Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
New Mac Pro and an 'Extreme' Chip: Here's What the Latest Rumors Say
- Thread starter MacRumors
- Start date
- Sort by reaction score
I think it's likely the case that Apple made a token effort to stick Apple Silicon into the Mac Pro chassis purely to avoid the bad PR of otherwise outright cancelling it (again), particularly as it was basically only relevant for about two years from its release to the release of Apple Silicon. Let's be honest, for 90% of the potential customer base for a machine like this, the most valuable feature would be to allow using NVIDIA GPUs. Given that's never going to happen, the machine is effectively pointless and I would imagine enough time has now passed that they could just quietly sunset it without nearly as much flak as they would've gotten had they done it back in 2021.
Agreed, but my hunch is most manufacturers want to shy away from producing modular products. With those types of products, though of benefit to the consumer, they may not necessarily benefit the manufacturer's bottom line to the same extent, so modularity is not prioritised.
In consolation, at least the new Mac Pro seems that it will be fast and powerful. For professionals, that's got to count for something.
When Apple releases the MacPro with an M5 processor, they will be building on my work: Twenty years of groping to prove the things I'd done before were not accidents. Seminars and lectures to rows of fools who couldn't begin to understand my systems. Colleagues. Colleagues laughing behind my back at the boy wonder and becoming famous building on my work. Building on my work!!
Last edited:
I think the next Mac Pro will be announced at WWDC with an 'Ultra' chip. Don't think 'Extreme' chip is cancelled completely. Might be released in M5 variant.
It depends how much Apple could charge for dedicated expansion cards.Agreed, but my hunch is most manufacturers want to shy away from producing modular products. With those types of products, though of benefit to the consumer, they may not necessarily benefit the manufacturer's bottom line to the same extent, so modularity is not prioritised.
In consolation, at least the new Mac Pro seems that it will be fast and powerful. For professionals, that's got to count for something.
If they were to offer cards for all aspects for example:
MX Pro / Ultra Cards
Current generation M series SoC with built in air cooling.
I estimate Apple could charge as much as the equivalent Mac Studio but at greater margins given they would not need to include a new power supply, and the enclosure for an SoC card would be much simpler to manufacture than a whole computer case, etc...
MXG Card
Based on the current generation M series, but focused purely on graphics cores and RAM for users requiring extreme graphics performance.
If performance can rival even an equivalent generation NVidea RTX XX80 card Apple could probably charge at least XX90 series prices accounting for the Apple tax and that customers are locked into a single GPU supplier.
MXN Card
Based on the current generation M series, but focused purely on neural engine cores and RAM for users requiring dedicated AI/ML performance.
I'm less sure on what pricing would look like for this. Asus' AI card based on Google Coral Edge is currently £1,300 on Amazon, I'd expect Apple could charge at least the same.
The key caveat would be that cards would need to be forward and backwards compatible with each other for this to be a viable product line. If I can't keep my perfectly good M4 Ultra card when I upgrade my GPU to the M6G, or if upgrading from an M4 Ultra card to an M6 Ultra card means my expensive M4G and M4N cards are incompatible, then the upgradability isn't practical because you still have to shell out for almost all new parts in each upgrade cycle.
Last edited:
Well, the current 2023 Mac Pro does allow for an upgrade kit to the 2025 internals. The kit would need two components (at most): the logic board and (probably) the Thunderbolt input/output board:
Everything (including heat sink and PCIE) is on the logic board, switching it out is straightforward. The internal SSDs are already modular, so they can be brought over. That in itself is a significant savings.
It wouldn’t be cheap, but the point is that it is already possible. Nothing would need to change in the current design:
Mac Pro (2023) Logic Board - Apple Support
Last edited:
What I find baffling about the Mac Pro debacle is that it is really of Apple's making. There was no need to position this product as an uber-expensive niche model. A tower case is hardly the pinnacle of engineering and nor does it carry with it substantial additional material cost. PCIe slots are hardly revolutionary engineering. If Apple had simply produced a Max version and priced the Pro at a small premium above the Mac Studio for the expandability, I bet it would have sold in decent numbers. It might have cannibalised Studio sales to some extent, but as Steve Jobs said, if you don't cannibalise yourself, someone else will.
I don't get why Apple would kill its own sales in this way? I mean, fair enough, I get that the MacBooks and Mac Mini etc are their best sellers, but who in their right mind would go and buy a new Mac Pro or Mac Studio right now with their 2 gen old chips? Surely it would be better if Apple updated all systems across the range together - then introduce and add the Ultra/Extreme chips when available?!
We have the M4 Max chip already - so theres nothing stopping Apple from updating the Studio with the M4 Max right now, considering that will be an option at some point regardless. Add the Ultra later!
We have the M4 Max chip already - so theres nothing stopping Apple from updating the Studio with the M4 Max right now, considering that will be an option at some point regardless. Add the Ultra later!
I had no idea how hard naming things would be until I became a dev.
Mac Pros were generating sales when they started at £2k. All the designers I knew had them. Then Apple started charging £4,500+. Irrespective of it being a business ex and iMacs etc, the Mac Pro had a 100%+ price increase.
Back in the 90s I dreamt of buying a computer with 4 discrete processors. I would then build a solitaire game where each suit was run by a dedicated processor… 🥸
Which is why Apple would never waste the money to produce them. Apple’s bread & butter is the Air’s and MacBook Pro’s. EVERY other Mac they release is just some iteration of what they focused on improving for the baseline, Pro and Max chips. That allows them to spread the R&D over the entire line. The VERY tiny unit sales of the Mac Pro means it’s not worth the effort to create a unique infrastructure just for 100,000 a year. Making a return on that investment would be really challenging.
And, if someone wants the “fastest Mac”, it’s not like they’re going to find it anywhere else in the industry, so there’s not even a critical need to produce any multi-die solution. Just throw more cores on the Ultra version and there will literally be nothing on the market that can do what it can do.
When Apple did their mea-culpa regarding the Mac Pro, they literally spelled out how they were going to do things going forward (for those that listened). They see dollars aligning to performance levels and, anyone that wants a performance level will need to buy into that performance level going forward. Then, by raising the performance levels of the lowest end so dramatically, they created a world where someone that would have absolutely required some desktop beast would be able to do just as well on a laptop. Many have taken them up on that and dropped the desktop and, for folks that don’t need internal storage or networking cards, they’ve effectively cannibalized the Mac Pro’s reason for existing.
The Mac Pro is expensive and niche because the real number of people that require that level of performance is exceedingly small. So, the cost is directly tied to supply and demand. Few will ever be sold, but those that ARE sold will be going to those for whom price is no object, they simply must have the fastest Mac AND internal expansion and there’s nothing else on the market that fits that description.
Many of those designers were using desktops because laptop processors of the time couldn’t compete against desktops. With Apple Silicon, for £2k today, many of those designers would opt for a laptop with performance that run circles around those old Mac Pros AND, if they need to visit a customer site, they can take their own computer with them. We’re no longer in the age where single core performance of laptop chips are artificially constrained in order to create a tiered structure of chips for customers to buy (at least, not in the Apple Silicon world). So, for most day to day processes an average designer is not going to notice a huge difference between working on a Mac Pro or working on a MacBook Air sitting on an ice pack.
Some may be used to sitting near a big tower just because it “feels” right, but a large number have just gotten a MacBook or MacBook Pro and have left desktop dreams behind.
I don't see the commercial/defense/scientific/industrial markets buying Macs, I see them buying Linux equipment for supercomputing.
Apple finished losing those markets, even without having finished winning them, when it withdrew the XServe a long time ago.
The days of upgrading your Mac without directly paying Apple are all but over.
I would love to see a gaming rig somewhere in this mix of chips and products.
When I worked in those fields I designed a lot of systems around them. That continues today. There's a group within Apple that reaches out to and assists in those markets.
A Mac Pro, especially the rack mount version, is an outstanding computer to build systems around.
It’s useful to remember that the Ultra is already multi-die advanced packaging. Two Max is multiple dies. This fact doesn’t contradict what you’re saying, indeed it reinforces it.
Experimenting with a quad Max design is a natural progression from the dual Max design, but we can only guess about the differences between a four-way silicon interconnect and the two-way silicon interconnect in the Ultra. I think it’s safe to assume it would have to be much larger, and that costs a lot more. So they can probably make it work, but it’s too expensive?
I love Gelsinger’s recent rant about this problem: “Speaking about yield as a % isn't appropriate. Large die will have lower yield, smaller die - high yield percentage. Anyone using % yield as a metric for semiconductor health without defining die size, doesn't understand semiconductor yield. Yields are represented as defect densities.”
In short, large dies cost a lot more.
Last edited:
Sounds like:
+) M4 Ultra is not cancelled : Still due out 2025_June on N3E.
+) M4 Extreme never existed
+) M5 Extreme has been cancelled (was being designed on N2P)
+) The High bandwidth face-to-face chiplet architecture that Broadcom specialises in and licensing out to Google will now be leveraged by Apple in a new DataCentre only AI chip (to allow huge scale out with minimal DataCentre power requirements)
I speculate this will result in the next MacPro moving from an 2025_Oct release to a 2025_June release with an M4_Ultra (based on 2 x M4 Maxes) and will no longer be M5 based.
When TSMC's muti chiplet / multi process rolls out in 2026 (on N2X) /2027 (on A16) - Apple will presumably be able to choose how many GPU tiles are desired, and leave the leave the interconnection optimisation to the routing tools.
+) M4 Ultra is not cancelled : Still due out 2025_June on N3E.
+) M4 Extreme never existed
+) M5 Extreme has been cancelled (was being designed on N2P)
+) The High bandwidth face-to-face chiplet architecture that Broadcom specialises in and licensing out to Google will now be leveraged by Apple in a new DataCentre only AI chip (to allow huge scale out with minimal DataCentre power requirements)
I speculate this will result in the next MacPro moving from an 2025_Oct release to a 2025_June release with an M4_Ultra (based on 2 x M4 Maxes) and will no longer be M5 based.
When TSMC's muti chiplet / multi process rolls out in 2026 (on N2X) /2027 (on A16) - Apple will presumably be able to choose how many GPU tiles are desired, and leave the leave the interconnection optimisation to the routing tools.
With the minor exception of every Mac besides the Mac Pro, of course. Your point is still valid, but they are only dragging their feet with the MP; the rest of the Mac line has been fairly regularly updated since the AS transition.
And that makes sense seeing Mac Pro sales are probably a fraction of 1% of total Mac sales. I'm glad Apple feels that's enough to keep it going. The rack mount version is a beast. I especially like the 1.2 kilowatt power supply. And of course the slots.