Apple announces M3 Ultra—and says not every generation will see an “Ultra” chip

It could explain why we’re getting an M3 Ultra this deep into the M4 rollout.

See full article...

 

[BB]

Looks like M3 has AV1

https://www.apple.com/newsroom/2023...-most-advanced-chips-for-a-personal-computer/

And, a new media engine now includes support for AV1 decode, providing more efficient and high-quality video experiences from streaming services.

 

[Got Nate?]

what is the point of an M3 Ultra vs the M4 Max?

Memory ceiling, memory bandwidth, and GPU cores. The Ultra is an AI monster.

 

[Still Breathing]

what is the point of an M3 Ultra vs the M4 Max?

See my reply above regarding my use-case. Twice the CPU cores, twice the GPU cores, twice the available RAM.

It's too much of a good upgrade to turn down over my M4 Max MBP... Today turned out to be an expensive day. A 32-core ultra/512GB (the RAM is important) machine for me, and a new MBA for the kid's birthday - I'd already promised him he was getting one when it came out, and he was good to wait.

 

[Ryan B.]

An M4 Max has about the same power as an M1 Ultra.

The ultra had the advantage that no new design was needed. Just an interconnect to double cost and power. I could imagine they can design a max-like chip with 16 or 20 instead of 12 performance cores. I can’t quite see why you wouldn’t base it on an M4 Max with extra cores instead of M3 with extra cores.

What I don't understand is, since the "Ultra" chips are just two Max chips fused together, shouldn't it be (relatively) trivial for them to make an M4 Ultra given the M4 Max already exists?

These chips have a lead time measured in years (approximately two most of the time). That is what we know for sure. Now, come with me on a journey of speculation to try and explain what happened here.

It looks to me, standing here on the outside, that the M4 was rushed because it was needed in order to drive the tandem OLED in the iPad Pro. If that's true, it seems likely that resources were pulled from the Ultra team in order to make that happen.

This then put the Mac Studio in a bit of a sticky spot. It hadn't seen an update in several years. Apple could either finish the in-progress work on the M3 Ultra and push the update out, or it could delay it even further in order to make an M4 Ultra. If they had done that, we would likely have had a situation where Apple had an M5 Max by the time the M4 Ultra was ready.

Skipping a generation on the Ultra may have been the only thing they could manage to do if they didn't want it to remain a generation behind forever.

 

[msadesign]

Let’s take a broader view.

Computers these days are so mind-numbingly fast for nearly all of us for the usual daily tasks. Who actually notices the speed bump?

And before you press the down vote button, lol, my own computer use as an architect involves some fairly hefty files. True, I’m not into video and the like; those guys really use the horsepower. For us, max ram is more important.

I do love me some very high-end technology. I do. I love it when files snap open! I’ve owned all of them since ’84. But unless you’re a bonafide geek, which I am, only the spec sheet differentiates these crazy fast M chips. When I buy computers for my guys, they get two year old refurbs.

Maybe I’m missing something?

Edited to add: wow! Reading these comments has really schooled me! I can see just how parochial my views and use case really are. Still, for my small studio, my buying plans remain the same. I’m just not even close to running with the big dogs.

Often I learn more from the comments than from the article.

 

[stilgars]

See my reply above regarding my use-case. Twice the CPU cores, twice the GPU cores, twice the available RAM.

It's too much of a good upgrade to turn down over my M4 Max MBP... Today turned out to be an expensive day. A 32-core ultra/512GB (the RAM is important) machine for me, and a new MBA for the kid's birthday - I'd already promised him he was getting one when it came out, and he was good to wait.

Twice the count, but 25% less efficient ... Ultimately, benchmarks should show that the M3 Ultra is only 30-40% better versus the M4 Max.
But twice the RAM, indeed (but only 35% more bandwidth).

 

[Still Breathing]

Twice the count, but 25% less efficient ... Ultimately, benchmarks should show that the M3 Ultra is only 30-40% better versus the M4 Max.
But twice the RAM, indeed (but only 35% more bandwidth).

Without taking contention etc. into account:

M4 Max : 16 M4 cores => 1.25x 16 M3 cores => 20 M3 cores
M3 Ultra : 32 M3 cores / 20 M3 cores => 1.6x the speed of the M4 Max

This doesn't take into account the GPU performance increase - as I said above, a huge amount of the stuff I use these beasts for now is using Metal and GPU compute. 80 GPU cores, all with access to 512GB of RAM is ... breathtaking. I chew up RAM, CPU and GPU like there's no tomorrow with all the image-registration stuff.

I've also been thinking about how I can leverage the "Neural Engine", which is effectively a lot of matrix multipliers. I do a lot of matrix multiplication... The Ultra has 2x the NE core count as well.

 

[Still Breathing]

Let’s take a broader view.

Computers these days are so mind-numbingly fast for nearly all of us for the usual daily tasks. Who actually notices the speed bump?

And before you press the down vote button, lol, my own computer use as an architect involves some fairly hefty files. True, I’m not into video and the like; those guys really use the horsepower. For us, max ram is more important.

I do love me some very high-end technology. I do. I love it when files snap open! I’ve owned all of them since ’84. But unless you’re a bonafide geek, which I am, only the spec sheet differentiates these crazy fast M chips. When I buy computers for my guys, they get two year old refurbs.

Maybe I’m missing something?

It's entirely possible there isn't a good case for you to upgrade your machine, if it's serving you well right now. For me (just as a home-user) there is a compelling case because of my astrophotography hobby.

 

[Ryan B.]

Let’s take a broader view.

Computers these days are so mind-numbingly fast for nearly all of us for the usual daily tasks. Who actually notices the speed bump?

And before you press the down vote button, lol, my own computer use as an architect involves some fairly hefty files. True, I’m not into video and the like; those guys really use the horsepower. For us, max ram is more important.

I do love me some very high-end technology. I do. I love it when files snap open! I’ve owned all of them since ’84. But unless you’re a bonafide geek, which I am, only the spec sheet differentiates these crazy fast M chips. When I buy computers for my guys, they get two year old refurbs.

Maybe I’m missing something?

Edited to add: wow! Reading these comments has really schooled me! I can see just how parochial my views and use case really are. Still, for my small studio, my buying plans remain the same. I’m just not even close to running with the big dogs.

Often I learn more from the comments than from the article.

My personal use case (software development) can always use more speed. The shorter the compile-link-load-debug cycle is, the more productive I can be.

That being said, I tend to keep my personal machines for a really long time. The M1 Max I own today replaced a 10-year-old machine.

 

[2TurnersNotEnough]

See my reply above regarding my use-case. Twice the CPU cores, twice the GPU cores, twice the available RAM.

It's too much of a good upgrade to turn down over my M4 Max MBP... Today turned out to be an expensive day. A 32-core ultra/512GB (the RAM is important) machine for me, and a new MBA for the kid's birthday - I'd already promised him he was getting one when it came out, and he was good to wait.

As an Apple shareholder, you have my gratitude

 

[ebeshore]

Honest question: what are the people who need the Ultras doing with them?

Astronomical image processing. I take about 20GB of data a night with my remotely-hosted telescope. Calibration and processing of hundreds of images scales nicely with core count on these chips with fast interconnects and on-chip RAM.

 

[Ryan B.]

My point is - the M1 and M2 Ultra chips are just two Max chips fused together with an interposer. There shouldn't be any additional engineering work for an M4 Ultra that wasn't already done for the M4 Max, which is already shipping (and has been for months), so this doesn't make sense to me.

And I’m saying there’s no way it’s that simple. I’m far from an expert in the field, but saying, “Just glue them together, how hard could it be?” seems reductive to me.

 

[plectrum]

Let’s take a broader view.

