It will be interesting to see what Apple offers as a reason to get the MacPro over the Studio. I could see just high speed storage slots.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Silicon Mac Pro Configurations: Everything We Know
- Thread starter MacRumors
- Start date
- Sort by reaction score
How this is different than using TB with intel iGPU (which uses shared RAM with CPU)?
You could keep using the iGPU with unified memory, but perhaps disable that, when you need more RAM to CPU and use only dGPU(eGPU), which has plenty of fast enough discreet memory for itself.
You mean something like RAM and SSD expansion? Lemme load it up with 512GB of RAM and a RAID 5 SSD array.🤓🤓🤓
Not really, it is a very small subset that requires more than 128GB RAM, and you can always add SSDs on the side. I was thinking more of something uniquely new.
The way Apple’s designed their Apple Silicon graphics system, the CPU is not expecting to prepare data, then send it over a bus to a GPU. It’s writing the data to its RAM and leaving it there for the GPU to read. The GPU reading from a position on the SoC is quite performant. The GPU attempting to read and modify data over a TB bus (which is what it would have to do the way the systems are currently architected) would be dramatically slower to the point of likely breaking the OS.
24gb max on M2.
Maximum unified RAM by that logic would be:
M2 pro = 48gb
M2 Max = 96gb
M2 Ultra =192gb
M2 Extreme=384gb
Is 384gb RAM enough for a pro machine where the previous generation maxed out at 1,5TB?
I would say no. Apple will have to do something to get more RAM in there.
Maximum unified RAM by that logic would be:
M2 pro = 48gb
M2 Max = 96gb
M2 Ultra =192gb
M2 Extreme=384gb
Is 384gb RAM enough for a pro machine where the previous generation maxed out at 1,5TB?
I would say no. Apple will have to do something to get more RAM in there.
If M-SoC does not currently use PCIe other than for storage controller, they should just implement more PCIe lanes to a next chip.
I wonder why PCIe is fast enough for the rest of computer industry?
Why all new most powerful dGPU's use PCIe? Because PCIe criplles their performance?
This is the bandwidth other systems expect from their PCIe solutions.
This is the bandwidth available to Apple Silicon GPU’s accessing main memory.
M1 68.25 GB/s
M2 100GB/s
M1 Pro 200 GB/s
M1 Max 400 GB/s
Apple’s Silicon systems are architecturally different and that makes their GPU solution so different that one can’t expect that solutions that are acceptable for non-Apple Silicon systems would be acceptable for Apple Silicon systems. So far, Apple’s been indicating that this is the way it’s systems will work going forward, but if they decide t change it, they’ll let us know at WWDC first.
Last edited:
We are stuck thinking M2 series will be the same as M1. ONLY thing we know, is that M2 added 24gb, a very odd number of Max RAM for M2 VS 16gb Max for M1 (single M1 vs single M2 systems).
1x24GB
4x24=96GB
8x24=192GB
16x24=384gb
32x24=768GB
64x24=1532GB
I doubt this is a coincidence. Somehow 24GB for M2 baseline translates to 1,5TB at the very high end M2 Mac Pro.
Apple cant release a M2 Mac Pro with less MAX RAM than the Intel counterpart it replaces.
Last edited:
All I know, is that the top spec M2 Mac Pro could be a monstrous beast. With a matching monstrous price tag.
And why did apple suddenly start pricing the Intel Mac Pro that stupidly high? Perhaps they knew that packing that much silicon in a system WOULD be that expensive down the line, so they started the extreme price hike already with the Intel Mac Pro systems to absorb the shock come Apple Silicon Mac Pro... (Sure Xeons are expensive af too..)
Anyhow, this is the sort of rumours I came to Macrumors for way.. way waaaaay back in the day. At the very least Apple Silicon makes it more interesting again.
And why did apple suddenly start pricing the Intel Mac Pro that stupidly high? Perhaps they knew that packing that much silicon in a system WOULD be that expensive down the line, so they started the extreme price hike already with the Intel Mac Pro systems to absorb the shock come Apple Silicon Mac Pro... (Sure Xeons are expensive af too..)
Anyhow, this is the sort of rumours I came to Macrumors for way.. way waaaaay back in the day. At the very least Apple Silicon makes it more interesting again.
LPDDR5X would allow up to 1TB of RAM with a M2 Extreme (quad M2 Max SoCs) configuration...
macOS cannot address 192 CPU cores...
Apple can do whatever they want, there is nothing that says they HAVE to meet or exceed the 1.5TB maximum RAM of the 2019 Intel (7.1) Mac Pro...
Yeah, the price hike between the 2013 Mac Pro & the 2019 Mac Pro couldn't have anything to do with more expensive components...
- Xeon CPU
- Custom double-sided eight slot motherboard
- Custom 1,400W power supply
- Custom highly-machined desk side chassis
- Custom MPX GPUs
Last edited:
dGPUs do not use "main memory".
You can come up ideas why dGPU's are slow but in real world benchmarking they are not.
Why?
Re-read:
Even though it’s a big difference, it’s not immediately understandable, especially if all ones knowledge of performant GPU architecture comes from studying Nvidia and AMD’s solutions. Fortunately, there are a decent number of articles out there where you can learn why PCIe is not currently an option for Apple Silicon and, unless Apple makes significant changes to their architecture, it not likely to be a future option.
Thank you for telling me that there is a reason, without telling what it is.
In benchmarks A-chips are not faster than other GPUs, so the pcie bandwidth doesn't seem to cause a problem.
I do understand and accept that Apple has not designed A-macs to be used with dGPU.
But I believe, if they would, those who need different or second gpu, would like macs more.
Btw, was Apple the first company to put 2 gpu's in a laptop? Where the other gpu really used shared memory with cpu?
I did tell you what a few differences are, but telling you those doesn’t help you if you don’t understand WHY Apple designed their Silicon that way. There are several good articles out there that describe how Apple Silicon’s GPU works, how it’s different and why you won’t see PCIe options (again, unless Apple makes big changes, which they’re free to do) that should help you with your “why’s”.
So...
There might be 3 scenarios for new mac pro:
There might be 3 scenarios for new mac pro:
- Trashiest can
Even less expandability, but this time maybe with corners.
How well that went last time? - Real expandable workstation, with ryzen perhaps?
- Mac Studio II
With just beefier chips, nothing else.
1 - Mmm... ASi Mac Pro Cube... Mmm...
2 - LOL, no... We moving to Apple silicon now, so how do AMD chips make any sense...?
3 - Double the SoCs, double the ports...
I think we will get M2 Ultra / M2 Extreme-powered ASi Mac Pro headless workstations with at least four PCIe slots...
I hope we also get Apple ASi GPU / GPGPU add-in cards...
I guess @Unregistered 4U has told us, that dGPUs are not possible with AS. Or at least "feasible".
Why, I'm not sure.
Since AS-macs do have TB, which does carry PCIe, I'd guess that enabling dGPU/eGPU would be just a software thing.
I can hear the YouTubers already "my wheels export video faster than your PC"
I'm still wondering why Apple wouldn't include full pciE busses in the future versions of m-chip?
Is there an explanation?
(Not just to lock out 3rd party solutions.)
Btw, doesn't TB-specs require carrying pciE links?
If m-chip's TB does not do this, can Apple still call those ports TB?
Is there an explanation?
(Not just to lock out 3rd party solutions.)
Btw, doesn't TB-specs require carrying pciE links?
If m-chip's TB does not do this, can Apple still call those ports TB?
Last edited: