Apple says not every Apple Silicon generation will get an Ultra

AppleInsider

Apple released the M3 Ultra alongside M4 generation chips -- but here's why we probably won't see the M4 Ultra.

Image credit: Apple

On Wednesday, Apple announced its new Mac Studio. As anticipated, it launched with an M4 chip -- the M4 Max, specifically.

But it also launched with the M3 Ultra. This left many people, including AppleInsider staff, wondering what Apple's reasoning was.

Ars Technica found out. When it asked Apple why the M3 Ultra made an appearance alongside M4 devices, the answer was simple.

According to Apple, not every chip generation will get an Ultra tier. While the first three generations have received an Ultra tier, with the M1 Ultra debuting in March 2022 and the M2 Ultra debuting in June 2023, it seems this may not be the case going forward.

The M3 Ultra is built with Apple's UltraFusion architecture, which links two M3 Max chips over 10,000 high-speed connections. These connections allow the two chips to function as a single unit.

Apple says it delivers 1.5 times the performance of the M2 Ultra and up to 2.6 times that of the M1 Ultra. The new GPU features dynamic caching, hardware-accelerated mesh shading, and ray tracing, making it well-suited for graphics-intensive tasks like 3D rendering and gaming.

And, as Numerama has learned, the M4 Max does not have UltraFusion connectors. The lack of connectors means that, as of now, an M4 Ultra is impossible without a re-engineering effort.

An Apple spokesperson has told Numerama that it opted to upgrade the Mac Studio now, rather than wait.

The backward compatibility of Thunderbolt 5 with the M3 architecture meant there was no reason to delay the upgrade. Otherwise, the company would have had to design an M4 Ultra from scratch or wait until the M5 Max.

So while Apple hasn't said explicitly that it won't release the M4 Ultra, it seems to strongly imply that we may not see another Ultra chip until the M5 generation -- or even later.



Read on AppleInsider

muthuk_vanalingam

Many people in this forum used to say that Apple has made plans for next 5 years, or even 10 years. The question is - Was it part of Apple's roadmap to not have M4 Ultra for Mac Studio from the beginning OR it was part of the plan but Apple ran into unexpected issues and had to change their plans in the last minute?

handsomesmitty

Watching Miani's DoobToob response to today's release, he's wondering why Apple would do this, and also why Apple state that possibly future chips might not get an 'Ultra' upgrade.

Dood--MONEY! Money has always been Apple's motivation with his incremental upgrades. Imagine what an M4 Ultra's processing power could be, and then consider why any Ultra buyers would NEED to upgrade ever again? Or how many people would settle for an M4 Max Studio instead of going the laptop routes.

"Money Money Mooooney!"

melgross

HandsomeSmitty said:

Watching Miani's DoobToob response to today's release, he's wondering why Apple would do this, and also why Apple state that possibly future chips might not get an 'Ultra' upgrade.

Dood--MONEY! Money has always been Apple's motivation with his incremental upgrades. Imagine what an M4 Ultra's processing power could be, and then consider why any Ultra buyers would NEED to upgrade ever again? Or how many people would settle for an M4 Max Studio instead of going the laptop routes.

"Money Money Mooooney!"

They’ve been saying that since the M1 came out. It’s likely just sales. This is expensive on all levels. If Apple isn’t getting the sales it needs to be profitable from it, it will wait a generation to reintroduce it. This is the way all companies work. Though Google would just discontinue it altogether.

applepoor

Go to Apple's Mac Studio site and do a compare with the new Ultra Studio and the M4 Max like I have in my new 16" Mac Book Pro with 128GB of memory and 8TB SSD.

When you see where the numbers are double in the Ultra what they in the M4 Max, I think for marketing purposes Apple choose "M3 Ultra" so they have "M4 Ultra" for the Mac Tower later. The number that will verify this thought will be the single core speed of the M3 Ultra vs the single core speed of the M4 Max.