Computers these days are so mind-numbingly fast for nearly all of us for the usual daily tasks. Who actually notices the speed bump?

And before you press the down vote button, lol, my own computer use as an architect involves some fairly hefty files. True, I’m not into video and the like; those guys really use the horsepower. For us, max ram is more important.

I do love me some very high-end technology. I do. I love it when files snap open! I’ve owned all of them since ’84. But unless you’re a bonafide geek, which I am, only the spec sheet differentiates these crazy fast M chips. When I buy computers for my guys, they get two year old refurbs.

Maybe I’m missing something?

Edited to add: wow! Reading these comments has really schooled me! I can see just how parochial my views and use case really are. Still, for my small studio, my buying plans remain the same. I’m just not even close to running with the big dogs.

Often I learn more from the comments than from the article.

It depends on what you do, but for developers there's often a solid business case for faster machines. The biggest cost for many business is salaries. If buying a new machine can make your devs x% more productive, then you can earn back the cost of the machine in a relatively short length of time.

Even if you have a long refresh cycle, even making somebody $1 an hour more productive adds up over the working year.

 

[repoman27]

Exact, it is clearly revised die for the combined M3 Max, considering they have Thunderbolt 5 and can reach 256 GB each (=512 total)

The up to 512GB DRAM capacity is achieved by using dual 8-high stacks of 32-Gbit LPDDR5-6400 dies per module x 8 modules. The M3 Max topped out at 16-Gbit dies and 4 modules for the 128GB configuration. So no changes to the SoC necessary.

I'm beginning to wonder if Apple was sandbagging a bit and included Thunderbolt 5 in the M3 generation for testing but only enabled Thunderbolt 4 functionality for whatever reason.

Also really curious if the Ultra is manufactured using TSMC N3 or N3E.

It's possible the M3 Max die is a chop of the M3 Ultra die.

 

[Still Breathing]

As an Apple shareholder, you have my gratitude

Hey, I have shares too

 

[Still Breathing]

And I’m saying there’s no way it’s that simple. I’m far from an expert in the field, but saying, “Just glue them together, how hard could it be?” seems reductive to me.

Agreed. The word "just" is carrying a lot of water there, IMHO.

 

[caramelpolice]

And I’m saying there’s no way it’s that simple. I’m far from an expert in the field, but saying, “Just glue them together, how hard could it be?” seems reductive to me.

It's not literally as simple as gluing them together obviously - they use die-to-die interconnects which requires its own design considerations. But those considerations would have already been made when the Max chip was designed and shipped. Otherwise, yes, M1 and M2 Ultra were very literally two M1/M2 Max dies joined at the interconnect interface.

Given that M3 Max had no interposer, I'm wondering if Apple is moving away from UltraFusion altogether and M3 Ultra is actually just its own unique die. Nowhere does Apple seem to mention M3 Ultra using UltraFusion like the previous generations.

EDIT: Well, that's wrong. The M3 Ultra press release does say it uses UltraFusion with two M3 Max dies. So hell if I know what's going on.

 

[2TurnersNotEnough]

It depends on what you do, but for developers there's often a solid business case for faster machines. The biggest cost for many business is salaries. If buying a new machine can make your devs x% more productive, then you can earn back the cost of the machine in a relatively short length of time.

Even if you have a long refresh cycle, even making somebody $1 an hour more productive adds up over the working year.

Last place I worked, they upgraded our mobile app development team to M1 Max MacBooks Pro when they came out after seeing build times for the app get cut in half (saving about 2-1/2 minutes per build of the iOS app). With the time savings, it penciled out to recouping the cost of the laptops in around 2 months.

 

[chuckharrington]

Good thing they didn't announce the MK Ultra...am I right, huh, huh....ugh

 

[breze]

Software development thanks to the higher core count, LLMs thanks to unified memory and the higher RAM limit allowing you to use larger models than even the highest-end nVidia H200 cards (albeit slower processing due to a slower GPU/NPU).

