Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Doesn't Apple (apparently) saying "the M4 Max lacks UltraFusion" suggest that the M3 Max does have it, and therefore that the M3 Ultra is two Max's glued together?

(It could also be a translation error?)

Yes, M3 Ultra most-likely does use UltraFusion to link two M3 Max together, but as Zdigital2015 has noted, these M3 Max SoCs have TB5 controllers (instead of the TB4 controller on the M3 Max at launch) and the addressable memory capacity are much higher, which implies these M3 Max also have new memory controllers. So these "Max" are not the same "Max" that launched in the MacBook Pros in October 2023.
 
Last edited:

MacRumors reported on this tweet, and the commenters flogged him

Fair enough, but he also said the M4 Ultra would be released later, which Apple is saying is impossible. So my point still stands.
 
  • Like
Reactions: picpicmac
So... the rumour I heard is that the Mac Pro will have something new.

A discreet GPU/AI Board that can compete with the 5090!

Presumably with an M5. The M4 was announced in May 2024 with the iPad Pro so it'll be WWDC I guess?
 
Fair enough, but he also said the M4 Ultra would be released later, which Apple is saying is impossible. So my point still stands.

Impossible "right now".

Apple could conceivably add UltraFusion to the M4 Max at a later date, as may or may not have been done to the enhanced M3 Max used in the new M3 Ultra.
 
Doesn't Apple (apparently) saying "the M4 Max lacks UltraFusion" suggest that the M3 Max does have it, and therefore that the M3 Ultra is two Max's glued together?

(It could also be a translation error?)
I'm not sure we can make that assumption. Perhaps the inclusion of Thunderbolt 5 and the expanded DRAM capacity are simple changes that didn't require any changes to the M3 Max die, but what if the M3 Ultra is the first member of a new family of SoCs like the M3 Pro that are distinct from just being doubled versions of the smaller sibling. Am I incorrect in saying that the M3 Pro is it's own beast and so is the M3 Max and the M3 Ultra?
 
I'm not sure we can make that assumption. Perhaps the inclusion of Thunderbolt 5 and the expanded DRAM capacity are simple changes that didn't require any changes to the M3 Max die, but what if the M3 Ultra is the first member of a new family of SoCs like the M3 Pro that are distinct from just being doubled versions of the smaller sibling. Am I incorrect in saying that the M3 Pro is it's own beast and so is the M3 Max and the M3 Ultra?
Yes, I believe the M3 Pro and M4 Pro aren't chops of the M3 Max and M4 Max.
 
A GPU with access to over 400 Gigs of RAM. If you can find a physical GPU with 400 Gigs of ram available to it, I bet you can’t get it for the price of a Mac :)
 
Screenshot 2025-03-05 at 22.46.43.png

The above passage is in Apple's new room press release. I have no idea why they didn't make that part clear in the product pages so you guys could have seen this easier.
 
View attachment 2488780

The above passage is in Apple's new room press release. I have no idea why they didn't make that part clear in the product pages so you guys could have seen this easier.
Because Apple often oversimplifies the copy in its main marketing messages and leaves technical information for users to dig through in obscure or less visible places. Thanks for pointing this out.
 
Impressive specs. Looks like there won’t be an Ultra version of M4. It should be good enough for the Mac Studio
 
  • Like
Reactions: mganu
Yes, I believe the M3 Pro and M4 Pro aren't chops of the M3 Max and M4 Max.
There’s definitely a slow evolution in place with regard to interconnects and Apple segmenting their products. There’s also a lot of open questions that even frustrate me and I don’t pour over articles on processor technology like I used to do when I was younger…all I can say is that I really miss Byte magazine and Anandtech.
 
Last edited:
  • Like
Reactions: CWallace
25% is nothing to sneeze at.

I wouldn't be surprised if even with medium-sized Xcode projects, the M4 Max comes out ahead, because even moderately high-end users simply don't saturate 32 (vs. 16) cores (presumably 24p8e vs. 12p4e) that often.
No, I won’t dismiss 25% faster SC like it’s nothing, but I would probably buy an M4 Max (40c GPU/128GB/4TB) for serious Xcode work (day job) as opposed to spending on an M3 Ultra, which I would want if I was dealing with 6K footage all day long and only coding part time in Lua (DaVinci Resolve), JavaScript or Python. My opinion, for what it’s worth, is that for most prospective M3 Ultra owners, the SC performance hit is a good enough trade off give the MC and GPU performance gains.

Again, how the M3 Ultra performs is a moot point until it’s released. I’m looking forward to those scores, no matter how synthetic they are.
 
  • Like
Reactions: CWallace
Looking forward to reading through these comments to find the people who can see the silliness on this approaching and point out the details that will be lost on article writers. At least in terms of initial hype.

I don't personally care about the performance of these machines, though I have been advising others on thier next laptop. This will help drive down the prices of the previous versions - so keep their 10% better numbers, prices are what matters.. the performance will be good enough.

Keep the analysis coming guys, thx in advance.

~ mac user persons, inc.
 
Apple should be optimizing Comfy UI and similar AI tools for these beasts. With the amount of unified RAM that they have, they would put the 4090 to shame if they helped the developers optimize the code. But they are missing this opportunity and letting much cheaper NVIDIA cards outperform the Maxes and Ultras by a wide margin. It’s sad.
 
Apple somehow managed to broke through the addressable memory to from M3 Max's 128GB (x2 to be 256GB) to 512GB
And if there are eight LPDDR5 SDRAM chips, the question is who is making the 64GB chip for Apple? Samsung currently advertises their largest LPDDR chip at 32GB, Micron at 16GB.

The M4 Max can use four 32GB LPDDR to get the 128GB total RAM.

Apple could have put 16 32GB LPDDR on the M3 Ultra daughter board (that is home to the SoC) but that means doubling up the memory rows on the board. Sounds tricky to me because the PCB will need to route a great many lines in a small space, but I guess that is why Apple has high-paid EEs.
 
the SC performance hit is a good enough trade off give the MC and GPU performance gains.
Looking at the spec page, the M3 Ultra Studio can support four 8k monitors, while the M4 Max, like in the MBP, only supports one 8k monitor.

So, for those who just happen to have four Apple XDR displays collecting dust in the closet....
 
I guess going from an m2ultra to this will make no sense :(

Depends upon workload. There was generational changes to hardware ray tracing starting at M3-series. M3 picked up AV1 decode. Workload pressing the max RAM capacity of M2 Ultra.

In contrast, workload totally confined to P cores is more incremental, evolutionary changes.

It is far more so aimed at M1 (and Intel systems; large screen iMac , Mac Pro 2019 ) user base. And folks that grew out of coverage with their M2 Max (or lower) systems.
 
In my opinion, the M3 Ultra is the server chip that is being deployed in their AI farm. Previously, I had not anticipated seeing this technology in a commercial product.

Apple stating that the M3 Ultra is two M3 Max (probably a slight variant to laptop version) doesn't really make this a 'server chip' in any significant way. More so the chip that Apple is using in their servers. ( very similar way have deployed Minis/MacPro to provision the Apple XCode cloud stuff.)

The motherboard might be different , but the SoC doubtful. At leas for the inference Private Cloud Computing stuff.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.