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 Working on Redesigned Mac Pro With Smaller Form Factor and Apple Silicon Chip
- Thread starter MacRumors
- Start date
- Sort by reaction score
Do I sound like someone who doesn't know or remember why Apple wouldn't make what I/we want ?
Yes.
"... Thunder X3 ...
The design comes in either 1 or 2 socket configurations, and the inter-socket communication uses CCPI (Cavium Cache Coherent Interconnect) in its 3rd generation, with 24 lanes at 28Gbit/s each, between the two sockets. .."
Previous iteration X2 had dual sockets also .
"..
ThunderX2 - Cavium - WikiChip
ThunderX2 is a family of 64-bit multi-core ARM server microprocessors introduced by Cavium in early 2018 succeeding the original ThunderX line.
en.wikichip.org
Also
"... Ampere’s Altra is a realized version of Arm’s Neoverse N1 enterprise core, much like Amazon’s Graviton2, but this time in an 80-core arrangement. ...
..., offers support for FP16 and INT8, supports 8 channels of DDR4-3200 ECC at 2 DIMMs per channel, and up to 4 TiB of memory per socket in a 1P or 2P configuration. Each CPU will offer 128 PCIe 4.0 lanes, 32 of which can be used for socket-to-socket communications implemented with the CCIX protocol over PCIe. This means 50 GB/s in each direction, and 192 PCIe 4.0 lanes in a dual socket system for add-in cards. Each of the PCIe lanes can bifurcate down to x2. ... "
Ampere’s Product List: 80 Cores, up to 3.3 GHz at 250 W; 128 Core in Q4
Generally, yes.
given that 128gb is 8 times smaller than 128GB ... yeah that is way off. (b -- bit , B -- byte ).
The "need" for triple digit RAM is more so primarily driven by the working set data footprint. Bigger data sets need more RAM so that the data is 'closer' to the cores than more persistent ( at rest ) data storage.
Dual socket servers are actually a declining market segment. Approaching the point where going to get > 100 cores per socket. Amazon didn't even try to cover two package systems with their ARM Neoverse N1 implementation. Even the cloud hosting folks start to run out of customer bases that actually need that kind of scale. They exist. Not going away. But also not a generally a growing pool of users.
There is non uniform memory access (NUMA) on the package. Throw multiple package NUMA on top of that and it becomes more issues to chase down in implementation and in OS stack complexity.
Apple is highly unlikely to do any dual socket anything. As mentioned these systems are declining; Apple generally avoids shrinking markets. Additionally, macOS currently can't really handle anything with more than 64 cores. iPhones and iPads don't need anything like that number of application core counts. Neither do MacLaps. Apple probably has extremely low motivation to "fork" the core kernel and scheduler just for a niche of a niche of a niche product. Some monstrous Mac Pro isn't really a driver of major future macOS development paths. Apple is highly likely going to pursue something like 28-38 clocked faster cores before anything along the lines of going wider-and-slower clocks. The stuff that is at the highest end of "embarrasingly parallel" they are also likely to farm out to Tensor/AI/ML cores or to GPU cores. Which has about zero impact on the core application OS kernel schedulers ( so can easily share with the iPhones. ) .
The bigger is though is that Apple is generally into making smaller systems . The Mac Pro 2019 was an exceptional corner case. But that bigger isn't about the CPU. The Mac Pro 2019's 'biggness' is primarily more so driven by having 8 slots ( 4 of which is double wides ... so pragmatically 12 slot widths that contribute to the height (in the vertical orientation. ) . Generally in Macs though Apple likes smaller logic boards so there is more space for batteries in mobile devices.
Thunderbolt v4 has the same speed/bandwidth and generally the same latencies as Thunderbolt 3. If TBv3 on the Mac Pro 2019 ( and v2 on the MP 2013 ) didn't solve the audio guys 'problems' then TBv4 isn't going to be make that much more of a substantive impact.
It isn't really just an "audio guys" problem. One of the more prevelent Mac Pro 2009-2012 slot occupiers in the slot survey in this forum ( not going to dig it up but its buried in the history) is for add-in M.2 SSD ( and storage carriers. ). If Apple isn't going to put M.2 sockets on the logic board..... that demand isn't really going to change. Mac baseline product design is mirroring the iPhone/iPad where there is one and only one nominal storage drive. In the Mac Pro, user base that notion breaks down and not just with "audio guys" subset.
There are a decent variety of PCI-e cards that folks need in slots that fall out of the "audio video" peep-hole that folks want to push the Mac Pro into.
This interconnect does what that PCI-e v4 , v5 , and v6 doesn't cover? Probably a whole lot of nothing. Especially for the use cases listed below
Storage modules that are based on PCI-e. ... does nothing more.
GPGPU with no video output feedback ... does nothing more. (GPU with video out only on the card edge ... ditto )
FPGA that has a chip that has a PCI-e interface on it .... does nothing more.
Neural Engine Module .... if this is Apple's then probably not on a module. Probably soldered to the motherboard.
