That is a terrible benchmark for performance. People use Mac Pro's for far more demanding work than Logic Pro and Final Cut Pro. And these machines do not exist in a vacuum. ARM CPU's are more than capable of scaling into the 100's of cores and Apple should be pitting their pro machine against the best workstations.
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.
Bloomberg: Apple Silicon Mac Pro May Feature Up to 32 High Performance Cores, 128 Core Option for Graphics
- Thread starter MacRumors
- Start date
- Sort by reaction score
That is a terrible benchmark for performance. People use Mac Pro's for far more demanding work than Logic Pro and Final Cut Pro. And these machines do not exist in a vacuum. ARM CPU's are more than capable of scaling into the 100's of cores and Apple should be pitting their pro machine against the best workstations.
This is in line with my expectations. No particular predictive skill, in my mind I was thinking that the current ones are already at the top end of what consumers can reasonably purchase. So, they wouldn’t have to add an insane number of CPU to maintain that spot. BUT, I’d think that, to provide an analog to AMD/Nvidia, they’d likely go heavy on the GPU. We’ll see if this pans out. I mean it IS from Big Hack Bloomberg.
The way I read it:
The AS-MPro will use a SoC that has several M3x chiplets (similar to some AMD CPUs) so there is no need to design special MPro-only silicon (apart from what is used to connect the chiplets).
The top end MBP will come with a SoC with 1 or maybe 2 of these chiplets.
I also predict that upgrading the RAM after the fact will be exclusive to the Intel-MPro.
128cores and no Realtime Raytracing 🤣, so it's useless for me.
Well they haven't said no hardware ray tracing just yet, lets hold out hope...
If you can fit everything on one piece of silicon, you are always better off doing so. Chiplets (which is just a marketing term for a form of MCM, or multi-chip module) are a necessary evil for when you cannot. MCM designs also may provide you with more supply chain flexibility - mix and match parts so you don’t guess wrong which combinations will need to be dabbed in greatest volumes, etc. But there’s never a performance or power advantage to do that.
Well, I mean, check Apple’s website for the MacPro. You won’t see ANY comparisons to anything other than the prior fastest MacPro. Apple doesn’t care that you can buy a faster computer elsewhere and hasn’t for awhile. They know that, if you’re using macOS, then UTMOST PERFORMANCE is of less interest than the familiarity with the OS. Because, everyone that needs UTMOST PERFORMANCE moved on to a more flexible system that provides it 3-4 years ago, probably longer. They’ve already replaced all those customers with new customers anyway (around half of Mac sales are to folks that have never owned a Mac before).
I’m sure there are folks that figure “Apple will do just like the other CPU vendors when it comes to the high end”. But, based on the reality of the Apple Silicon systems actually for sale today, it’s clear that their vision is starkly different from AMD/Intel. I mean, you could pretty much say, “What would Intel do?” and the answer would be the first thing to exclude
350+ watts and hyperthreading are two things that come to mind 
Sounds like it could be a winner, almost sounds like a Mac mini pro unit and not a replacement for the full tower. Or maybe they will have both. If they can come in with a mid size desktop tower at around $3000 starter it should sell well.
But they are going to have to have like 6 thunderbolt 4 ports to make up for the loss of expansion slots.
But they are going to have to have like 6 thunderbolt 4 ports to make up for the loss of expansion slots.
Well, I mean, check Apple’s website for the MacPro. You won’t see ANY comparisons to anything other than the prior fastest MacPro. Apple doesn’t care that you can buy a faster computer elsewhere and hasn’t for awhile. They know that, if you’re using macOS, then UTMOST PERFORMANCE is of less interest than the familiarity with the OS. Because, everyone that needs UTMOST PERFORMANCE moved on to a more flexible system that provides it 3-4 years ago, probably longer. They’ve already replaced all those customers with new customers anyway (around half of Mac sales are to folks that have never owned a Mac before).