The new AMD Strix Halo (and desktop variants theof may give it a run for the money, though, and of course the nVidia Digits.

My 2022 $5000 M1 Ultra 128GB Mac Studio is obsoleted by the M4 Mac Mini, however, so these machines have a short relevance span, even if they are usable for more than 3 years,

Sounds terribly obsolete, want me to dispose of that e-waste for you? I’m an environmentalist or whatever so I’ll even do it for free.

 

[thequietplace]

In the context of AI they didn't increase the memory bandwidth. You can get the same memory bandwidth on the M2 Studio. Yes, yes, of course you can get 512 gigabytes of uRAM for 10 grand with M3 Ultra. The the question is if a LLM will run with usable performance at that scale?

The point is there's diminishing returns despite having enough uRAM with the same amount of memory bandwidth even with increased processing with M3 Ultra. Therefore there must be a min-max performance ratio between memory bandwidth and memory pool in relation to the processing power. Taking a broader look at the available hardware.

So far the "reasonably priced" seem to be NVIDIA digits (memory bandwidth unknown), (AMD Ryzen™ AI Max 385 products with up to 256 GB/s ) like Framework Desktop, M1 64gb, M2/M3 128gb studio/ultra with same memory capacity.

It's the first time anything else is competing with Apple with unified memory. The GPU market isn't competitive enough for the amount of VRAM needed for larger models.

The AMD Ryzen™ AI Max 385 products devices I think a second generation will be significantly better than what's currently on offer today. Rationale below...

For a max spec processor with ram at $2,000, this seems like a decent deal given today's market. However, this might age very fast for three reasons.

Reason 1: LPDDR6 may debut in the next year or two this could bring massive improvements to memory bandwidth and capacity for soldered on memory.

LPDDR6 vs LPDDR5 - Data bus width - 24 bits, 16 bits Burst length - 24 bits, 15 bits Memory bandwidth - Up to 38.4 GB/s, Up to 6.7 GB/s

Reason 2: - AMD Ryzen™ AI Max is a laptop chip with limited PCI lanes and reduced power envelope. Theoretically, a desktop chip could have better performance.

Reason 3: PCI lanes are limited AMD Ryzen™ AI Max 385 products. No announcement so far has released 16. PCI 5.0 express slot. Technically possible within AMD Ryzen™ AI Max 385.

I'm excited to see other processors enter the market with high-speed RAM besides Apple. The RAM most likely won't be modular. Framework tried to include camm ram but they were told it wasn't feasible by AMD. I'm not sure if that's a processor architecture design issue or signal integrity issues at 256 GB/s with the camm standard. Logically if camm standard doesn't work with 256 GB/s due to signal integrity there's going to be more signal integrity issues at 500 GB/s and 800 GB/s. This leads to three questions.

1. Will there be socketed CPUs (not soldered like the mobile chips) with soldered RAM?
2. Would a dedicated GPU through PCI express provide a significant performance advantage utilizing unified memory 256 GB/s in combination with a processor like AMD Ryzen™ AI Max?
3. Dedicated GPU's are going to start seeing upward pressure from APU/NPU with unified memory. This may lead to the demise of the PCIe within consumer products downsizing our options for modular peripherals and storage. A trend like Apple Mac Pro workstation which has PCIe but doesn't allow dedicated graphics and is only available in the workstation line. In this case not a technical limitation but market segmentation.

 

[ibad]

Knowing nothing about the details, but making a completely uneducated guess: You have to have two perfect M3 Max chips and a perfect interposer and you've gotta join them up perfectly. I'm thinking this is an issue with yields. If doing it were easy, we wouldn't have seen delays in the M3Ultra rollout and the M4Ultra would be available on day one. (Do the chip designs have additional P and E cores to improve yields? )

If that's the case, then I guess we can expect an M4 Ultra when yields improve? Maybe in a year?

