Would have bought an M4 Ultra Studio - decided not to get the M3 Ultra or maxed out Max - anyone else?

[G5isAlive]

Yes, we kind of do. LPDDR5 are made mostly by Samsung, Micron, or SK-Hynix. With the M3 Max using four LPDDR5, if you want 256GB of RAM then each LPDDR has to be 64GB. None of the manufacturers ever announced they could do that, and only recently did Samsung claim they could make 32GB LPDDR5x, though they don't have it on their sales page. The M4 Max will use four of those to get 128GB.

As for TB5, while it is true there is a separate IC for the interface, to handle the power delivery requirement, the handshakes, etc., the M3 SoC still has to have IO controllers that are designed to handle the bandwidth, and the display controllers on the SoC have to deal with multiple 8K displays.

As for interconnect between the two sides of the Ultra SoC: a noticeable fraction of the surface area of a die for a large SoC is for wiring all the pieces together, and notably the memory controllers. One does not merely paste two such die together. The engineers will have to had laid out the wiring ahead of time knowing that an interconnect is going to be implemented.

Not sure what you’re arguing.. seems you start with its not possible for a max to have 256 gb ram… so how does the ultra have 512? Agreed two max’s are not glued together, never said it was, the interconnect is old news. But still where does the ultra get its extra memory from? As for tb5, the extra secret sauce is in fact off the chip. The fact that part of the controller is on the chip doesn’t say it never was before. People want to say it’s not the original max being used in the ultra when Apple says it is.

But as Spock says, if you rule out the possible, then no matter how improbable, the impossible is likely. The ultra is two maxes connected the same way they always have been. No mystery.

 

[picpicmac]

so how does the ultra have 512?

That's a question that I asked when it was announced. No one here was able to answer me. From what I have sussed out of the memory market news, it does seem that the players are claiming they are working on higher density chips. So it is possible that, say, Samsung is making something special for Apple. If so, it probably costs a lot.

I also put forth the idea that Apple engineers may have found a way to essentially multiplex two 32GB LPDDR from one controller. That's why they bandwidth did not go up (current advertised bandwidth implies 8 LPDDR, not 16.)

 

[Bug-Creator]

From M1/M2 Ultra, we know that the Ultra performance isn't exactly 2 Max performance.

That proves nothing, it is just a basic fact that adding more cores never scales exactly.

We've been told that's what they are. I have no reason to disbelieve that information.

"We" have been told that M1Ultras were made by fusing 2 M1Max after waver production. It seems save to assume that was also true with M2Max/Ultra.

We know from die shots that early M3Max lacked that ability and it has been suggested for the M4Max too.

What we don't know is how M3Ultras are made. Is it 2 revised M3Max with that connection prepared still getting fused after waver production. Or is it 2 revised M3Max designs placed side by side with the connection made at the same time?

Why? Again we can only guess, but I wouldn't be surprised if signal quality was an issue. With M1Max you had all those traces going nowhere (were they terminated in any way?). With M1Ultra signals had to go from die 1 to the interconnect and then die 2 which for sure added both resistance and some capacity.
Node shrinks combined with higher bandwidth may have made that unfeasible.

Or maybe someone realized that by leaving out the interconnect they could squeeze 10 more M3Max dies on a wafer.

As reviewers are saying: "They have basically set it up for failure by basing it on last year's chip"

Define "failure".
Macs are an afterthought to iPhones.
Desktop Macs are an afterthought to MacBooks.
MacStudio is an afterthought to MacMini.
MacStudio MxUltra is an afterthought to MacStudio MxMax.

Apart from bragging rights there are only very few real use cases for such configs.
Reviewers trashing these have a 100% overlap with reviewer who 10 years ago trashed the 12" MacBook over not being good at video editing. Read they live in their own little bubble refusing to acknowledge that the outside world even exists.

 

[Yimbaz]

I think we will see more and more videos like this one, where people are finding it hard to recommend the M3 Ultra:

"They have basically set it up for failure by basing it on last year's chip"

"The Ultra chip as it exists now, is fatally flawed"

Here's a tip that will greatly increase your own mental health and quality of life: Don't watch YouTube videos with titles and thumbnails like that one. It's designed to make you feel bad.

 

[snak-atak]

I'll consider getting a base M4 Max to tie me over for now.

Curious to know what you’re using right now and in what ways is it inadequate?

 

[G5isAlive]

That's a question that I asked when it was announced. No one here was able to answer me. From what I have sussed out of the memory market news, it does seem that the players are claiming they are working on higher density chips. So it is possible that, say, Samsung is making something special for Apple. If so, it probably costs a lot.