In both the M1 family and the M2 family of four trim lines, the single core speed for each of the four trim lines were nearly identical. So we will have to see if the M3 Ultra has a single core speed like the M3 family or M4 family.

keithw

While it's nice that the M3 Ultra is now finally out, why did it take them over a year to release it?  (The M3 line came out on October 30, 2023!) Why didn't they release the M4 Max Studio at the same time as the M4 Max MBP?  If they had, I may have saved a few thousand $$$ since I got tired of waiting and bought the MBP.   And is the single core performance of the M3 Ultra the same as the M4 Max? Enquiring minds want to know...  But I guess with the 512MB memory capacity and the 80 graphics cores on top of the 32 CPU cores, the M3 Ultra should be killer LLM machine.

edited March 5

keithw

HandsomeSmitty said:

Watching Miani's DoobToob response to today's release, he's wondering why Apple would do this, and also why Apple state that possibly future chips might not get an 'Ultra' upgrade.

Dood--MONEY! Money has always been Apple's motivation with his incremental upgrades. Imagine what an M4 Ultra's processing power could be, and then consider why any Ultra buyers would NEED to upgrade ever again? Or how many people would settle for an M4 Max Studio instead of going the laptop routes.

"Money Money Mooooney!"

You're absolutely right!  By waiting until now to release the M4 Max Studio, they made an extra $1500 from me when I bought the M4 Max MBP back in December.  I was tired of waiting since they hadn't updated the Studio line in 2 years (and I was running a 2017 iMac Pro which now is a dinosaur.)

programmer

keithw said:

While it's nice that the M3 Ultra is now finally out, why did it take them over a year to release it?  (The M3 line came out on October 30, 2023!) Why didn't they release the M4 Max Studio at the same time as the M4 Max MBP?  If they had, I may have saved a few thousand $$$ since I got tired of waiting and bought the MBP.   And is the single core performance of the M3 Ultra the same as the M4 Max? Enquiring minds want to know...  But I guess with the 512MB memory capacity and the 80 graphics cores on top of the 32 CPU cores, the M3 Ultra should be killer LLM machine.

The thing people don’t seem to understand when whining about not getting this or that is that these products take a huge development effort to create.  And then when they have been designed, there is a complicated balancing act about fab capacities and yields.  This isn’t some blokes with a drill press pumping out aluminum parts from their garage.  This is light years beyond that in terms of complexity.  Apple (and the rest of the industry) are pulling off miracles, and forum trolls pour hate on them because doing that takes no brain cells and you don’t even have to get off the couch.

If I were to speculate wildly, I would suppose that the process node (N3B) used for the M3 series had some issues, and pretty much only Apple used it.  So to address the issues, Apple moved faster on the M4 using the newer process (N3E), and eschewed the ultra connector to get that line out faster.  This may have freed up M3-capable capacity, which they can now use for the M3 Ultra… and IIRC (and this is even more speculative) the first process did have some advantages over the later one (they removed features from N3E to make it work better), which may play better to what high end chips like the ultra need.  So rather than spending the time to make the ultra connector work using N3E, they are probably focused on N3P, which is apparently what comes next.

leehericks

• Not building the ultra fusion connector into M4 is an interesting choice made long before. I'd love to know the reasons. Maybe since it doesn't really yield twice the performance of two Max chips, they have hit a bottleneck where it just isn't performant and cost-effective to do this over creating a chip with beefier cores. We may see a new architecture in M5 or M6?
• Given that, I guess they had a difficult choice to make for updating the Mac Studio. It has been 3 years so an upgrade was needed, but if the M4 Max was the most powerful chip available for it people also would have complained.
• It's interesting that Apple seemed to move quickly to get away from this N3B process node, and is now coming back to it with an M3 iPad Air and this Ultra. Maybe they have a huge stock of binned chips? Does the Ultra get fabricated fused together or can they combine existing M3 Max chips later?

nubus

keithw said:

While it's nice that the M3 Ultra is now finally out, why did it take them over a year to release it?  (The M3 line came out on October 30, 2023!) Why didn't they release the M4 Max Studio at the same time as the M4 Max MBP? 

Seems Apple had M3 Ultra planned, decided to make a bigger splash by moving to M4 Ultra introducing something new, failed, and had to restart the M3 Ultra project. The main difference between M3 Max and M4 Max relevant in a desktop is the new Neural Engine. Perhaps it didn't work in M4 Ultra?

M3 Ultra and M4 Max should be similiar on AI. For other tasks M3 Ultra is much faster and the 512 GB limit is awesome. The new Studio is late but much appreciated.

9secondkox2

Apple doing damage control. 

The m3 max was also reported to lack the ultra fusion interconnects. 

