News Apple Is Struggling to Build Mac Pro Based on Its Own Silicon: Report

[Admin]

Apple's own SoC-based Mac Pro is behind the schedule as company considers moving its production to Vietnam.

Apple Is Struggling to Build Mac Pro Based on Its Own Silicon: Report : Read more

 

[PlaneInTheSky]

Mac Pro systems are often used for cinema and video production, and such workloads are getting more demanding as resolutions and color depths increase.

When did resolution increase. Content has never moved on from 4k because it is pointless. I think you can count the movies shot in native 8k on one hand.

 

[Glock24]

An Epyc/Threadripper workstations would be cheaper, support more RAM and offer a ton more flexibility and upgradeability. Unless you rely on Apple software, then there's no point in this Mac Pro.

 

[bit_user]

A Mac Pro based on the alleged M2 Extreme processor would cost around $10,000

Adjusting for inflation, isn't that close to the starting price of the existing Mac Pro, when it launched?

I'm exaggerating, but not by much. I also seem to recall the fully-configured price of a Mac Pro was like $40k or $45k, back then.

Anyway, I think a bigger issue might be that even 384 GB isn't enough memory for some workstation applications. CXL.mem would be a good option for expandability, but it'd required a lot of software work to enable.

One of the features that Apple's Mac Pro has — and something that the company's other systems lack — is upgradeability. A user of previous Mac Pros would often buy the tower, then install a new graphics card, add more memory, or Apple's Afterburner accelerator. It is unclear whether an Apple Silicon-based desktop would be upgradeable, but from what we see with the Apple Studio machine, the company is reluctant to offer such capability even to its professional customers.

Well, Mac Minis have never had good upgradability, I think. But, the main issue is probably how many PCIe lanes the M2 Max has, and what speed. If it had 16, then a quad-chip configuration could get you 64 lanes.

the current-generation machine was launched in 2019.

...using a special Cascade Lake variant, with 64 PCIe 3.0 lanes instead of the normal 48 you'd get in that socket. However, it only had up to 28 cores. Perhaps they regretted not going with ThreadRipper, which would've given them PCIe 4.0 and up to 64 cores.

 

[bit_user]

An Epyc/Threadripper workstations would be cheaper, support more RAM and offer a ton more flexibility and upgradeability. Unless you rely on Apple software, then there's no point in this Mac Pro.

I agree with you in terms of what I'd buy. However, there's no denying that the current Mac Pro has superior cooling, built-in Thunderbolt, and the GPUs have an over-the-top link that even the normal AMD Pro cards don't support.

https://www.apple.com/shop/product/MW732AM/A/radeon-pro-vega-ii-duo-mpx-module

 

[cia1413]

When did resolution increase. Content has never moved on from 4k because it is pointless. I think you can count the movies shot in native 8k on one hand.

The resolution of what you render down to to hand off to people is 4k sure, but cameras are now available and commonly used up to 12k raw files. The editors use the extra space to zoom in and reframe and special fx guys always want ultra high rez to do super special fx stuff to.

 

[SSGBryan]

I was running Mac Pro's for 15 years (and Power Macs before then).

The clowns running Apple never did understand the Mac Pro user base. The 1,1 to 5,1 Mac Pro (the original cheesegrater) was an incredible general purpose workstation. It was price and performance competitive with Windows based workstations. It could handle anything I threw at it (3D art - Blender, Zbrush, Poser, Hexagon, Shade, Carerra, Vue, etc). Creatives that needed horsepower loved those things.

Then we got the trashcan. Overpriced, and I would have to add $2,000 to the price to replace the functionality that Sir Idiot Boy yanked out of it, so Apple could have a Cube 2.0. It could only do 1 thing well - video production (up until the video cards started dying by the bucketload). And no way to replace the video card BECAUSE THE VIDEO ROM WAS ON THE MOTHERBOARD, not the video card. Could max the CPU or the GPU, but not both, because Sir Idiot Boy didn't understand the customer base didn't care what it looked like - we were busy creating.