If Apple goes to mulitple packages in a system. They are likley soldered. The Mac Pro SoC ... quite highly likely soldered if has the Security Enclave processor in it ( just like the T2). If Apple needs multiple chip dies then most likely will get some form of multiple chip modules. All wound up in a package..... not a slot.
The main data transport of a MPX bay in the Mac Pro is via PCI-e. The MPX connector only handles.
Power -- which any for the use cases above the 75W of the standard PCI-e slot is largely more than plenty enough for the usecase.
DisplayPort output to embedded TBv3 (or TBv4 in futre) sockets. -- no of the above really get into that.
PCI-e feeds for the Thundebolt controllers on the card. -- again none of the above really touch that need or issue.
A half size enclosure would mean a refreshed look at the "externally modular" Mac. It would have to reply on Thunderbolt 4 without a doubt to offer the same expandability as the current line.
This could however really affect those with legacy slot in cards and devices.
This could however really affect those with legacy slot in cards and devices.
“not set to come out in 2020 and will presumably be released in 2021 or 2022”
And the follow up model will not be set to come out in 2021 or 2022, but will presumably be released in 2023 or 2024, or 2025. New models will presumably be released when they are released.
And the follow up model will not be set to come out in 2021 or 2022, but will presumably be released in 2023 or 2024, or 2025. New models will presumably be released when they are released.
Depends upon how many they make and how many products stuff a single implementation into. That has worked fantastically well for the 80+ million run rate of the . The chips designers are basically a fixed ( albeit yearly if want to hold onto them long term ) cost.
Paying for something isn't a semiconductor problem. It is a revenue generating problem.
Apple can do it cheaper than Intel if they lower the CPU product diversity much lower than Intel to correspond to their order of magnitude lower volumes. Bow wave off of the iPhone cores work and it gets even cheaper. And if Phones units sales are plateauing adding Macs to the pile of systems paying for basic uniformly shared infrastructure also amortizes the costs over more units also.
Apple would be spending more if tried to match the breath of feature offerings that Intel and AMD have. If Apple drops the Mac line up onto a fewer number of SoC implementations than it is on now then .... that is partially to save money and hence ... yes ... about making more to toss into the Scrooge McDuck money pit. Apple isn't going to radically lower Mac system prices. Apple slapped large 20% markups on Intel processors list prices and passed that on to their customers.
This new "smaller" Mac Pro version is quite suggestive that Apple is cranking down the capabilities to match the SoC rather than "spending more money" . The "full size" Mac Pro would be a real test if Apple really is trying to stay in the same CPU package product diversity or basically executing a long term retreat. That retreat ... saving money is extremely likely a major factor of that.
Even if Apple does a full size Mac Pro SoC if the refresh times are once 3-5 years then once again money is likely a major factor there too. Having overlapping , pipelines design teams so that can pump out new designs every 12-14 months costs more money than just a scalar , non-pipelined team who chuck out a new design every once in a while in their "spare" time. That is less expensive too. ( Mac Pro and iMac Pro going into Rip van Winkle modes ... cheaper to execute ).
Can’t help wonder why it took them the better part of a decade to come out with a replacement for the previous MacPro. Since when it was finally available they took about a 10min break before announcing that the platform is basically dead to them and that the replacement is already on the way..
Where was this sense of urgency back in 2014? 🤔
Where was this sense of urgency back in 2014? 🤔
As long as Apple has entangled the CPU package into keeping the secure enclave and the encryption keys/biometrics/data .... nope ... it will highly likely be soldered to the motherboard.
The I/O bandwidth the CPU package needs to provision is much, much, much greater than what Afterburner deals with. Gould be in a Mac Pro 2009-2012 context where the whole RAM+CPU subassembly comes out. But again run into the current era entanglement of the security subsystem. If the keys to the internal storage drive 'walk off' the data is pragmatically toast.
Long term Afterburner stands a decent chance of just being explicitly designed into the SoC package. Or at least the media subsystem coupled to the GPU. The more it sticks to just doing ProRes only the more likely this will collapse into just explicit implementation once get to a set of ProResRAW capabilities that are mature and are reasonably high demand.
I'd expect a free , single width , bus power slot just because there are more folks with M.2 SSD "caddy" add in cards. The Afterburner would just "happen to fit" that same slot motivated by the larger demand.
One larger fan covering the zone that the CPU and the single , bus power x16 slot is in would be a stretch to characterize as being the "one cooling system.". Both individual components would have decoupled cooling implementations directly attached. Yes there would be a strong dependency of the system provided fan to move air but that is only part of the system and only limited , physically detached coupling.
Don't even have to go to AMD. Just take the 'M' off of that CPU model number and that would be $3-4.5K less.
Pointing at the Intel Xeon 'M' CPU costs is largely just misdirection. Even the Intel 2021 models probably won't charge that much. Those models were priced at as a relatively short term money grab by Intel (leverage a short term feature gap with competition ) to try to collect enough money to 'buy' their way out of the bonehead moves they had made. Long term, likely that is going to turn out to be yet another bonehead move added to the pile. They took that extra money and spent a decent chunk of it on more silliness that didn't address the root cause problem.