And we are WELL into the M4 generation. 

Why the heck did they wait soooo long to launch an m3 ultra? SHOUld have  happened last year if that was the case. 

Perhaps this is an indicator that m5 is coming sooner than later and the decision was made to forgo the m4 ultra as a result? 

Super bummed we don’t get to see the m4 ultra show what it can do vs Nvidia 5090. 

eriamjh

The Ultra chips likely have a significant yield problem.   Considering they go into products that sell the least by volume and mix, those costs are even more significant. 

Apple has multi-year plans, I’m sure, but plans are fluid and change with presidents and tariffs (nowadays).   

The Mac Studio was talked about being cancelled a recently.   It probably sells circles around the Mac Pro and will continue to do so.  

They’re very small circles, though.   Even the iMac is rumored to only be 3% of Mac sales.    

fidonet127

Is 512 enough memory for those who needed 1.5gb of memory?

dewme

programmer said:

keithw said:

While it's nice that the M3 Ultra is now finally out, why did it take them over a year to release it?  (The M3 line came out on October 30, 2023!) Why didn't they release the M4 Max Studio at the same time as the M4 Max MBP?  If they had, I may have saved a few thousand $$$ since I got tired of waiting and bought the MBP.   And is the single core performance of the M3 Ultra the same as the M4 Max? Enquiring minds want to know...  But I guess with the 512MB memory capacity and the 80 graphics cores on top of the 32 CPU cores, the M3 Ultra should be killer LLM machine.

The thing people don’t seem to understand when whining about not getting this or that is that these products take a huge development effort to create.  And then when they have been designed, there is a complicated balancing act about fab capacities and yields.  This isn’t some blokes with a drill press pumping out aluminum parts from their garage.  This is light years beyond that in terms of complexity.  Apple (and the rest of the industry) are pulling off miracles, and forum trolls pour hate on them because doing that takes no brain cells and you don’t even have to get off the couch.

If I were to speculate wildly, I would suppose that the process node (N3B) used for the M3 series had some issues, and pretty much only Apple used it.  So to address the issues, Apple moved faster on the M4 using the newer process (N3E), and eschewed the ultra connector to get that line out faster.  This may have freed up M3-capable capacity, which they can now use for the M3 Ultra… and IIRC (and this is even more speculative) the first process did have some advantages over the later one (they removed features from N3E to make it work better), which may play better to what high end chips like the ultra need.  So rather than spending the time to make the ultra connector work using N3E, they are probably focused on N3P, which is apparently what comes next.

Thank you x 10 for injecting some sanity into the conversation. First of all, none of the so-called Apple soothsayers who claim to know what Apple’s plans really are are privy to what Apple is actually planning that’s not been externally announced. 

Second, even if they were somehow able to get a snapshot in time of a committed roadmap, anyone who’s worked in any development organization knows that plans change based on priorities, actual progress, current conditions, constraints, and often, funding. The only thing that is constant in business is that things will always change. If you can’t adapt to change you go extinct. 

Third point, these chips are incredibly complex. I remember when the Intel 80386 came out and it was seen as a massive step up in the world of personal computing. It had slightly under one million transistors, which was astonishing at the time. 

The M3 Ultra has more than 180 billion transistors. 

Building these things is a lot harder than making a PB&J sandwich. When the 80386 arrived they needed a massive amount of computing resources simply to verify the design and test the fabrication while using the best fab and test machinery available. I can only imagine what it must take to verify the design and build and detect flaws in an M3 Ultra. All of the machinery need to fabricate the chips has to be built and verified as well. 

The M3 Ultra is one of the most complex consumer SoCs ever developed. It’s a fantastic achievement regardless of what name is attached to it. 

edited March 6

danox

muthuk_vanalingam said:

Many people in this forum used to say that Apple has made plans for next 5 years, or even 10 years. The question is - Was it part of Apple's roadmap to not have M4 Ultra for Mac Studio from the beginning OR it was part of the plan but Apple ran into unexpected issues and had to change their plans in the last minute?

Apple, gave voice to what they’ve been doing essentially all along for the last 10 years their upper division computers have basically been on a 2-3 or 4 year schedule, (take your pick) the Mac Pro, and the big screen iMac have been orphans and the Mac Studio has basically been released every 2-2.5 years, with nothing but internal excuses some of the upgrades Apple made recently are mainly due to them using the Mac Studio as a server in their Apple Intelligence/AI efforts (dragged along kicking scratching I might add).

