MP All Models Realistically when could a revised/new Mac Pro be announced?

[kvic]

This is illogical. Pci4 doubles band with but isn’t important but When pci5 doubles performance it will be important.

Disagree. Both are giant bumps and both are super important. Pci6 will again double throughput and will be hugely important. Ssd speeds alone make it important. Now it isn’t clear what will happen with pci7, and one wonders could they can possibly keep doubling performance like this forever, or will we hit a plateaued.

People who use MacPro7,1 as large disk-based database sort of applications in a server-like environment will see benefit from the increased PCIe bandwidth in v4.0. I don't know if this is popular among MacPro7,1 user base as they could be better served by PC workstations like Threadripper or Threadripper Pro.

Hardware applications such as Apple Afterburner Card can't even saturate PCIe 3.0. With v4.0, Apple can stamp two chips to support 12 streams (instead of 6 atm) of 8K ProRes raw? I doubt Apple has the incentive to do so in a niche segment of an already niche market.

Server guys always want higher PCIe bandwidth that's driving the progress. And progresses were first made by AMD a couple years ago. Intel was forced to follow and a bit late to the party. People are focussing on PCIe 5.0 not only because of its further increased bandwidth but it opens up new (hardware) applications never seen before. Read up Compute Express Link (CXL) if you're interested.

On PC side, workstations are easy to receive PCIe 5.0 (and CXL) support. On Mac Pro's with Apple chips, I'm not so sure. Perhaps Apple Car or some Apple Secret Project will take off to justify to do so.

 

[deconstruct60]

W-3300 series indeed comes with PCIe 4.0, for example, 32-core Xeon W-3365.

Personally I won't expect PCIe 4.0 something to be excited about. Workloads such as 4K video production (perhaps the most common & demanding workload a Mac Pro faces in daily operations) could hardly see benefit both GPU & storage performance wise.

In some context it will. First, MPX Duo modules. With PCI-e v3 you have two. x16 PCI-e v4 GPUs trying to share a single x16 PCI-e v3 backchannel to the CPU ( so that is about a x8 v3 split between them). With PCI-e v4 you would pragmatically have a x16 v3 split between them. The DUO would be no worse than the single PCI-e v4 x16 GPU modules in that respect. It is going to be much easier to actually get the line speed ups.

Second, Afterburner's inputs and output are two vastly different sizes.

PCI-e v3 x16 --> 128Gb/s.

One uncompressed 10 bit color 8K 60Hz data stream. -->. 62 Gb/s.
" " " @ 30Hz data stream --> 30 Gb/s.

4 * (the 30Hz stream ) --->. 120 Gb/s.

Technically not completely saturating the x16 input, but if have to walk and chew gum at the same time ( where something else from CPU also needs substantive bandwidth). Afterburner probably caps out at four , highest fidelity, 8K streams for a reason other than the FPGA ran out of "horsepower" .

PCIe 4.0 is a nice bump in the spec. Due to prolonged lifespan of v3.0, PCIe 4.0 will be very short lived and be replaced by PCIe 5.0 in workstations in 2023. PCIe 5.0 is something to be very excited about in add-in card markets/applications in general. IMO, PCIe 5.0 is a huge uncertainty if Apple will tit-for-tat follow the PC industry trend.

Will Apple even. tit-for-tat for. PCI-e v4 at more than x1 bundles over the short term? Pretty good chance that Apple doesn't want to go there and that an Intel model that does helps "kick the can" to 3rd or later generation M-series.

If Apple is out to kill off 3rd party dGPUs (if not dGPUs altogether ) then that will take lots of 'air' out of the sails to more rapid adoption of the versions past v4.

One of the immediate benefits of the new Ice Lake Mac Pro is the memory sub-system, both capacity & performance wise. All SKUs come with 8 memory channels (instead of 6 in MacPro7,1) and support max capacity of 4TB (instead of 0.7-1TB in MacPro7,1).

