MP All Models 40 core ASI with PCI-E slot is around the corner!!

[innerproduct]

So you want to put a server oriented CPU inside a workstation? It doesn’t seems smart.

Please try to broaden your views on this a little. The point here was that an arm based cpu/gpu solution for mac pro doesn’t at all have to be a derivative of m1/m2 but could rather be something similar to the grace/hopper solution that nvidia builds. Seems to me like a high performance system that is in line with what one should be expecting from the richest company in the world when they are totally free to make a workstation for the future.
For clarity’s sake: no, I do not expect this actual SoC to be a part of the new macpro.

 

[PineappleCake]

That's hilarious, only a very very small set of people need that much memory, even in a workstation.
You can be sure that Apple knows exactly the percentage on MacPro 7.1 user that have installed 1.5TB, and I would be extremely surprised to discover that more than 5% of the user have done that.
You can build a workstation with less than 500GB of RAM and be sure that it will cover 95% of the workstation market needs.
You can start a pool and ask how much RAM people have installed on the 7.1 or on their PC workstations.

I know the YouTuber MKBHD uses over 750GB of RAM when he used a Mac Pro 7,1.

 

[PineappleCake]

The latest Intel Xeons support up to 4TB of memory. So 1TB is totally doable if Apple wants too

 

[Mago]

RAM capacity is not matter on how much do I need now, but on how much probably I'll need in foreseeable workstation lifetime.

When I read comments on "500 MB are enough RAM for 95% workstations users", is when I realize there's are fake workstation user among us.

a server class SOC as Nvidia's Ada Lovelace is something way different than workstation, these specialized Apu or dpu are designed along it's consumer needs and its optimization parameters , there are large models you can split among multiple low RAM nodes without losses in efficiency and there are models that solves quite fast in large RAM nodes all depends on the specific algorithm you're running, Even there are models this barely uses RAM but bus speed (as signal analysis) and models that requires huge ram and relatively low processing power (as some crypto miners).

The Main selling point on workstation is flexibility you can Taylor it to quite niche or extreme use cases while mainstream personal computer's just care on what that 90% does (sending the remaining 10% to workstation market)

 

[sirio76]

I know the YouTuber MKBHD uses over 750GB of RAM when he used a Mac Pro 7,1.

So just because a YouTuber use over 750GB everybody should do the same? Beside the fact that I still want to see how it uses that much memory for video editing Youtube videos..

Again, the fact that Intel supports up to 4TB doesn't mean that anyone should.

 

[Apple Knowledge Navigator]

I know the YouTuber MKBHD uses over 750GB of RAM when he used a Mac Pro 7,1.

He never said that needed all that RAM on a daily basis, only that 750gb was what was installed.

 

[sirio76]

When I read comments on "500 MB are enough RAM for 95% workstations users", is when I realize there's are fake workstation user among us.

Sorry but that's hilarious too, professional workstation users typically keep their system for 3 years, only amateur and small freelance keeps their system for a decade. In 3 year is very unlikely that your RAM needs will quadruple, unless you have made a terrible choice from the start and have acquired a system with way less RAM than your typical workflow requires.

If you tell me that it would be good to have user upgradable RAM I agree with you, since I prefer to avoid OEM taxes.

 

[fuchsdh]

I know the YouTuber MKBHD uses over 750GB of RAM when he used a Mac Pro 7,1.

I frankly doubt MKBHD actually used that much RAM, though. In the grand scheme of workloads, 8K video editing is certainly hefty, but I don't see where you'd actually be utilizing that much memory. Granted that MacOS likes to hog as much RAM as possible in case it's needed for utilization, but I would be highly surprised if he was actually running into serious memory constraints with much less RAM, even if he's one of those guys who likes to leave every single program he might use in a week open all the time.

For comparison, I can't max out 128GB doing 4K video tasks, and processor and GPU are far more important there for encodes and the like than RAM.