The baseline cost of a Mac Pro has a Xeon W chip in it that is in the $799 range. The basic floor price of the Mac Pro isn't not being driven by the high end Xeon W 'M' processor variants that Apple selected. The non CPU only related factors are the bigger drivers.
This is all wrong. The marginal cost of cpus is the cost of wafer starts. So it makes no difference whether apple has less diverse offerings than Intel.
This is the Mac Pro Mini, the Mac Mini Pro would be like the current Mac Mini but with a dedicated GPU.
I'd like to see a tower with less PCIe slots and less RAM options for a lower price. I guess I'm not the only one but my guess Apple had this is store to offer a overpowerfull tower first then a smaller version for less.
The iMac Pro was a panacea, a binky, a nuck. It was made to scam enough 'pro users' to maybe keep them quieter. It's not a 'pro machine'. It has thermal issues, and adding memory is basically a complete disassemble. I think it looks like a panicked moment for Apple to 'put something out there', and not a real 'here's red meat for pro users'.
They packed them with lots of memory, and barely acceptable video, but it's 'black', and everything matches. *shrug* It could have been worse I suppose. I have one. I've had issues with it freaking out. Hard to tell if it's heat, or just what it was designed to do. I had two cases open with Apple, but since the workload has plummeted, it's been pretty stable. It was fun getting it to the point it was rolling over and whimpering. I was surprised how quickly it did that, but my Mac Pro (5,1) would have been a puddle by then. I just didn't expect the total psychotic break it had. Sad... *shrug*
I would have loved to own a new new new new Mac Pro, heck, even a trash can, but too rich for me. I don't have enough jam for that... My first house was $20,000. Spending over that for one computer is noting I can do, but if it's that capable, I could put it to work, at times.
The haves have not a clue...
Maybe when they hit ebay...
Given that they can fit an A14 inside a teeny tiny iPhone which has single core performance higher than that of a Mac Pro, I can’t see them having issues fitting one with a bunch more Cores that still requires minimal cooling and space. Now remember there’s a GPU in an iPhone too and it isn’t all that terrible. I bet Apple could scale up that GPU tech too while still requiring minimal space.
Heck, they could shrink a Mac Pro quite substantially while still having room for a large GPU like the 3090 or 6900. The current Mac Pro is pretty huge mostly because of the many PCI slots it accommodates. A half sized Mac Pro would still fit the majority of what people need.
Heck, they could shrink a Mac Pro quite substantially while still having room for a large GPU like the 3090 or 6900. The current Mac Pro is pretty huge mostly because of the many PCI slots it accommodates. A half sized Mac Pro would still fit the majority of what people need.
Sigh. Fixed costs and variable costs are different. You are the one off in the weeds wrong. Wagging your finger at wafer costs ( which are variable costs ) is plain misdirection.
Marginal costs are a component of variable costs. The costs of making one more additional die is largely going to be dominated by the wafer costs. But you need something to make in volume quantities before start to incur that marginal costs. The R&D and set-up for that production run is a fixed cost. That fixed costs also needs to be paid for. For low enough production runs that is significant factor. All that design time occurred when that product was generating no revenue. When the run rate is high enough the contribution to total costs relatively drops to a lower contribution.
Can use some hand waving "rob Peter to pay Paul" rationalization that the current products pay the upfront for the future products.
Apple is paying a humungous amount for wafer starts. But they are also getting humungous amount of revenue for the working dies that come out of those wafer starts to. And there really isn't relatively that much of a time gap between money spent and money earned. In this narrow aspect, Intel really wasn't buying them much in recent years. Apple couldn't buy 7nm from Intel. So really didn't make a difference how much less Intel was charging.
The yearly salary for a high end silicon design engineer is way more than a even the new relatively much more expensive 5nm wafer starts. The cost is incurred for multiple years before start genering any revenue. The wafer s being more expensive for bleeding edge nodes means ( if end users prices are about the same) means there actually less margin to pay for the fixed cost overhead out of the wafer.
[ Apple is usually not using process shrinks to make the dies smaller. They have been packing more "stuff" in to approximately the same size dies. when the wafer costs go up and end user price stays the same the margin actually goes down. ]
Apple regularly stuff "hand me down" A-series SoC into multple products. Partially because they can get away with it if the SoC is "overkill" for the lower performance tier product. But it is also a way of increasing margins by using already "paid for" SoC for yet another product.
I hope to make my 2019 Mac Pro last 5 years into 2025. I feel like at that point, it has easily earned it's worth, so I hope everything is fully supported (are you listening Adobe) at least until then.
Technically, I had one good job earlier in the year that paid for it outright, but I want it to actually earn me money over time.
I won't be mad if they announce an ARM Mac Pro however. It's a good sign that they're looking to make this architecture work for pro users. I feel like they need to make this point as the initial ARM products (iphone/ipad) could be blamed for Apple's lack of focus on the pro market some time ago.
Technically, I had one good job earlier in the year that paid for it outright, but I want it to actually earn me money over time.
I won't be mad if they announce an ARM Mac Pro however. It's a good sign that they're looking to make this architecture work for pro users. I feel like they need to make this point as the initial ARM products (iphone/ipad) could be blamed for Apple's lack of focus on the pro market some time ago.