Pretty doubtful Apple is going to add more DIMM slots to the motherboard. The cheapest , quickest path forward for them would be to either keep the 12 DIMM slots have now with a different controller to slot mapping. ( e.g., straight 1-to-1 mapping for 4 slots and 2-to-1 mapping for another 4 slots ( 4 + 2*4 = 12 ). Akin to the 3 1:1 and 1 2:1 of MP5,1 . Or just backsliding back to 8 DIMM slots and leveraging higher single DIMMs to reach new relatively higher cap, but less than max of other systems. )

Apple probably won't promote 4TB because M-series is going to be no where even remotely close to that. If they stick with soldered on SoC memory modules , then probably caps out at 512GB or so. If W-3300 models tops out at 2-3TB that is still many multiples over where the "half sized" Mac Pro will peak out at.

One more thing.. I'm guessing the refreshed Mac Pro will come with an exclusive option of the mystically absent W6900X Duo, and support for next-gen & highest-end AMD Radeon RX 7000 series GPUs. Something to be very excited about.

IMHO the DUO W6900X is highly unlikely to ever ship. The 6800 DUO has more performance over the 6900 on workloads that scale in both core and/or VRAM memory utilization and and is $1000 cheaper to boot. It has significantly better $/Performance. By pricing the single 6900 at $1000 higher , they painted themselves into a corner. Doubling the 6900 price point isn't gong to be competitive. Apple OCD probably will hold that to a doubling in price ( high but highly consistent pricing model policy ).

Apple probably has some availability problem with the 6900. They are priced to reduce demand while adding a hefty contribution to the Apple "Scrooge McDuck" money pit. Apple is probably taking some of the potential 6900 Duo profits via the singles at the inflated price. Some users are going to have software which paints them into the corner of "have to" buy the 6900.


As for next gen RX 7000 support for macOS intel. If Apple keeps AMD GPUs completely frozen out of M-series systems I wouldn't hold my breath on those coming. Even more so if Apple releases a "half sized" Mac Pro on M-series that slices off most of the bottom half of what is left over of the Mac Pro userbase.

The more RX 7000 (and higher) series GPUs look to leverage the x16 PCI-e v4+ connection for shared memory traffic the more deployments into x4 Thunderbolt external enclosures are going to be more limited. If Apple is cutting down the number of Mac Pro AMD can deploy to also then at some point the relationship is quite likely to crumble.

If the relationship is already on choppy waters that too is probably another contributor to the 6900 odd ball pricing.

So I believe MacPro7,1 will depreciate a little faster as the refreshed Mac Pro will be quite interesting, definitely not boring. I guess this is good news for a lot of people on this sub-forum as most seem to be tinkerers who do not need the horse power of a latest & highest-end Mac Pro.

If Apple holds to their current path of "no dGPUs on M-series" then MP7,1 systems probably won't crater in pricing anytime soon. Many of the folks who jumped from. 5,1 to 7,1 probably won't like the "half sized" Mac Pro that Apple is going to pitch. There probably won't be a wholesale stampede off 7,1. However, folks who stretched budgets to buy the 7,1 aren't going to rush off either.

Most likely the 8,1 is being more targeted at the MP 2008-2013. hold outs who still haven't moved. And a few of the iMac Pro folks in their upgrade zone who weren't particularly happy with the iMac Pro limitations.

The overall number of 8,1 sold will probably be lower than the number of. 7,1 sold when compare the two lifecycle sales over two years . The 8,1 may sell more if Apple goes back into "Rip van Winkle" mode and just doesn't retire it for more than several years. Especially, if Apple comes in with an incrementally more affordable "half sized" model to take a deep slice of the "Mac Pro userbase".

 

[deconstruct60]

People who use MacPro7,1 as large disk-based database sort of applications in a server-like environment will see benefit from the increased PCIe bandwidth in v4.0. I don't know if this is popular among MacPro7,1 user base as they could be better served by PC workstations like Threadripper or Threadripper Pro.

Databases are more about random I/O. Mac Pro systems tend to have heavier streaming sequential data streams I/O issues to tackles. That can still drive higher end bandwidth.

Hardware applications such as Apple Afterburner Card can't even saturate PCIe 3.0. With v4.0, Apple can stamp two chips to support 12 streams (instead of 6 atm) of 8K ProRes raw? I doubt Apple has the incentive to do so in a niche segment of an already niche market.

four completely uncompress HDR 8K video streams aren't trivial. Even more so if trying to push that to a Duo MPX module where there are two GPU data consumers that may have other concurrent data streams to pull in.