Over 2,000 days later (yep, that is how long we were stuck with that PoS.) We got another overpriced PoS that was thoroughly obsolete the day it launched. Every single subsystem had been superseded. Video, IO, storage, everything.

The base model (@ $6,000USD - before the $400 non locking wheels) could be outperformed by a $1,200USD Ryzen based system.

If Timmy & Sir Idiot Boy had just said in their 2017 apology tour, We are leaving the PC space, because we can't compete there. I could have left 2 years earlier and jumped on the Ryzen train then, rather than waiting till that PoS was released.

The only people buying the 7,1 are video folks that are too scared to leave the Apple prison.

Apple users really don't understand how far behind they are. But hey, they can now use their iFart apps on their "desktop" systems.

 

[bit_user]

If Timmy & Sir Idiot Boy had just said in their 2017 apology tour, We are leaving the PC space, because we can't compete there. I could have left 2 years earlier and jumped on the Ryzen train then, rather than waiting till that PoS was released.

Uh, so Timmy is "Tim Apple" and Sir Idiot Boy is Johnny Ives? If so, at least he's gone, right?

P.S. thanks for your post. Interesting insights, as I've not touched a Mac in over 20 years.

 

[thestryker]

When I see stories like this I can't help but wonder if Apple's loss of talent to Nuvia (now at Qualcomm) has hurt them for this space. They seem to keep trying to scale what they have rather than do a proper new design and that doesn't seem particularly smart.

 

[systemBuilder_49]

Apple is realizing it will take additional time to train its sheeple to buy a non-upgradeable workstation(s). Until they can train them, they will be forced to offer workstations based on PCI bus and unsoldered ram which can be upgraded. Also the aftermarket for mac workstations will collapse with M2 cripplestations but that's not apple's concern.

 

[Kamen Rider Blade]

When I see stories like this I can't help but wonder if Apple's loss of talent to Nuvia (now at Qualcomm) has hurt them for this space. They seem to keep trying to scale what they have rather than do a proper new design and that doesn't seem particularly smart.

How many CPU Engineers did they losoe to Nuvia?

 

[PlutoDelic]

I was running Mac Pro's for 15 years (and Power Macs before then).

The clowns running Apple never did understand the Mac Pro user base. The 1,1 to 5,1 Mac Pro (the original cheesegrater) was an incredible general purpose workstation. It was price and performance competitive with Windows based workstations. It could handle anything I threw at it (3D art - Blender, Zbrush, Poser, Hexagon, Shade, Carerra, Vue, etc). Creatives that needed horsepower loved those things.

Then we got the trashcan. Overpriced, and I would have to add $2,000 to the price to replace the functionality that Sir Idiot Boy yanked out of it, so Apple could have a Cube 2.0. It could only do 1 thing well - video production (up until the video cards started dying by the bucketload). And no way to replace the video card BECAUSE THE VIDEO ROM WAS ON THE MOTHERBOARD, not the video card. Could max the CPU or the GPU, but not both, because Sir Idiot Boy didn't understand the customer base didn't care what it looked like - we were busy creating.

Over 2,000 days later (yep, that is how long we were stuck with that PoS.) We got another overpriced PoS that was thoroughly obsolete the day it launched. Every single subsystem had been superseded. Video, IO, storage, everything.

The base model (@ $6,000USD - before the $400 non locking wheels) could be outperformed by a $1,200USD Ryzen based system.

If Timmy & Sir Idiot Boy had just said in their 2017 apology tour, We are leaving the PC space, because we can't compete there. I could have left 2 years earlier and jumped on the Ryzen train then, rather than waiting till that PoS was released.

The only people buying the 7,1 are video folks that are too scared to leave the Apple prison.

Apple users really don't understand how far behind they are. But hey, they can now use their iFart apps on their "desktop" systems.