 

[AlphaCentauri]

I’m writing it as a music producer, there’s never enough RAM for large sample libraries. I have 64GB now and I will need at least 128-256GB for my next computer. And I’m not doing ML, AI, 3D or fluid simulations, just making music… 🤣

 

[sirio76]

I’m writing it as a music producer, there’s never enough RAM for large sample libraries. I have 64GB now and I will need at least 128-256GB for my next computer. And I’m not doing ML, AI, 3D or fluid simulations, just making music… 🤣

Then why don't you install 256GB?

 

[Mago]

Sorry but that's hilarious too, professional workstation users typically keep their system for 3 years, only amateur and small freelance keeps their system for a decade. In 3 year is very unlikely that your RAM needs will quadruple, unless you have made a terrible choice from the start and have acquired a system with way less RAM than your typical workflow requires.

If you tell me that it would be good to have user upgradable RAM I agree with you, since I prefer to avoid OEM taxes.

I used to switch every two years angry not being capable on yearly basis.

Corporate workstation users (90%+) rarely touch it's RAM modules even IT department rarely does ram upgrades mostly it's considered a failure, often what does is migraine legacy workstation to less demanding Dept as could be from AI research to cad and rendering. Often that migrations don't need ram upgrades.

What an actual workstation user needs is an rig capable for the workload a project need or the project (s) needs until the next update cycle, often workstation pays itself in matter of months but some unfortunate cases depends on residual value to avoid Red balance.

Indeed a pro workstation user often don't care on diy ram upgrades even by IT department (warranty support is paramount), but capability and resale value (that's not the user but the accountants advice).

So quite easy to spot an gamer pretending being an Pro workstation user.

 

[AlphaCentauri]

Then why don't you install 256GB?

Into what? 🤣

I’m rocking Intel Mac Mini at the moment. I’m waiting for new Mac Pro as I don’t want to be confined into non-upgradable Mac Studio Ultra, if I can avoid it.

 

[theluggage]

When I read comments on "500 MB are enough RAM for 95% workstations users", is when I realize there's are fake workstation user among us.

I think this is all missing the real-world issue. The real questions are (a) whether there are enough potential customers to make it economical for Apple to produce the custom silicon needed for an Apple Silicon "real workstation" and (b) whether that would offer a big enough advantage over generic PC Xeon/Threadripper hardware (which offers a much wider choice and greater specialisation of hardware and software than a one-size-fits-all Mac).

Apple have a great scheme with the M series where, by almost literally chopping bits off or doubling-up Mx Max dies - a single SoC die design covers everything from the 14" MacBook Pro to the Studio Ultra. They're also milking the performance advantage of having integrated GPUs (that are much faster & more MacOS optimised than competing integrated GPUs) and unified RAM, and the low power draw makes for long battery life and/or small, quiet machines.

The wheels seem to come off that for anything beyond the Mx Ultra which might just stretch to 256GB of Unified RAM and maybe 16 lanes of PCIe (going out on a limb and guessing that half of the TB4 ports could be re-configured as 4 lanes each of PCIe) - which might actually be useful for some users, but isn't going to satisfy "real workstation" users who actually need 1TB+ RAM and/or quad high-end discrete GPUs.

Would it break any of the laws of physics for Apple to make their own ARM-based Xeon/Threadripper rival? No - but economically, making a whole new class of processor just for the Mac Pro would be hugely expensive and could easily turn into a vanity project. That would also involve throwing away the advantages of unified RAM/integrated GPUs as well as the economies that come from re-purposing the Mx Max die. To what advantage? A more power-efficient processor - in a machine with a ton of non-low-power DDR5 and 2-4 hot, sweaty AMD Space Heater GPUs? Maybe a few more CPU cores (but AMD are already setting a high bar there)?

The workstation market is probably were Apple Silicon's major party trick - performance vs. power consumption - is least impressive. It's great for laptops and small-form-factor, and becomes important again in the high-density data centre (hence Ampere and Google making server-class ARM chips) - but on a personal tower system who's #1 priority was performance? Not so much.