I would not be surprised though is Apple lowered the development costs and only provisioned PCI-e v4 to slots 1 and 3 though. Keeping the Afterburner sitting on a better backhaul provisioned x16 PCI-e v3 link. If Apple feeds two x16 PCi-e v4 links into an updated Plex PCI-e switch that is better foundation to provision out 5-6 slots of PCI-e v3 to than two x16 v3 are now. The Plex switch gets highly bandwidth oversubscribed if put in 2-3 fast SSD disks along with 1-2 Afterburner cards. PCI-e v4 could dramatically reduce that bandwidth pressure without having to roll it all the way out to all the end user slots.

Server guys always want higher PCIe bandwidth that's driving the progress. And progresses were first made by AMD a couple years ago.

Chuckle. IBM Power 9 had PCI-e v4 back in 2017. AMD wasn't 'first' here at all. PCI-e v4 isn't "dying prematurely" here . It is more so that PCI-e v3 is being dropped very late by lower end systems.

Intel was forced to follow and a bit late to the party. People are focussing on PCIe 5.0 not only because of its further increased bandwidth but it opens up new (hardware) applications never seen before. Read up Compute Express Link (CXL) if you're interested.

Intel's rapid skip to PCI-e v5 has more to do with their multiple log jammed product development processes than normal evolutionary adoption rate for PCI-e versions. Not a good place to set the expectation metric.

Intel is also likely going to leverage CXL a bit like how AMD/Nvidia leverage Infinity Fabric / NVlink. This time though making a trade-off to be more open standards so can loop in more 3rd parties. CXL+PCI-e v5/6 will have some overlaps in replacing Intel's QPI link ( for connecting CPU packages ) as Intel puts more multiple dies inside the package. The PCI-e foundation link will be left on the outside interpackage connection workloads (and possibily be used on some inside links at PCI-e v6 levels) .

On PC side, workstations are easy to receive PCIe 5.0 (and CXL) support. On Mac Pro's with Apple chips, I'm not so sure. Perhaps Apple Car or some Apple Secret Project will take off to justify to do so.

PCI-e v5 and 6 have issues traveling as far as PCI-e v2 (and v3) did. CXL products commonly seen in the aisle at BestBuy/Walmart/Microcenter probably won't happen over the next several years.

If Apple is going to coupled two separate packages that are far apart into a shared memory construct , they are pretty likely to go the proprietary route as AMD/Nvidia.

 

[kvic]

In some context it will. First, MPX Duo modules. With PCI-e v3 you have two. x16 PCI-e v4 GPUs trying to share a single x16 PCI-e v3 backchannel to the CPU ( so that is about a x8 v3 split between them). With PCI-e v4 you would pragmatically have a x16 v3 split between them. The DUO would be no worse than the single PCI-e v4 x16 GPU modules in that respect. It is going to be much easier to actually get the line speed ups.

Ice Lake W-3300 has the same 64 lanes but PCIe 4.0 speed for each lane. Hence, I think the slot config and bandwidth allocations will remain the same in the refreshed Mac Pro. I would be very surprised for the price of MacPro7,1, Apple can't absorb the extra cost and limit PCIe 4.0 to two slots.

PCIe bandwidth is not as fungible as people would think they're. On paper, 8-lane PCIe 4.0 equal to 16-lane PCIe 3.0. Add-in card designs aren't flexible enough to work optimally in both cases. Add-in cards require delicately designed to take advantage of increased bandwidth. Current generation of GPUs show PCIe 4.0 makes little difference in performance though they operate at v4.0 speed. Whether next-gen Radeon RX7000 can make better use of v4.0 speed remains to be seen. I believe Mac Pro owners will be able to find out, perhaps even Radeon RX8000 series that follow.

Second, Afterburner's inputs and output are two vastly different sizes.

PCI-e v3 x16 --> 128Gb/s.

One uncompressed 10 bit color 8K 60Hz data stream. -->. 62 Gb/s.
" " " @ 30Hz data stream --> 30 Gb/s.

4 * (the 30Hz stream ) --->. 120 Gb/s.

Technically not completely saturating the x16 input, but if have to walk and chew gum at the same time ( where something else from CPU also needs substantive bandwidth). Afterburner probably caps out at four , highest fidelity, 8K streams for a reason other than the FPGA ran out of "horsepower" .