I’m sure there are folks that figure “Apple will do just like the other CPU vendors when it comes to the high end”. But, based on the reality of the Apple Silicon systems actually for sale today, it’s clear that their vision is starkly different from AMD/Intel. I mean, you could pretty much say, “What would Intel do?” and the answer would be the first thing to exclude
At the time of the new Mac Pro it was the most powerful Intel Xeon you could shove in the machine as where the 2010 models. The people buying machines with over 1 TB RAM and 28 cores haven't suddenly went away and are now looking for even more powerful machines. Apple putting out only 32 cores on the top end Mac Pro means they are leaving a lot of performance on the table. Many professionals need the utmost performance from macOS machines and if 32 cores is the top end the cynic in me jumps to the conclusion that the 64 core and 128 core machines are being artificially held back to make new SKU's in the future.
... but the RAM and SSD will be "unified" and soldered to one assembly...
It seems pretty likely, if the specs in this post are true.
There will be a CPU chiplet with 16 large cores and 4 small cores.
Some MacPros will include 1 of these, others will include 2.
There may be separate chiplets for the I/O and GPU, or these may be integrated into a single non-CPU chiplet.
Or these functions may be distributed across the CPU chiplets, like first generation AMD Zen (which may mean 32+8 MacPro will have more I/O than 16+4).
GPU may (indeed it's most likely thinking about it) even be off of the CPU for Mac Pro, so they can ship more powerful discrete GPUs suitable for a workstation in varying configurations.
... but the RAM and SSD will be "unified" and soldered to one assembly...
That might actually be interesting. I would love to see something like a mini blade server where you just slot in compute cards that have some cores and RAM that are able to communicate with the others in the system as one unit. I don't believe it is possible just now, and multi socket designs always have design issues, but it would still be a cool take on a pro machine that you can upgrade by just slotting in new blades.
With chiplets you can bin (select for performance and power characteristics) at the chiplet level, and pair the best chiplets together when assembling.
This means there is a performance and/or power advantage to using chiplets.
Additionally at the moment TSMC have good yields even on 5nm, but smaller silicon always yields better, so this is another gain - if yield wasn't so good, the mathematics for smaller chiplets improve a lot.
And you can reduce the number of different silicon designs you need to do, which may not be an issue for Apple given the amount of R&D money they have, but there is always a headcount limitation.
The downside is losing on-die communications. You can use a silicon interposer to route signals, without losing too much (often use for HBM memories), or through package signalling, with a bandwidth/latency limitation (AMD do the latter in multi-die Zen processor implementations, you will see the monolithic APUs get better latency figures).
That is pretty amazing graphics performance. If it scales with the number of cores, which it should because graphics operations are highly parallelizable, here are some comparisons.
Current M1, with 8 graphics cores, does 2.6 TFlops. That is just below an NVidia GTX 1650 at 2.9 TFlops.
Extrapolating, 64 graphics cores would do 20.8 TFlops, which is just over the NVidia RTX 3070, at 20 TFlops.
128 graphics cores would do 41.6 TFlops, which beats the NVidia RTX 3090, at 36 TFlops.
Current M1, with 8 graphics cores, does 2.6 TFlops. That is just below an NVidia GTX 1650 at 2.9 TFlops.
Extrapolating, 64 graphics cores would do 20.8 TFlops, which is just over the NVidia RTX 3070, at 20 TFlops.
128 graphics cores would do 41.6 TFlops, which beats the NVidia RTX 3090, at 36 TFlops.
So Apple's headless desktop offerings will consist of:
$600 M1 mini with 8 cores.
$x,xxx Mac Pro with 20 or 40 cores.
For the love of pete, can we just get something in the $2,000 range that has 10 or 12 cores?
$600 M1 mini with 8 cores.
$x,xxx Mac Pro with 20 or 40 cores.
For the love of pete, can we just get something in the $2,000 range that has 10 or 12 cores?
It would just be nice to have a blazing fast (and efficient) Mac that shuts the PC gear heads up for a while. They'll still squeal about price, but it would be great to blow some Intel/AMD doors off...