Had a mate who rode this train. Inbetween, i made him a Hackintosh. His thoughts on all this is that he preferred the fear of the unknown Hackintosh trouble to the limits he's feeling with the current cheese grater. Mainly for performance.

Apple has divorced this user base, and developers have followed through, many jumping to Windows/PC.

Of course it's funny in the long run for Apple, they're not losing any money about it, it feels like a hobby that lost interest to say the least.

Imagine a fully open sourced macOS x86 or licensed to other OEMs. A grown up ma can still dream of (impossible) solutions i guess.

 

[bit_user]

When I see stories like this I can't help but wonder if Apple's loss of talent to Nuvia (now at Qualcomm) has hurt them for this space.

I think their pace of innovation in their cores has definitely slowed, and that's a shame. I'd like to see them with ARMv9-A and especially SVE2.

They seem to keep trying to scale what they have rather than do a proper new design and that doesn't seem particularly smart.

The M1 Ultra was pretty amazing, though. Just scaling that up 4-ways would be enough. in terms of CPU and GPU performance. As I mentioned, the tough issues are going to be scaling PCIe and memory capacity. Adding some CXL links could potentially solve both in one go, because CXL supports switching (though maybe not until 2.0?).

 

[bit_user]

How many CPU Engineers did they losoe to Nuvia?

I think the issue was less about the absolute number and perhaps more the fact that it was some of the most senior and elite veterans who left. That's just the sense I got, but maybe it's wrong.

I can understand getting bored of working at a place or maybe disliking your boss, but the sense I got was that they were primarily motivated by greed. Given how Apple's stock has performed, all of them must've already been multi-millionares. But, maybe that counts as middle-class, in Sillicon Valley.

We should also consider that they might've lost other engineers to AI & other chip startups.

 

[Coffee Fueled Curmudgeon]

Uh, so Timmy is "Tim Apple" and Sir Idiot Boy is Johnny Ives? If so, at least he's gone, right?

I assume Sir "form over function" Jonny Ives. He may be gone, but his practices are very much present in Timmy.

 

[crimsonfilms]

I was running Mac Pro's for 15 years (and Power Macs before then).

The clowns running Apple never did understand the Mac Pro user base. The 1,1 to 5,1 Mac Pro (the original cheesegrater) was an incredible general purpose workstation. It was price and performance competitive with Windows based workstations. It could handle anything I threw at it (3D art - Blender, Zbrush, Poser, Hexagon, Shade, Carerra, Vue, etc). Creatives that needed horsepower loved those things.

Then we got the trashcan. Overpriced, and I would have to add $2,000 to the price to replace the functionality that Sir Idiot Boy yanked out of it, so Apple could have a Cube 2.0. It could only do 1 thing well - video production (up until the video cards started dying by the bucketload). And no way to replace the video card BECAUSE THE VIDEO ROM WAS ON THE MOTHERBOARD, not the video card. Could max the CPU or the GPU, but not both, because Sir Idiot Boy didn't understand the customer base didn't care what it looked like - we were busy creating.

Over 2,000 days later (yep, that is how long we were stuck with that PoS.) We got another overpriced PoS that was thoroughly obsolete the day it launched. Every single subsystem had been superseded. Video, IO, storage, everything.

The base model (@ $6,000USD - before the $400 non locking wheels) could be outperformed by a $1,200USD Ryzen based system.

If Timmy & Sir Idiot Boy had just said in their 2017 apology tour, We are leaving the PC space, because we can't compete there. I could have left 2 years earlier and jumped on the Ryzen train then, rather than waiting till that PoS was released.

The only people buying the 7,1 are video folks that are too scared to leave the Apple prison.

Apple users really don't understand how far behind they are. But hey, they can now use their iFart apps on their "desktop" systems.