I also put forth the idea that Apple engineers may have found a way to essentially multiplex two 32GB LPDDR from one controller. That's why they bandwidth did not go up (current advertised bandwidth implies 8 LPDDR, not 16.)

Fair enough, so back to my original point, a 256 gb m3 max does exist, just not in the wild.

 

[m1maverick]

some reviewers are doing the click bait thing and saying that..

It never surprises me to see the number of people who defend Apple no matter the circumstances.

Agreed, the M3 Ultra isnt for everyone. But the ultra series never was.

Whether it's for everyone or not is not the point. The point is Apple released a new, top end product based on their last generation technology while moving everything else to their current technology. For those who need the capabilities of the Ultra it seems reasonable the use of current (M4) technology would be preferable to last generation (M3) technology. The same can be said of those who don't need an Ultra.

There may be some technical reason why Apple cannot product an M4 Ultra chip. However Apple hasn't stated as much. Their "There won't always be an Ultra versions of an specific Mx technology" doesn't do anything to offer insight into why an M4 Ultra doesn't currently exist.

Please stop criticizing people who feel that releasing a new product based on last generation technology when everything else has moved to current technology. Especially when the product utilizing last generation technology is the high end offering.

 

[seek3r]

Like many people I don't really need an Ultra, but wanted to buy one nonetheless - and I would have had it been an M4, but I do not want last year's chip with poorer: single core performance, power consumption, resale value and being one year/generation closer to being phased out. I contemplated the 128GB version of the M4 Max, but the cost (for the unbinned version) is so close to the Ultra that it just feels like a rip-off or us being pushed into buying something that we might not have otherwise. As I said, like most people, I don't really 'need' an Ultra or maxing out a Max.

Pretty sure most people would have been happy to wait a little bit long for a M4 Ultra, and it seems this is yet another 'lets screw as much money as we can out of people' scheme from Tim Cook/Apple, perhaps as a way to promote sales of the Mac Pro if it magically gets it first/at the end of the year. Who knows. But what I can confidently say is that people would have preferred an M4 Ultra.

I'll consider getting a base M4 Max to tie me over for now. I guess many others will skip the M3 ultra or the maxed out Max too, in the hope it might prompt Apple to start giving people what they want again.

While I generally agree in that I would have considered an M4 ultra but I want better single core perf so I’m likely to pick up a max personally it’s worth pointing out that higher core count chips having lower single core perf is actually pretty common. Usually it’s to keep within the same general power budget for a generation of chips, that’s not the reason here, but the actual end result is the same.

 

[seek3r]

That proves nothing, it is just a basic fact that adding more cores never scales exactly.



"We" have been told that M1Ultras were made by fusing 2 M1Max after waver production. It seems save to assume that was also true with M2Max/Ultra.

We know from die shots that early M3Max lacked that ability and it has been suggested for the M4Max too.

What we don't know is how M3Ultras are made. Is it 2 revised M3Max with that connection prepared still getting fused after waver production. Or is it 2 revised M3Max designs placed side by side with the connection made at the same time?

Why? Again we can only guess, but I wouldn't be surprised if signal quality was an issue. With M1Max you had all those traces going nowhere (were they terminated in any way?). With M1Ultra signals had to go from die 1 to the interconnect and then die 2 which for sure added both resistance and some capacity.
Node shrinks combined with higher bandwidth may have made that unfeasible.

Or maybe someone realized that by leaving out the interconnect they could squeeze 10 more M3Max dies on a wafer.




Define "failure".
Macs are an afterthought to iPhones.
Desktop Macs are an afterthought to MacBooks.
MacStudio is an afterthought to MacMini.
MacStudio MxUltra is an afterthought to MacStudio MxMax.

Apart from bragging rights there are only very few real use cases for such configs.
Reviewers trashing these have a 100% overlap with reviewer who 10 years ago trashed the 12" MacBook over not being good at video editing. Read they live in their own little bubble refusing to acknowledge that the outside world even exists.

It’s also possible the ultrafusion component on M3 has yield complications (it is made on a transitory node after all) so maybe either there were 2 different tape outs to produce M3 Maxes, one with ultrafusion for the highest end where worse yield can be built into price and one without for everything else or all M3 maxes are initially created with the interconnect, binned, and the part of the wafer that has the interconnect stripped out on chips where it didnt work (I dont know enough about the SOC layout to know if the second is possible).