10-bit 4:4:4 @60Hz 8K = 80Gb/s
10-bit 4:4:4 @30Hz 8K = 40Gb/s
10-bit 4:2:0 @60Hz 8K = 40Gb/s
10-bit 4:2:0 @30Hz 8K = 20Gb/s
8-bit 4:2:0 @30Hz 8K = 18Gb/s

Guess what their test conditions were when Apple said Afterburner Card could handle 6 simultaneous 8K streams.

The transition away from x86 is quite different from PowerPC. x86 is and will remain as the dominant player in productivity and data centres for foreseeable future. The cut-off timing of x86 Mac may well depend partially on market acceptance of the half-size Mac Pro with Apple chips as well as how well plans for Apple Car or some Apple Secret Project work out.

 

[Morgonaut]

Performance of the new upcoming W-3300 series CPUs are nothing impressive, sometimes it even falls behind 2 generations old 3175. So if I should buy new Mac Pro then I would wait for the Apple Silicon one. But my current 2019 MP I believe will serve me well also in next few years. If it do the job now very well and the job will not dramatically change, then there's no reason to upgrade.

 

[ZombiePhysicist]

Performance of the new upcoming W-3300 series CPUs are nothing impressive, sometimes it even falls behind 2 generations old 3175. So if I should buy new Mac Pro then I would wait for the Apple Silicon one. But my current 2019 MP I believe will serve me well also in next few years. If it do the job now very well and the job will not dramatically change, then there's no reason to upgrade.

Despite my lust for the likely PCIe4 and extra cores, I'm not upgrading from my 2019 either (unless in the hyper unlikely scenario that they would offer an upgrade path to swap motherboards). There are pro's and con's to getting the last intel machine they likely will ever make. For those that need the intel compatibility, it may be worth a go. Some that held off on the 2019 might see the more modern guts as a reason to jump on it.

Even now knowing the pci4 version is coming, I think made the right choice. I got my 2019 in 2019, so getting 2 years of use for me was well worth it. And it's far from clear that I'll want to jump on the first apple chip Mac Pro.

The annoying thing is I might have to get the first apple chip Mac Pro in that, with every Mac Pro and apple's amazing hatred of the pro/enthusiast market, you're never sure if it may be the last Mac Pro (see Mac book, iMac Pro, cube, etc.).

 

[Morgonaut]

With PCIe gen 4 most people will barely see any benefit. Better wait and spend the money on Apple Silicon MP next year

 

[ZombiePhysicist]

With PCIe gen 4 most people will barely see any benefit. Better wait and spend the money on Apple Silicon MP next year

Disagree. Basically a huge proportion of Mac Pro owners do a LOT to get the faster SSDs on their system. PCI4 double's that speed. This makes MOST EVERYTHING DONE with regard to storage on the Mac Pro that much better.

 

[Morgonaut]

Disagree. Basically a huge proportion of Mac Pro owners do a LOT to get the faster SSDs on their system. PCI4 double's that speed. This makes MOST EVERYTHING DONE with regard to storage on the Mac Pro that much better.

If you so disagree do you think you need MILLIONS of MB/S for good performance? ) No. I do 6K video production with terrabytes of footage for each project and all I need is speeds up to 10 Gb/s and low latency and that's what I have even ever network. I tested super fast 4x NMVe in stripe array even on my channel and I saw no real benefit of having this disk array in my Mac Pro and waste PCIe slots over running our whole production over network.
Super fast internal storage does not mean super fast performance or whole fast production.

 

[ZombiePhysicist]

If you so disagree do you think you need MILLIONS of MB/S for good performance? ) No. I do 6K video production with terrabytes of footage for each project and all I need is speeds up to 10 Gb/s and low latency and that's what I have even ever network. I tested super fast 4x NMVe in stripe array even on my channel and I saw no real benefit of having this disk array in my Mac Pro and waste PCIe slots over running our whole production over network.
Super fast internal storage does not mean super fast performance or whole fast production.

Yes, I do. 6k. Drones shoot 8k these days. You might want to sit down. This will be shocking:

Your use case is not everyone else's use case.

And yes, super fact performance does mean a whole lot faster for moving files, for manipulating files, for going through lots of files.