Only Apple knows how well the 2019 Mac Pro has been selling, and hence what the potential market is. Even that assumes that the existing users are ready to move on to Apple Silicon (which will inevitably kill off a few software products) or even the latest MacOS (sometimes a bigger deal than re-building for ARM). Meanwhile, the desktop computer market is probably dwindling anyway - squeezed between ever more capable laptops and pay-for-what-you-need-when-you-need-it cloud computing.

 

[Mago]

I think this is all missing the real-world issue. The real questions are (a) whether there are enough potential customers to make it economical for Apple to produce the custom silicon needed for an Apple Silicon "real workstation" and (b) whether that would offer a big enough advantage over generic PC Xeon/Threadripper hardware (which offers a much wider choice and greater specialisation of hardware and software than a one-size-fits-all Mac).

I think you have a Gurman idea on how Tech profits, I've no enough spare time to discuss it, wait for the new ASi Mac Pro it will answer your everything you are not aware.

 

[majus]

Today the developer RC of Ventura 13.3 was released, so it should go golden no later than April 4. I think the new Mac Pro will also be announced no later than then.

 

[spaz8]

I have to agree with the 95% of Mac Pro users would put less than 500GB of RAM in their machine. I'm also not sure how Apple knows how much ram we all have (phone home?). I sure wouldn't pay the Apple tax on 1TB of ram. The 64 GB in my trashcan was 3rd party waiting to be installed before the system showed up.

I think there is a significant amount of Mac users that just need/want more than the 128 GB of ram the M1 Ultra offers, and that forces you into the Mac Pro camp. Frankly if you need more than 24 or 32 GB you've got about 3 choices of Apple products and 1 is a M2 laptop.

If the new ASi MP topped out at 256 GB of ram (the minimum I would want in a build), i'd be sad, but I think it would address a lot of users needs for the next 3 years.

I think this is another reason you don't see a M2 Ultra Mac Studio.. a 192 GB Ram Ultra would probably cannibalize a ton of potential Mac Pro buyers.

I'm still hoping for the daughter card approach.

 

[Mago]

If the new ASi MP topped out at 256 GB of ram (the minimum I would want in a build), i'd be sad, but I think it would address a lot of users needs for the next 3 years.

Simply it wouldn't happen as it would be an quite expensive error not for Mac Pro sales, but for the entire macOS-evry_i_sh1tOS, as it would constraint coders creativity and it's related products: AI(most demandingly),AR/XR (creative workflow also ram hungry), and Mac rebirth as scientific/engineering tool (there are now ASi Metal 3 native support in widely used stem-programming languages as Julia, tensorflow pytorch, Rust).

I think even at loss apple will provide a reasonable capable ASi Mac Pro, maybe not as powerful for everything as an threadripper workstation (first generation at least l but enough powerful for most use cases (actually quite rare to see an 48 or 64 threadripper at engineering and most r&d, often 32 cores seems standard Now, but it will change, the problem is despite how multi core hardware is available now there still tons of applications that can't scalate to multiple threads neither process, some by obsolete coding and quite few by algorithmic restrictions, both issues being addressed slowly.

For Mac Pro success it should offer nothing less than current Mac Pro 7,1 in every respect, what is most compromised is the GPGPU as no M2-Ultra/Extreme GPU setup is an alternative to an Nvidia RTX 4070, and this is an shame, but nothing out this world as it has solution on AMD GPUs coming back to Mac systems but likely restrictions to compute/file rendering only, and memory flexibility have always been just apple didn't want you to be aware ASi support discreet DDR5 ram modules, and that 'unified' memory just an fancy branding for the industry old Shared Memory concept quite common in APUs.

 

[Boil]

For Mac Pro success it should offer nothing less than current Mac Pro 7,1 in every respect

I would think the actual number of end users who really need more than 384GB of RAM is quite small, a niche within a niche within a niche...?

Nicheception...?

...what is most compromised is the GPGPU as no M2-Ultra/Extreme GPU setup is an alternative to an Nvidia RTX 4070, and this is an shame, but nothing out this world as it has solution on AMD GPUs coming back to Mac systems but likely restrictions to compute/file rendering only...