 

[Howard2k]

There may be some technical reason why Apple cannot product an M4 Ultra chip. However Apple hasn't stated as much. Their "There won't always be an Ultra versions of an specific Mx technology" doesn't do anything to offer insight into why an M4 Ultra doesn't currently exist.

Do you think that producing a base M4 and producing an M4 Ultra would be the same level of difficulty? Or would it be harder to produce the M4 Ultra? And why would this be if there is a difference in production difficulty?

 

[Bug-Creator]

or all M3 maxes are initially created with the interconnect

If all M3Max had been created with the interconnect they could have made Ultras from the start, no matter how bad the yields were. Given that the market for the Ultra is much smaller than the Max they would have needed catastrophic yields to get to a point where they couldn't find enough good Max pairs to make the Ultras.

Which leaves the following:
a) TimApple just likes screwing his most loyal customers over on the products with the highest margins.
b) The couldn't make the interconnect work at first and only figured it out later, but couldn't apply that knowledge to the M4Max/Ultra.
c) M3Ultra is in some way so different to 2xM3Max that they couldn't be binned one way or the other as late as with M1/M2 based chips.

 

[Bug-Creator]

Do you think that producing a base M4 and producing an M4 Ultra would be the same level of difficulty?

It should be the same difficulty as long as the interconnect isn't the problem.
BUT with a the less than stellar yields to be expected from a new process (or just new design) bigger chips tend to produce more waste.

As long as noone can point me to a plausible business reason on why Apple would delay the M4Ultra on purpose I stick to some sort of production problem.

 

[Howard2k]

It should be the same difficulty as long as the interconnect isn't the problem.
BUT with a the less than stellar yields to be expected from a new process (or just new design) bigger chips tend to produce more waste.

As long as noone can point me to a plausible business reason on why Apple would delay the M4Ultra on purpose I stick to some sort of production problem.

Right, that's my understanding too, the yield and interconnect. From what I can find it's probably 10-15 times more wasteful to produce the Ultra variant than the base variant, based on typical yields, and then there is the interconnect issue too.

Even then with the two perfect MAX variants, an interconnect issue means it's no good.

But as yields improve through the production lifecycle then it makes sense that we'd get the Ultra towards the end, not at the beginning.

 

[Bug-Creator]

Even then with the two perfect MAX variants, an interconnect issue means it's no good.

Yes, but.....

The M3Ultra got released after all products with M3Max have been discontinued negating the need to cut 1 Ultra into 2 Max. Add the design change needed over the M3Max with no interconnect and it is highly likely that these new Ultras come as fixed pair on the wafer with no after works need to fuse them.

Or Apple is feeding a bunch of new M3Max Studio/MBP into the supply chain that end up as "refurbished" or NOS on 3rd party store. Does not sound very Apple.

 

[Howard2k]

Yes, but.....

The M3Ultra got released after all products with M3Max have been discontinued negating the need to cut 1 Ultra into 2 Max. Add the design change needed over the M3Max with no interconnect and it is highly likely that these new Ultras come as fixed pair on the wafer with no after works need to fuse them.

Or Apple is feeding a bunch of new M3Max Studio/MBP into the supply chain that end up as "refurbished" or NOS on 3rd party store. Does not sound very Apple.

But fixed pair on a wafer means an even larger surface area and increases the yield difficulty.
And you're right, the discontinuation thing is strange too. I doubt Apple is still producing them for sparing/repairs, but then they did just introduce the M3 iPad Air.

But ultimately I don't know, just speculating.

 

[Bug-Creator]

they did just introduce the M3 iPad Air.

Sure, but thats a different chip that can't be paired or cut to create a different one.

As for repairs, when was the last time anything Apple got a chip level repair (1st party).
If your M3Max dies under warranty/AppleCare you will get a new computer, might be a refurb depending on local consumer rights and how much you are prepared to make a stink about, or a similar M4Max SKU.

 

[m1maverick]

Do you think that producing a base M4 and producing an M4 Ultra would be the same level of difficulty? Or would it be harder to produce the M4 Ultra? And why would this be if there is a difference in production difficulty?

I don't think it would be any more difficult than producing an M1 / M2 / M3 Ultra.

 

[Howard2k]

I don't think it would be any more difficult than producing an M1 / M2 / M3 Ultra.

Right, but that's not what I asked. I asked whether the M4 and M4 Ultra would be different.

 

[m1maverick]

Right, but that's not what I asked. I asked whether the M4 and M4 Ultra would be different.

OK, yes...it would be more difficult to produce an Mx Ultra over a an Mx.

 

[picpicmac]

a) TimApple just likes screwing his most loyal customers over on the products with the highest margins.