 

[gnasher729]

Fwiw it’s nice of Apple to admit they are using M3 but it’s super confusing to the consumer that M3 Ultra is better than M4 Max

M3 is just the third generation. Each generation has quite a few different models with massive performance differences.

 

[gnasher729]

what is the point of an M3 Ultra vs the M4 Max?

Higher performance. The improvement from 3rd to 4th generation is 10%, but ultra would have many more cores than max.

 

[Thunderracker]

So the maxed out Mac Studio is $14099. But I am not getting the Final Cut Pro license because I don't want to waste money. lol.

 

[Dano40]

Honest question: what are the people who need the Ultras doing with them?

Someone will find a use for 512 gigs of UMA memory that runs at low wattage, certainly Apple is using them behind the scenes on their server farm for Apple Intelligence….

 

[Dano40]

At $4000 before addons, I don't think a consumer will be considering the M3 Ultra at all. The Pro who knows what they need will lap these up, but the Ultra is not targeted at the consumer.

Part of the reason for their existence with 512 gigs is the fact that Apple is using them behind the scenes, and if they are using them, why wouldn’t Apple also sell them to the public too?

 

[Dano40]

When the M4 first came out there were reports that the architecture of the M4 was quite different from the M1 through M3 architectures. This led to two different speculations in the industry.
1) The new architecture would eliminate the need for the same type of interposer between the two M4 Max implementations making it "easier" to create the equivalent of the M4 Ultra. It is very possible that Apple has run into issues joining two M4 Max into one M4 Ultra due to this new architecture.
2) The way the Ultras were constructed made doing the whispered Mx "Extreme" (effectively four Max chips welded together for a true, high end workstation chip for the Mac Pro) an impossibility. The new M4 architecture was rumored to get around this blockage by making the way the Mx chips could be bound together more technically achievable.

All this was speculation around the time the first M4s started showing up. We'll just have to see if an M4 Ultra does show up based on this new architecture.

Apple probably will jump to the M5 ultra from the M3 ultra.

 

[repoman27]

So the maxed out Mac Studio is $14099. But I am not getting the Final Cut Pro license because I don't want to waste money. lol.

Omitting FCP is fine, but you'll still want a display, keyboard, trackpad and AppleCare+, which brings the total to... $17064 USD. And that's with just a single Studio Display, not even a Pro Display XDR!

edit: With 8 Pro Display XDRs you can get the total up to $74592, but that's probably a little excessive.

 

[benwaggoner]

Very diminishing returns, except when that $1500 offsets against time saved over the 2-3 years the Mac will be in use for creative pros saving them a minute here and there over these 2-3 years - it adds up.

Yeah, spending an extra $10K/year to increase the productivity of a senior digital media professional by 10% is normally a good deal. It's remarkable just how pennywise and pound foolish many enterprises are in terms of hardware upgrades. Lots of upgrades will pay off in just a month compared to the cost of salary, office space, benefits, admin overhead, etcetera for the person getting that upgrade.

 

[benwaggoner]

Agreed. The word "just" is carrying a lot of water there, IMHO.

The word "just" should break assumed to be carrying a lot of water by default, really.

"I don't know much about this, but how hard could it be?" is almost always answered with "a whole lot harder than you think, for myriad reasons you aren't aware exist."

 

[Terakh]

man that is insane - man 512GB paired with a cpu... what would the cost of such a beast be and life expectancy of it. i can imagine it still being faster than what an average joe will have 10 years down the road...

 

[caramelpolice]

Agreed. The word "just" is carrying a lot of water there, IMHO.

The word "just" should break assumed to be carrying a lot of water by default, really.

"I don't know much about this, but how hard could it be?" is almost always answered with "a whole lot harder than you think, for myriad reasons you aren't aware exist."

I feel like my post is being fundamentally misunderstood.

I'm not saying 'surely they can just glue two M4 Maxes together it's easy'. I'm saying that, if they ever planned to fuse M4 Maxes together, that design work would had to have been done during the development of M4 Max because the Ultra chips are built with Max dies. If you can ship two Maxes you can ship an Ultra unless there is some extremely specific problem with designing or manufacturing the interposer.

