M4 Mac Announcements

Anyone know the model no. of the M4 Max Studio (i.e., ##,#), so I could search for it in GB? I don't expect they gave it a higher max clock than the M4 Max MBP, but would like to confirm.
Mac16,9 but I’m not sure there are any results in the searchable database yet. Scores were only submitted a few hours ago and it only gets updated every 12 hours I believe.
 
There are M4 Max Studio scores there, but they can’t be searched easily.
1741381668503.png


1741381741966.png
 
HBM in particular depends on what Baltra is supposed to be and how much overlap/dual-use design Apple wants with its consumer chips.
The thing is Baltra shouldn't overlap with their consumer chips, it should be seperate line and made for server use only.

Its most likely the most boring option, that it will be a 64 core, 160GPU with 1TB or 2TB of LPDRR5X with 2TB/s of Mem Bandwidth. Apple also has this Carbon Zero so they likely won't even use server based IP or boost up clocks.

The 4x M4 Maxes will be connected via advanced packaging simliar to AMDs. In the end Apple's "server" chip is based a Mac SoC.
 
The thing is Baltra shouldn't overlap with their consumer chips, it should be seperate line and made for server use only.

Its most likely the most boring option, that it will be a 64 core, 160GPU with 1TB or 2TB of LPDRR5X with 2TB/s of Mem Bandwidth. Apple also has this Carbon Zero so they likely won't even use server based IP or boost up clocks.

The 4x M4 Maxes will be connected via advanced packaging simliar to AMDs. In the end Apple's "server" chip is based a Mac SoC.
In that case, LPDDR5X and multiple M4 Maxes, I’d classify it as dual use :) - I was referring to building a bespoke die with HBM just for server use that I view as less likely. But, given that Apple seemingly went back to modify the M3 Max die to make the M3 Ultra, I could also potentially see them modifying a “server” die with HBM to use a consumer variant that is otherwise the same with LPDDR. Though I suspect that’s still a bigger change than what they did with the M3 Max, adding ultra fusion and TB5.
 
Last edited:
There are M4 Max Studio scores there, but they can’t be searched easily.
View attachment 34117

View attachment 34118
Thanks! The M4 Studio has the same 4.5 GHz max clock as the M4 MBP, as expected. Would have been cool if they were able to take advantage of the far greater thermal dissipation capability of the Studio to bump it, but alas....
 
Has anyone raced a MBP Max against a Studio Max? I mean, marathon, not sprint. Perhaps the Studio can hold the higher clock for longer?
 
Has anyone raced a MBP Max against a Studio Max? I mean, marathon, not sprint. Perhaps the Studio can hold the higher clock for longer?
I would say undoubtably better than the 14", but in my experience the 16" holds clocks high fairly consistently, although I don't have experience with its M4 Max variant specifically
 
Something that jumped out at me....

I wonder if Apple will start offering the ability to rent compute time in their private cloud being powered by M3 Ultra during the next WWDC.

This would definitely be interesting to me as a path for Apple to further explore and push availability to private LLM compute for the masses (over here the 512GB memory ultra with the 16TB ssd is close to 20k!)
 
Something that jumped out at me....

I wonder if Apple will start offering the ability to rent compute time in their private cloud being powered by M3 Ultra during the next WWDC.

This would definitely be interesting to me as a path for Apple to further explore and push availability to private LLM compute for the masses (over here the 512GB memory ultra with the 16TB ssd is close to 20k!)
I expect MacStadium will offer that as soon as they get their hands on the M3 Ultra Studios (see https://www.macstadium.com/ and https://www.macstadium.com/bare-metal-mac ).
 
Apple should allow CPU overclocking on the Studio or at least the Mac Pro. All that cooling just to run at the same clocks as a MacBook Pro.
 
All that cooling just to run at the same clocks as a MacBook Pro.
Clocking is not all there is too it, though. If you are running heavy loads, the SoC will down-clock when it starts to get too hot. All that cooling means that the Studio probably down-clocks later, or less, compared to the MBP.
 
Apple should allow CPU overclocking on the Studio or at least the Mac Pro. All that cooling just to run at the same clocks as a MacBook Pro.
Sometimes when you are using a SOC architecture, you may not have designed your PLL/clock dividers in such a way to allow the CPU clock to scale up without also scaling up other components that shouldn’t be scaled up. You can also run into aliasing problems - you increase the speed of the CPU, but whenever your CPU has to talk to other components, they can’t keep up, your FIFO buffers fill up, and then you have to stall. You can design around it (bigger FIFOs, more fine-grained clock domains, etc.) but then you are doing a lot of work that helps you for a very low volume product and which does nothing in your high volume product.
 
Sometimes when you are using a SOC architecture, you may not have designed your PLL/clock dividers in such a way to allow the CPU clock to scale up without also scaling up other components that shouldn’t be scaled up. You can also run into aliasing problems - you increase the speed of the CPU, but whenever your CPU has to talk to other components, they can’t keep up, your FIFO buffers fill up, and then you have to stall. You can design around it (bigger FIFOs, more fine-grained clock domains, etc.) but then you are doing a lot of work that helps you for a very low volume product and which does nothing in your high volume product.
This is a great explanation. Thank you.
 
I expect MacStadium will offer that as soon as they get their hands on the M3 Ultra Studios (see https://www.macstadium.com/ and https://www.macstadium.com/bare-metal-mac ).
Oh I’m sure you’re right. More what I was getting at is that the obvious market for the m3 ultra is those wanting the 512GB unified memory for very specific workloads like LLM. Apple is making such a big thing about AI and more recent WWDCs saw Apple push more into developing cloud services with Xcode cloud https://developer.apple.com/xcode-cloud/

I wouldn’t be surprised to see Macstadium sherlocked and rentable m3 ultras being made available direct from Apple at either this or next years WWDC. The ultra chip almost looks ideal for such workloads and it gives Apple another cloud service revenue stream. That’s where my 0.02 goes anyway!

Edit: I like Macstadium and certainly not advocating or wanting them to be sherlocked!
 
Sometimes when you are using a SOC architecture, you may not have designed your PLL/clock dividers in such a way to allow the CPU clock to scale up without also scaling up other components that shouldn’t be scaled up. You can also run into aliasing problems - you increase the speed of the CPU, but whenever your CPU has to talk to other components, they can’t keep up, your FIFO buffers fill up, and then you have to stall. You can design around it (bigger FIFOs, more fine-grained clock domains, etc.) but then you are doing a lot of work that helps you for a very low volume product and which does nothing in your high volume product.
Is it sufficient to do the needed modifications on a per-core basis, or does this require global changes to the chip? E.g., would you need to increase the voltage across the entire chip to enable boosting on a single core?

I ask because I'm wondering if you could create one or two super-P-cores, capable of running at very high clocks, into the Pro and Max chips, without having to modify the chip as a whole. If so, this wouldn't be that costly.

As you know, since most apps continue to be single-threaded (including CPU-demanding apps like Mathematica, Maya, and AutoCAD) (are there also significantly CPU-sensitive games?), even boosting just a single core can have significant practical value. How much you enable it could be device-dependent, providing further product differentiation. E.g., maybe moderate boost on the Pro Mini and Pro/Max MBP's, and high boost on the Studios.

What about Intel chips? They appear to allow significant overclocking, beyond their stated specs. Are they wastefully over-provisioning their PLL/clock dividers, FIFO buffers, etc. to enable this?

And separately, what about boosting the all-core clocks on the GPU's--what considerations would apply there?
 
Last edited:
Back
Top