The whole Tim-is-victimizing-me mindset is a plague on the Apple community. (And even worse, this kind of thinking is endemic to our current society.)

No, Tim is not victimizing the "loyal customers".

It should be the same difficulty as long as the interconnect isn't the problem.

The M3 series was made on TSMC's first commercially available "3nm" process.

The M4 series switched to a different process (N3E) that (allegedly, I nor anyone else here don't have access to first-hand data) is less difficult hence less expensive.

But, supposedly, one of the reasons that TSMC is switching quickly to an improved process (N3P) is increase density on a chip as N3E turned out, while easier to manufacture, couldn't get the density as high as hoped for 3nm.

Until the release of the iPhone 16E, Apple sold the iPhone SE which had a processor that was several generations older than the regular iPhone 16. It's not unusual for Apple to have products at sale at any given time which happen to use different generations of processors.

It's so odd, that the Apple community gets into fits over these things. Over in the x86/x64 world the offerings are all over the map, many different generations of processors sitting on the shelf right next to each other. And, one can easily find "influencers" who are pushing a previous generation of an Intel chip for some reason (e.g. overclocking.)

 

[Bug-Creator]

Until the release of the iPhone 16E, Apple sold the iPhone SE which had a processor that was several generations older than the regular iPhone 16

Sure a budget product being on an older chip is no surprise, wether it is due to that chip being cheaper or just plain gatekeeping.

The M3Ultra is different, a top tier product where price of a component is a secondary thought and nothing to gatekeep above (apart from the MP on an even older chip).
So a technical reason seems the most plausible one.

 

[rehkram]

It's not always just about money. I have found over time that available under-utilized resources should be a strong factor to guide what I should purchase.

For example, I'm currently sitting on licenses for Logic and Final Cut Pro but without, currently, the hardware resources to do them full justice. Being a very serious amateur musician and fairly good videographer/photographer, I really should be using those tools.

So my purchasing decisions tend to be based on best price / performance, future-proofing, and sufficient processing power to be able to use my existing under-utilized resources productively.

Except when it's a crazy impulse buy (my most recent one was a drone), which may result in short term satisfaction and long term lack of use. But sometimes you just have to dive in, right?

 

[Hajj.david]

To me, it seems obvious that they had a bunch of M3 chips laying around and they knew LLM farms will buy it up if they give a ridiculous amount of ram. Per EXO labs the cost is 6% of what Nvidia charges to run the same size LLMS that they can get with two interconnected studios.

 

[seek3r]

If all M3Max had been created with the interconnect they could have made Ultras from the start, no matter how bad the yields were. Given that the market for the Ultra is much smaller than the Max they would have needed catastrophic yields to get to a point where they couldn't find enough good Max pairs to make the Ultras.

Which leaves the following:
a) TimApple just likes screwing his most loyal customers over on the products with the highest margins.

I really doubt this is the reason for an M3 Ultra instead of an M4 Ultra, it would look a lot better for Apple to have the Ultra be an M4 and the Ultra based machines are expensive enough already to absorb that price. I *would* bet the reason we never saw an M3* Studio or Mac Pro was because they were already eating up most of the available N3B production for much more popular devices.

The reason for the M3 Ultra showing up now could be as simple as N3B capacity freeing up as everything else moves to M4s, may also explain why there's no M4 Ultra - Apple choosing how to allocate available manufacturing capacity.

b) The couldn't make the interconnect work at first and only figured it out later, but couldn't apply that knowledge to the M4Max/Ultra.

I seem to remember the reports were that yields were initially pretty terrible for M3 on N3B, could be it wasnt worth integrating it and making things worse

c) M3Ultra is in some way so different to 2xM3Max that they couldn't be binned one way or the other as late as with M1/M2 based chips.

I have my doubts on this, but eventually someone will do a full scan on the chips and we'll know

 

[JSRinUK]

Before the Studio was announced, I'd kind of set my mind on 128GB RAM. When it was announced, only the M4 Max was offered with 128GB, so that was where my focus was. This model was over my budget by a fair amount, but I could cover it so I was okay with that.

I did look at the M3 Ultra - even though it was even more over my budget (now venturing into "this is too much of a stretch" territory). Being an M3 Ultra wasn't the deal-breaker for me - the fact that it wasn't offered with 128GB RAM was. Having set my mind on 128GB, I wasn't prepared to pay £400 more to get 32GB less RAM (*and* potentially an older chip into the "bargain"). The next step (to 256GB) was just wayyyy too much money for me.

So, for me, the 128GB M4 Max was the "sweet spot".