Your argument is basically no one would ever want more than 640kb of RAM.

 

[Morgonaut]

Yes, I do. 6k. Drones shoot 8k these days. You might want to sit down. This will be shocking:

Your use case is not everyone else's use case.

And yes, super fact performance does mean a whole lot faster for moving files, for manipulating files, for going through lots of files.

Your argument is basically no one would ever want more than 640kb of RAM.

what are you talking about?? I never said anything about memory like you force to my mouth. Use what you want. Pay much more if you want. I just save on smarter places and my workflow is freaking fast and balanced and optimized. If you feel you need "millions of MB/s" per second, then buy any hardware you want. I just said it's not needed with good workflow. So looks like yours is probably not as good as you think if you need those "millions MB/s"

 

[deconstruct60]

Ice Lake W-3300 has the same 64 lanes but PCIe 4.0 speed for each lane. Hence, I think the slot config and bandwidth allocations will remain the same in the refreshed Mac Pro. I would be very surprised for the price of MacPro7,1, Apple can't absorb the extra cost and limit PCIe 4.0 to two slots.

Slot's 1 and 3 are "odd ball" for a typical PCI-e v4 system design since there are about as far as possible away from the CPU package as you can get on a motherboard. Apple very likely needs some PCI-e re-drivers to get the PCI-e v4 signal down there at full bandwidth performance.

The PLEX switch is closer. Depending upon the route taken to it it might need some re-drivers also. The switch itself is pragmatically a huge re-driver as a side effect. So there could be some 'free' wins there for at least a few of the end user slots it manages. Even if not though the backhaul increase in bandwidth is a win. ( lower needs to manually juggle the bandwidth slot allocations across pool A or B. )

There is lots of other stuff on the Mac Pro board though that have to route around and deal with. There are no wires so all the power is going through the board. There are 4 DisplayPort video output paths being mux'ed to the 4 different TB ports on the system.


Apple probably wouldn't have to absorb much there. The T2 (and associated drives modules ) are yet another year older and more paid for. Swapping that for some redrivers and incrementally more expensive PLEX switch could be pretty close to an even trade.

PCIe bandwidth is not as fungible as people would think they're. On paper, 8-lane PCIe 4.0 equal to 16-lane PCIe 3.0. Add-in card designs aren't flexible enough to work optimally in both cases. Add-in cards require delicately designed to take advantage of increased bandwidth. Current generation of GPUs show PCIe 4.0 makes little difference in performance though they operate at v4.0 speed. Whether next-gen Radeon RX7000 can make better use of v4.0 speed remains to be seen. I believe Mac Pro owners will be able to find out, perhaps even Radeon RX8000 series that follow.

PCI-e switches make it far more fungible than you are trying to portray here. The up/down shifting between PCI-e v4 and v3 can be managed by the switch. Similar with adjusting the lane width. For slots 2 , 4 , 5-8 it is more about how they share the bandwidth. The card design doesn't have to be in the loop on that at all ; as it is handled by the switch.

The Pool A and B are oversubscribed in more than a few Mac Pro configurations. PCI-e v4 added to the backhaul will be a substantive and quite transparent difference with zero changes to the PCI-e cards added.

10-bit 4:4:4 @60Hz 8K = 80Gb/s
10-bit 4:4:4 @30Hz 8K = 40Gb/s
10-bit 4:2:0 @60Hz 8K = 40Gb/s
10-bit 4:2:0 @30Hz 8K = 20Gb/s
8-bit 4:2:0 @30Hz 8K = 18Gb/s

Guess what their test conditions were when Apple said Afterburner Card could handle 6 simultaneous 8K streams.

If to get to 6 streams Apple has to turn on chroma compression how does that make Afterburner a low bandwidth card. The decoder of the 6 RAW streams has all of the chroma info when decoding a RAW file. Some of that is being dropped on the floor to push 6 streams ( crank the number up to a more sexy marketing number) . The Afterburner engine isn't the limiter there. The conduit off the card is the limiter.

That's why I mentioned uncompressed.

Your assertion back in post 26 was :

...Hardware applications such as Apple Afterburner Card can't even saturate PCIe 3.0. With v4.0, Apple can stamp two chips to support 12 streams (instead of 6 atm) of 8K ProRes raw? I doubt Apple has the incentive to do so in a niche segment of an already niche market. ...