Yes, and it hasn’t been the most powerful Intel Xeon in plenty of years. The Mac Pro and the iMac Pro have been the fastest Macs, but they’re not the fastest PC’s you can buy. Apple Silicon may not change that. The story doesn’t mention RAM capacity, but there’s no reason why it couldn’t have a TB of RAM, 32 CPU cores and 128 GPU cores. If a user hasn’t bought a Mac since 2010, I’m fairly certain a 32 core Apple Silicon will beat a 28 core Intel processor from 2010.
These systems WILL provide the utmost performance for macOS. They will absolutely be the fastest macOS machines available when they ship. If anyone uses macOS, NEEDS performance, but pauses their purchase because they want to see 64 before they buy… KNOWING that these have higher performance than the system they’re sitting in front of… Well, then I’d assume they don’t really NEED performance.
Are we just not going to talk about this claim? While I currently have more use for an Intel processor and thus this would be a potential benefit to me, it seems like a very odd move at this stage of the game.
Nothing will shut PC gear heads up. Folks will just buy the Mac with the performance they require and enjoy it! I think “delighting our customers and not “shutting up PC gear heads” is higher on Apple’s list of requirements for their future endeavors.
I truly hope that Apple will create a high-end Mac Mini using the 8+2 chip rumoured for the MBP, although it will be priced higher.
More cores than that, I don't know.
Jade 2C and 4c is suggestive that this is just one new die with interconnection built in and that they are doing chiplets. So this new Mac Pro is just 2 or 4 MBP 16" dies glued together. It is just one much bigger SoC package ( relatively speaking. Could still be in the 900mm range for four dies. TSMC is suppose to roll out some multiple chip packaging that spans 1200-1400mm so still room fro RAM around the edges. )
So this would be a "Mac Pro" with an iGPU and soldered on RAM. Wouldn't be surprising for Apple's playbook. One die multiple products. The Jade-C Chop die for the "small core count" MBP could even just be a die with one "hemisphere" turned off to harvest yield on some defects ( and ). Or perhaps an actually chopped off hemisphere that is cheaper to make. ( and Apple expects to sell volume on the Mini's , iMacs , and MBP's ).
If users are herded into buying Apple RAM these "Processors" ( really bundled SoCs) probably won't be cheap. ( good chance up in the legacy iMac Pro price range for system prices. )
The NeXT cube had slots. But a PCI-e slotless , soldered RAM Mac Pro is to draw a large number of "boo's and hisses" from a large fraction of the Mac Pro crowd. That is drifting back the MP 2013 ethos which got lots of "hate" for many years. ( actually somewhat even more of backslide if the RAM is soldered down ).
Half sized Mac Pro 2019 design could just be dumping the 2nd MPX bay and probably most of the non MPX bay slots ( slots 5-7 or 6-8 ) . If packing soldered on RAM to 4 substantively large-ish dies then the package size is still going to be big. So the CPU subsection of the 2019 design would get reused. The SoC probably doesn't have the I/O bandwidth to support all those additional slots. ( may even need an PCI-e hub add-on to the die to get the fan out for just 2-4 slots (e.g. an MPX bay , x16 , x4-custom I/O ) . The W580-W5700 range like iGPU means don't need to take up a PCI-e slot just for a Apple default GPU to be installed. So they can drop a MPX Bay and still have some empty slots for those that want them.
But yeah could see Apple coming back with a MP 2013 ethos design as a repise of the "can't innovate my ass" prance around the stage.
WWDC will tell if Apple opens up the 3rd party GPU driver window for anyone else but themselves. If it still locked down then probably bad for the Mac Pro ( at least the user internal expandable functionality Mac Pro ).
[/QUOTE]
Last edited:
I don't think it's odd at all.
There are (few) people who do need those insane amounts of RAM, PCIe GFX-cards and soon, but I don't see Apple investing in developing such a complex motherboard for a niche of a niche.
With a Xenon based MPro they can start with whatever eval board/schematic Intel sends them which is at least 2/3 of the way.
Congrats on the least perceptive comment of the day.
If the M1 Mac Mini was going to be the only Mac Mini, it would already be the only Mac Mini. Look at the Intel Mac Minis that are left in the lineup, they will replace those models with Apple Silicon model(s). It's literally impossible for Apple to telegraph their Mac Mini plans more obviously.