The thing that really hurts is that Apple has the ability in house to get these products out in a timely manner, 10 years ago it was a menagerie of third-party hardware companies that got in the way IBM, Motorola, Intel, AMD, and Nvidia that excuse is now gone most of the problems are now self-inflicted (iPhone fever). Apple is still doing very well, but you know the rest of the story they have to do better the lead Apple Silicon/Apple OS gives them won’t last forever.

https://www.reddit.com/r/LLMDevs/comments/1j46f5e/apples_new_m3_ultra_vs_rtx_40905090/   Multiple accessible markets available Phones/Tablets/Laptops/Desktops right now no one is even close with solutions in all four major forward facing computing areas with performance and low wattage capability.

https://www.reddit.com/r/MacOS/comments/1j47b6k/apple_announces_m3_ultraand_says_not_every/

edited March 6

danox

dewme said:

programmer said:

keithw said:

While it's nice that the M3 Ultra is now finally out, why did it take them over a year to release it?  (The M3 line came out on October 30, 2023!) Why didn't they release the M4 Max Studio at the same time as the M4 Max MBP?  If they had, I may have saved a few thousand $$$ since I got tired of waiting and bought the MBP.   And is the single core performance of the M3 Ultra the same as the M4 Max? Enquiring minds want to know...  But I guess with the 512MB memory capacity and the 80 graphics cores on top of the 32 CPU cores, the M3 Ultra should be killer LLM machine.

The thing people don’t seem to understand when whining about not getting this or that is that these products take a huge development effort to create.  And then when they have been designed, there is a complicated balancing act about fab capacities and yields.  This isn’t some blokes with a drill press pumping out aluminum parts from their garage.  This is light years beyond that in terms of complexity.  Apple (and the rest of the industry) are pulling off miracles, and forum trolls pour hate on them because doing that takes no brain cells and you don’t even have to get off the couch.

If I were to speculate wildly, I would suppose that the process node (N3B) used for the M3 series had some issues, and pretty much only Apple used it.  So to address the issues, Apple moved faster on the M4 using the newer process (N3E), and eschewed the ultra connector to get that line out faster.  This may have freed up M3-capable capacity, which they can now use for the M3 Ultra… and IIRC (and this is even more speculative) the first process did have some advantages over the later one (they removed features from N3E to make it work better), which may play better to what high end chips like the ultra need.  So rather than spending the time to make the ultra connector work using N3E, they are probably focused on N3P, which is apparently what comes next.

Thank you x 10 for injecting some sanity into the conversation. First of all, none of the so-called Apple soothsayers who claim to know what Apple’s plans really are are privy to what Apple is actually planning that’s not been externally announced. 

Second, even if they were somehow able to get a snapshot in time of a committed roadmap, anyone who’s worked in any development organization knows that plans change based on priorities, actual progress, current conditions, constraints, and often, funding. The only thing that is constant in business is that things will always change. If you can’t adapt to change you go extinct. 

Third point, these chips are incredibly complex. I remember when the Intel 80386 came out and it was seen as a massive step up in the world of personal computing. It had slightly under one million transistors, which was astonishing at the time. 

The M3 Ultra has more than 180 billion transistors. 

Building these things is a lot harder than making a PB&J sandwich. When the 80386 arrived they needed a massive amount of computing resources simply to verify the design and test the fabrication while using the best fab and test machinery available. I can only imagine what it must take to verify the design and build and detect flaws in an M3 Ultra. All of the machinery need to fabricate the chips has to be built and verified as well. 

The M3 Ultra is one of the most complex consumer SoCs ever developed. It’s a fantastic achievement regardless of what name is attached to it. 

Apple Silicon across the board along with all the variants of Apple OS are great achievements, and so far Apple hasn’t been resting on their laurels, their iterations over the last 25 years has been second to none in comparison to their competition, however, the current fate of Intel, IBM, Xerox, Kodak, and the death of Motorola of Schaumburg, Illinois means the job is never done.

