would translate to faster/better performance (thus providing a competitive edge)? Would I want to share that with another party? Not likely.

Not wanting to be reflexively negative to the idea, I still see a lot of issues. And not many pluses going in that direction.

Not to mention that Apple would never do it because of patents. They make a thing for you and the thing infringes a patent, so they get sued. What then? You’d end up having to enter into the sort of contract where you indemnify them for everything. So now you have a $100 million potential liability if apple gets sued (and they get sued a lot).

 

[Citysnaps]

Not to mention that Apple would never do it because of patents. They make a thing for you and the thing infringes a patent, so they get sued. What then? You’d end up having to enter into the sort of contract where you indemnify them for everything. So now you have a $100 million potential liability if apple gets sued (and they get sued a lot).

Perhaps that outcome could also extend to a MacPro, augmented by Apple based on design requirements and order by a 2nd party (say me or a company), and then somehow in some manner causing damage to a 3rd party (another company) who purchased the augmented Apple MacPro from the 2nd party. Not too likely, maybe, but I'd guess Apple would not want to shoulder that potential risk.

Related to that... I can see situations where people or companies engaging Apple to custom augment a MacPro for someone according to a set of requirements and specs, not fulfilling that someone's expectations. Either due a poorly written spec/requirements without enough (or with ambiguous) detail, or misinterpretation by Apple, or Apple not having the right skill set for the job, or whatever. And ultimately not meeting the requirements that someone had in mind. Resulting in an unhappy customer after likely spending thousands of $ for customization, perhaps blaming Apple.

 

[B01L]

So y'all want custom one-off bespoke motherboards...?

Or are we talking a motherboard with all the bells & whistles packed onboard...?

Both would be exceedingly expensive, and the latter would be ridiculous for those who do not need all the bells & whistles...

Standard motherboards with PCIe slots would be the most cost/market efficient...