Mac Pro 7,1 (2019) is actually a great design. It has better upgrade potential than the 5,1. More PCIe slots, TB ports, etc. The issue is... the CPU they chose. The Intel Xeon W series is a high-end CPU. I'm not sure why Apple did not choose the Xeon Bronze, Silver and Gold series back then. If you priced out an HP or Lenovo workstation with the same specs, the Mac Pro was better priced.... until you factor in the CPU. The HP can be configured with a Silver 12 core Xeon that was thousands cheaper than the MP 12/14 core.

 

[bit_user]

The issue is... the CPU they chose. The Intel Xeon W series is a high-end CPU. I'm not sure why Apple did not choose the Xeon Bronze, Silver and Gold series back then.

Because they had Intel build a special variant of the Xeon W that repurposed the UPI lanes and replaced them with another 16 PCIe lanes. That's why it has 64 PCIe lanes, instead of the normal 48. It also means you can't just put one of the server CPUs in it.

The UPI lanes are normally used for multi-CPU configurations. In most Xeon W CPUs, they're disabled and do nothing (because those CPUs are limited to single-CPU configurations). That must be where the idea came from to reuse them.

 

[palladin9479]

In other words, Apple can't figure out how to gouge it's faithful congregation enough so is holding off until the time comes where it can. Other vender already offer far superior hardware solutions that are upgradable and aren't vendor locked. If someone really is die hard anti-windows they can just use one of the multitude Linux distributions. The only folks who end up buying these things are those who can't operate outside of MacOS or on anything with a fruit logo.

Here we have Apple's next innovation for an even simpler user experience.

 

[Cbeehlz]

The M1/M2 architecture isn't fast because of some engineering magic - it's fast because it LACKS many of the underpinnings of a pro-level system that require system resources to run. That includes things like PCIe, SCSI, SATA, etc.
When Apple started touting "workstations" I wondered how they were going to get around the fact that there's a hard IO limit with what's available on a single M1 chip. I expected some engineering wizardry that would allow them to add a PCIe bus onto what's essentially a mobile CPU. Nope.
What we got was two CPU's stapled together in the same package - essentially doubling the number of USB/TB ports on the system. Creative, but in the sense of creatively cheating on a exam, not creatively inventing a light bulb or a new kind of golf club.

Apple is all about squeezing as much money as they can out of their clientele while taking as few risks as possible. That means everything is now essentially an iPhone and margins need to stay fat at all costs. If you're a Pro start reading up on Windows 11. You'll be using it sooner than you think if you want to keep working with high end systems.

 

[thestryker]

How many CPU Engineers did they losoe to Nuvia?

It's not so much how many as the who. The people who went over to Nuvia were the ones who wanted to do larger CPUs aimed at professional/server market, but Apple had no interest. So they lost a lot of their senior engineers who were behind their current slate of CPUs.

The M1 Ultra was pretty amazing, though. Just scaling that up 4-ways would be enough. in terms of CPU and GPU performance. As I mentioned, the tough issues are going to be scaling PCIe and memory capacity. Adding some CXL links could potentially solve both in one go, because CXL supports switching (though maybe not until 2.0?).

Performance wise, yes, but it's also an absurdly large chip and there's undoubtedly a lot of wasted silicon compared to if they just built a properly large chip. That market isn't a one size fits all, but that is the strategy behind the entire M line of products so far. CXL doesn't solve their lack of interfaces problem, but I agree it may be the route they choose for memory expansion as it could save board complexity and allow them to fleece customers with proprietary upgrade kits.

 

[bit_user]

The M1/M2 architecture isn't fast because of some engineering magic - it's fast because it LACKS many of the underpinnings of a pro-level system that require system resources to run. That includes things like PCIe, SCSI, SATA, etc.

Just PCIe, USB, and SATA are implemented in today's chipsets, and they don't use all that much die area, in comparison to how big some of the M1 SoCs are.

I expected some engineering wizardry that would allow them to add a PCIe bus onto what's essentially a mobile CPU.

The M1 series aren't "mobile", in the sense of being Phone SoCs. They are indeed laptop-oriented.