 If Apple has a advantage that you can use/press and not go too far out of bounds you have to pursue it that includes getting those Apple Silicon computing trucks out the door in a timely manner the time is now while most of the competition is struggling, Apple has/is getting the attention of a new generation of very smart people who want to get into AI programming and they are looking for solutions now Apple Silicon appears to be in the right place at the right time once again in comparison to the present competition,.

https://www.reddit.com/r/LocalLLaMA/comments/1j43us5/apple_releases_new_mac_studio_with_m4_max_and_m3/  Refer to the comment section the enthusiasm that is expressed in that section by most of the participants means there is a market, the computing trucks have to be made, Apple has to be a part of the mind share in this area of computing, they can’t let it go like they have in the 3-D CAD construction, and CAD/CAM world, particularly since they have Apple Silicon and a viable ecosystem today to put up a fight.

edited March 6

applepoor

To cover their costs to create and get the operational M3 Ultra, they will need to make a lot of them. As outsiders, we have zero clues how many of the M3 Ultra Mac Studios and what configurations might be going into Apple's AI farm. When we aerial photos of their building show that they cover dozens of acres, that could be lots of Mac Studios, perhaps in more plebeian garb with just the ports needed in there computer farm(s).

A few thousand individual unit sales will be some frosting around a candle but the cake I bet requires many thousands to cover costs even if being used in house.

programmer

ApplePoor said:

To cover their costs to create and get the operational M3 Ultra, they will need to make a lot of them.

I don’t think that’s true.  I suspect the original M3 Max design had the ultra fusion connector, but they just masked it off to make the die a little smaller.  Once they decided to start making ultras they stop cropping it, and voila!  The amount of additional design work could be virtually nil.  The markup on the ultras is pretty significant, so that reduces the number of sales needed to profit.  And if they are using ultras in servers (which might be why we’ve not seen them until now), selling to users is just helping amortize dev costs.

marvin

9secondkox2 said:

Super bummed we don’t get to see the m4 ultra show what it can do vs Nvidia 5090. 

The 5090 raw performance has actually been quite disappointing. They were marketing 70% increase in performance but the actual results haven't been that much. Here the 5090 is just 35% faster than the 4090:

https://opendata.blender.org/benchmarks/query/?compute_type=OPTIX&compute_type=CUDA&compute_type=HIP&compute_type=METAL&compute_type=ONEAPI&group_by=device_name&blender_version=4.3.0

The M4 Max is on page 2 with 5093 and M3 Max just below at 4130.

The Ultra is around 90% faster than Max so M3 Ultra should be 7847 (equivalent to Nvidia 4080) and M4 Ultra would have been 9676 (equivalent to Nvidia 4090), about a 25% difference.

Although M3 Ultra loses out a bit on raw performance vs the 5090, it has more memory. Most people will get the 16GB Nvidia model, some will get 32GB. Neither are good for AI. M3 Ultra allows for 512GB. It also won't melt a hole in the desk when it's running at full speed:

https://www.pcgamesn.com/nvidia/asus-rtx-5090-burned-out

danox

eriamjh said:

The Ultra chips likely have a significant yield problem.   Considering they go into products that sell the least by volume and mix, those costs are even more significant. 

Apple has multi-year plans, I’m sure, but plans are fluid and change with presidents and tariffs (nowadays).   

The Mac Studio was talked about being cancelled a recently.   It probably sells circles around the Mac Pro and will continue to do so.  

They’re very small circles, though.   Even the iMac is rumored to only be 3% of Mac sales.    

When Apple introduced the M4 iPad Pro the introduction increased sales by $1.5 billion in one quarter and those sales have been up since if you multiply by four that is $6 billion per year and a new big screen iMac being sold over the course of one year probably would do about the same yes that number isn’t up there with iPhones or the laptop sales, but it is a significant amount of very profitable dollars being left on the table (the markets are not the same many people have no interest in the laptop computer), note the sales of iPads we’re down from previous highs in the quarters leading up to the introduction of better M4 iPads.

The same scenario applies to towers and servers money billions is being left on the table by Apple up until this point, it’s good to see Apple offer/upgrade, the Mac Studio and going into the future Apple needs to offer more if they don’t want to lose that awakening AI mind-share for truck computers to continue go to AMD, Intel, and Nvidia and make no mistake they are the competition long-term to Apple hardware. Apple Silicon with its inherent advantages is at now at the crossroads where they can and should make life really difficult/miserable for the big three, Nvidia in particular has a huge unsustainable wattage problem.

edited March 8