So either the M4 Max was straight-up not designed to be fused like previous Max chips, or Apple has some other internal reason for holding back on an M4 Ultra, but it doesn't make sense for the reason to be a delay in chip design, because the design would have to have been done when the M4 Max was designed. (And obviously the M4 Max has been designed, manufactured, and shipping for months.)

 

[BadSuperblock]

Honest question: what are the people who need the Ultras doing with them?

Given that the main hallmark of the Ultra is the number of CPU and especially GPU cores, it's anything where GPU power is the potential performance bottleneck. This means video editors, 3D artists, AI features....

But honestly, the lower tiers are just fine for most people, who could be maximally productive with the base Mac Studio or a tricked-out MacBook Pro. For modest workloads those Macs are more than enough.

The Ultra starts to pay for itself when your workloads are heavy, like many video tracks with effects, very large 3D models, or batch-processing photo edits where GPU-intensive AI features were applied like noise reduction. But that is a small niche; the extra cores are wasted on light to moderate workloads.

For example this is a question you can ask yourself: How much is your professional time worth, like $500-$1000 a day? So if the speed increase of the Ultra could save you a few days of work over its service lifetime, paying the extra $2000 to get to the Ultra could be well worth it. For the highly paid it could be a no-brainer.

(For my work, that is not the case, so if I got a Mac Studio I can only justify the base model, which is quite powerful already.)

 

[OSB]

Not cheap, yes, but $1,500 also includes 20 more GPU cores.

Precisely this. My workflows wouldn't benefit from this, and I suspect that's true of most people, but I presume there are at least some users out there who are render- or GPU-compute starved even on a full-fat M2 Ultra. Value is always a personal question, but Apple obviously believe some number of buyers must care....

 

[BadSuperblock]

man that is insane - man 512GB paired with a cpu... what would the cost of such a beast be and life expectancy of it. i can imagine it still being faster than what an average joe will have 10 years down the road...

That isn't a guaranteed assumption. I often tell people that maxing out a system does not necessarily future-proof it.

For example I have a 2007 desktop tower which in theory is quite expandable and you can stuff it with all kinds of amounts of RAM and storage, replace the cards in the PCI slots, etc. But 10 years later it was starting to not do the job. Industry needs were changing, newer versions of OSs had features that used newer components and basically the old motherboard could not support them and were not compatible. I couldn't put in the latest graphics card. Its Intel Xeon CPUs lack some features and efficient hardware accelerators like the media engines and Neural Engine NPU that today you can get in a sub-$1000 MacBook Air...or an iPhone.

You said "RAM and CPU" but if you look at where the major performance advancements have been in the last 5 years, those are the two areas where the improvements have not been. The real performance gains have come from the GPU, the media engines, and (for AI) potentially the NPU in the future when more apps figure out what to do with it.

 

[Faceless Man]

Good thing they didn't announce the MK Ultra...am I right, huh, huh....ugh

That's the one with the MJ-12 cores.

Personally, I'm holding out for the M5 Ultimate.

 

[mschira]

the fact that someone would consider a machine obsolete the moment a better one appears just baffles me

Yea, my main office machine is a 2017! iMac maxed out with (I forget what it was) some Intel i7 and 64Gb of RAM. Was a monster back in the days, still works flawless today.
I do lot of the hardcore computing on a dedicated cluster.
I don't need the latest power on my desktop anymore with how easy cluster access became, but I still need a decent amount of local power (and RAM). So if I had specced that iMac smaller back then it would not suffice anymore.

Surprisingly, my M3Pro Macbook is not that much faster than my desktop.
Well, it is. Noticeable, but not making the iMac unusable (the biggger screen makes up for a lot...).
Yea. Thinking about a M4Max studio, but I can't find a screen that would be a real upgrade over my iMac. And THAT is shocking.