What we got was two CPU's stapled together in the same package

"Stapled" isn't the right word. The M1 Ultra's die-to-die interconnect remains the fastest ever created, by far. Way faster than AMD's MI200 series and even their new RX 7000 series. The RX 7000 touts 5.3 TB/s, but that's the aggregate figure for all 6 chiplets.

If we're to believe that the M2 Max can scale to a 4-way configuration, then Apple could solve their PCIe problem simply by adding 16 lanes to each Max.

IMO, the harder problem is expending memory beyond the 384 GB that could conceivably fit in-package.

Apple is all about squeezing as much money as they can out of their clientele while taking as few risks as possible.

The M1 Ultra was fairly ground-breaking, not low-risk.

If you're a Pro start reading up on Windows 11.

I use Linux. Never touched a Mac in more than 2 decades, FWIW.

 

[bit_user]

it's also an absurdly large chip and there's undoubtedly a lot of wasted silicon compared to if they just built a properly large chip.

Most of the die area is for the GPU(s). You have to compare it to a fast desktop CPU + high-end GPU, for the die area to make sense.

That market isn't a one size fits all, but that is the strategy behind the entire M line of products so far.

They have 4 different products, built from 3 different dies: M1, Pro, Max, and Ultra. How is that one-size-fits-all?

 

[thestryker]

Most of the die area is for the GPU(s). You have to compare it to a fast desktop CPU + high-end GPU, for the die area to make sense.

Yes, but it's extraordinarily ineffecient to manufacture huge silicon, and they're adding to it by hooking dies together rather than designing purpose built. Taking a design which was laptop focused then just duplicating it is a great way to waste silicon for things that don't help a product like the pro.

They have 4 different products, built from 3 different dies: M1, Pro, Max, and Ultra. How is that one-size-fits-all?

They're all the exact same thing just more of it as you go up the stack (I think the base model has slower memory). They all carry the same limitations which are hard wired due to the design of the processor. It would be like if AMD/Intel used their mobile APU design only and never changed anything across the entire product stack.

 

[bit_user]

Yes, but it's extraordinarily ineffecient to manufacture huge silicon,

Yield can be managed by having more CPU & GPU cores than they need, and just disabling the ones with defects.

The M1 Max is estimated at 432 mm^2, which is still smaller than the AD102 die in Nvidia's 4090. The latter is not only 609 mm^2 but also uses a smaller process node (TSMC N4).

and they're adding to it by hooking dies together rather than designing purpose built.

The M1 Max was purpose-built for a dual-die configuration. You cannot possibly have such a fast interconnect without the SoC being designed for it, from the ground up.

As for the notion that hooking the dies together is somehow bad, I don't know where you got that. There's already an industry consensus that multi-die is the best way to scale.

Taking a design which was laptop focused then just duplicating it is a great way to waste silicon for things that don't help a product like the pro.

If you think it's a waste, try talking to some Mac Studio owners or reading reviews of it.

They're all the exact same thing just more of it as you go up the stack (I think the base model has slower memory).

Certainly not. You clearly need to educate yourself more, if we're to continue this conversation.

https://www.anandtech.com/show/17024/apple-m1-max-performance-review

 

[palladin9479]

Yes, but it's extraordinarily ineffecient to manufacture huge silicon, and they're adding to it by hooking dies together rather than designing purpose built. Taking a design which was laptop focused then just duplicating it is a great way to waste silicon for things that don't help a product like the pro.


They're all the exact same thing just more of it as you go up the stack (I think the base model has slower memory). They all carry the same limitations which are hard wired due to the design of the processor. It would be like if AMD/Intel used their mobile APU design only and never changed anything across the entire product stack.

The M1 looked fast when it was released because Apple paid TSMC a ton of cash for access to the latest process node before anyone else. Then once everyone else got access to that same node the performance difference vanished. Phone chips work great for phones, tablets and thin notebooks, not so great when performance needs to scale deep instead of wide. Also integrated graphics can never compete with dedicated at scale due to thermodynamics.