This could also be accomplished with an Apple silicon solution, no need to reintroduce AMD to the mix...?

...and memory flexibility have always been just apple didn't want you to be aware ASi support discreet DDR5 ram modules, and that 'unified' memory just an fancy branding for the industry old Shared Memory concept quite common in APUs.

"Shared memory" in APUs usually just allocates a small specific amount of RAM to the iGPU, the Apple solution allows much more RAM to be used by the iGPU, and in a much more flexible manner...?

 

[sirio76]

So quite easy to spot an gamer pretending being an Pro workstation user.

If you are questioning who I am and what I do for living you can see some of my works here in the forum, I've posted something from time to time, you can consider me a very high end architectural visualization professional. And no, I don't game

 

[sirio76]

Into what? 🤣

I’m rocking Intel Mac Mini at the moment. I’m waiting for new Mac Pro as I don’t want to be confined into non-upgradable Mac Studio Ultra, if I can avoid it.

If you need 256GB of RAM and you have bough a Mac Mini, then you have made a bad choice

 

[Mago]

"Shared memory" in APUs usually just allocates a small specific amount of RAM to the iGPU, the Apple solution allows much more RAM to be used by the iGPU, and in a much more flexible manner...?

The concept never restricted ram % dedicated to GPU, it's done for marketing segmentation and Apu simplification as restricting amount in ram means you can discard GPU ram address lines which simplifies the design somehow.

you can consider me a very high end architectural visualization professional. And no, I don't game

With due respect your work at best is an analogy to gaming, I worked at Renderman coding the primitives to what now is very popular even we pioneered in ray tracing all on CPU only. Your work is not more demanding than gaming only when animations are so large it requires true graphics workstations or clusters is when it becomes closer to Pro HPC , you actually never ever aware what does an data science pro with 1 TB of ram, or an mechanical engineer with compute performance shaders or an chemical researcher with Julia and large protein models. I respect your work kid, but quite different mine, I've even had to design my own ASIC and FPGA compute devices if you understand what is that and why someone may require that then maybe we're colleagues.

 

[AlphaCentauri]

If you need 256GB of RAM and you have bough a Mac Mini, then you have made a bad choice

I wanted to buy Mac Pro 7.1 when it was released but due to Apple announcing transition to AS, I’ve decided against it and bought Mini as a “stopgap“ in early 2020, to get me through the turmoil of AS transition and incompatibility. Back then, 64GB was enough for me. Now, not so much.

 

[sirio76]

The concept never restricted ram % dedicated to GPU, it's done for marketing segmentation and Apu simplification as restricting amount in ram means you can discard GPU ram address lines which simplifies the design somehow.

With due respect your work at best is an analogy to gaming, I worked at Renderman coding the primitives to what now is very popular even we pioneered in ray tracing all on CPU only. Your work is not more demanding than gaming only when animations are so large it requires true graphics workstations or clusters is when it becomes closer to Pro HPC , you actually never ever aware what does an data science pro with 1 TB of ram, or an mechanical engineer with compute performance shaders or an chemical researcher with Julia and large protein models. I respect your work kid, but quite different mine, I've even had to design my own ASIC and FPGA compute devices if you understand what is that and why someone may require that then maybe we're colleagues.

That answer tells me that you know very little about modern rendering, my scenes can demand way (and I mean WAY) more RAM resources than any game will ever need. If you do not understand that further discussion are pointless

 

[Mago]

I didn't render, if any one or two images. I coded the software what rendered it, some of my C code still in use, I've worked at HPC most of my professional life (now I'm retired). I know one or two things not just on how software like redshift or Renderman (did you at least read about it?), When I coded some of it it was named Photorealistic Renderman, later simple Pixar's Renderman.

Something still cues me there's an outsider pretending being one of us.

C.u.

 

[PineappleCake]

This is Nvidias 72 core ARM CPU. I am sure Apple can do a 32 core CPU for the Mac Pro.