Realistically when could a revised/new Mac Pro be announced?

People who use MacPro7,1 as large disk-based database sort of applications in a server-like environment will see benefit from the increased PCIe bandwidth in v4.0. I don't know if this is popular among MacPro7,1 user base as they could be better served by PC workstations like Threadripper or...

forums.macrumors.com

Afterbrner is not saturating PCI-e v3 because it is lossy compressing the data . Pragmatically Afterburner is a quite high consuming PCI-e v3 bandwidth consuming card (on the output side). Primarily, because it only does part of the problem that needs to be addressed. It decodes RAW files, but it doesn't display them. To actually see what is being decoded it has to be passed via PCI-e v3 to another PCI-e v3 card ( a GPU card). That GPU card has its own input data requirements for the rest of the tasks it is being assigned. That multiple screen Afterburner output plus the rest of the data the GPU needs sets a pragmatic limit what is "high" and "saturating".


An updated Afterburner that was PCi-e v4 could get closer to the max 8K screen count with higher gamut output. That would have decent synergy given Apple is also trying to sell their "extreme' gamut $7-8K screens coupled to the the Afterburner cards. Would Apple "have to" push out an updated Afterburner card with a W-3300 powered Mac Pro? Probably not. However, the synergy with the very high gamut displays would go up if they did. ( also wouldn't hurt if Apple could shave $500+ off the price also while evolving the product. )

The transition away from x86 is quite different from PowerPC. x86 is and will remain as the dominant player in productivity and data centres for foreseeable future. The cut-off timing of x86 Mac may well depend partially on market acceptance of the half-size Mac Pro with Apple chips as well as how well plans for Apple Car or some Apple Secret Project work out.

The Apple Car is immaterial to the Mac Pro future. The dynamics in the Mac product space are what is signiificant. If Apple increased their MBP sales by 10% and the Mac Pro sales dropped by 10% , Apple would probably take that trade off all day long. MBP sales up by 5% and iMac up by 5% and Mac Pro sales drop by 10% .... same thing.

A "half sized" Mac Pro leveraging MBP oriented chiplets would be the cheapest path to hanging in a smaller Mac Pro user base market for Apple longer term.

One of the primary goals of the M-series SoCs is to put "old school" Mac Pro like performance inside of the smaller enclosures of the rest of the Mac line up. If Apple isn't pulling substantive number of "old school" Mac Pro users out of their old systems and into these newer smaller enclosures then they are not really accomplishing their primary goal. ( move laptop performance dramatic up and move desktop performance more so in the energy efficient zone. (the latter is clearly represented by the dramatic thinning out of the smaller screen iMac (the new 24" iMac). )

The "half sized" Mac Pro is quite likely going to be the more affordable option. ( e.g. something like $4K versus $6K. Or even more of a gap of $3.5K versus $6.5K ). When Apple increased the entry price 100% from the MP 2013 levels, Apple left a substantive set of users behind. If Apple doesn't bring dGPU to those half sized model they won't get all of those back but probably will get a substantial number whose add-in card needs are primarily non-GPUs.


The other long term big issues for "max slot" Mac Pro long term is if Apple continues to be Apple GPU only on macOS on M-series. Even more so if that stays iGPU only ( or hyper proprietary off SoC augment "dGPU" that presents as a iGPU). In contrast, if Apple lets AMD (and/or Intel) back into the 3rd party GPU driver building process then there are better evolutionary dynamics there for the desktop Mac GPUs. But as long as laptop GPU dynamics are Apple's only concern that is a big problem for the full size Mac Pro.


Apple's semi-secret project besides the car is the VR/AR glasses/headset. Bluntly, that project just likely puts Apple deeper into the monomaniacal iGPU focus. Foval (adaptive ) raytracing assist hardware and/or AI/ML upscaling resolution packed into an on package GPU. That is just more iGPU focus pressure that will look even better on bigger die allocation iGPUs.

 

[Korican100]

I don't know what you use your Mac Pro for, but an M1 mini is more than enough to tide you over until a newer MP comes out.

doesn't m1 struggle with anything besides hvec in fcpx w/ effects and transitions etc