Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
https://www.pcbuildersclub.com/en/2018/12/vega-2-what-amd-might-do-with-the-brand/

It can be a bit confusing with all the fuss about Navi, but I am pretty sure the mMP will have at least an option with a 7nm Vega, very likely two or three, given some combinations of 56 and 64 cores and the VRAM, as Apple already does with the iMP.

Coming up with a Navi-based mMP would seem a bit strange, given the first cards to arrive seem to be the mid- and low-specced that will be followed by the more powerful ones by the end of the year (if not later). On the other side, mounting an almost 2 years old Vega 14nm would just mean shooting on their foot (especially given how the stocks are doing), if they want to satisfy the anticipation they contributed feeding.

The question is: given it supports PCI-e 4, could it be a sign that Apple might move to the EPYC platform (and stick with the PCI slot) or that they want to implement PCI-e 4 on a proprietary daughterboard to allow it to work with a Cascade Lake Xeon?

I know they could just go with PCI-e 3, but wouldn't it be a wise marketing choice future-proofing the mMP, since we wouldn't expect them to upgrade it at least for another 18-24 months. (look at the iMP) By then, all pro GPUs will very likely have migrated to PCI-e 4.

Do you think that along with Vega 7nm, Apple will include a base model with a Navi card?
 
To get this thread back on track, if that is really even possible...

Lisa Su has said that 7nm Vega is not for the consumer market, & the new MI50 & MI60 cards are labeled as "datacenter GPUs" & have no external monitor connections...
They have Video Outs ;).

aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9GL1AvODA5MjY5L29yaWdpbmFsL1JhZGVvbi1JbnN0aW5jdC1NSTYwLmpwZw==
 
They have Video Outs ;).

aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9GL1AvODA5MjY5L29yaWdpbmFsL1JhZGVvbi1JbnN0aW5jdC1NSTYwLmpwZw==

Maybe an early prototype or engineering sample that needed the external monitor connection for testing purposes, but the exploded view image on the AMD website for the Radeon Instinct MI60 "datacenter GPU" shows no external port opening on the rear bracket nor ports on the PCB...

9g65vKJ.png
 
Maybe an early prototype or engineering sample that needed the external monitor connection for testing purposes, but the exploded view image on the AMD website for the Radeon Instinct MI60 "datacenter GPU" shows no external port opening on the rear bracket nor ports on the PCB...

9g65vKJ.png
You are showing renders of the GPU. renders do not have Video outs. Actual GPU, and this is not a render, because you can see the markings on the PCB:
mini-02_amd_radeon_instinct_mi60.jpg

Compare this to what David Wang holds in his hand during the Next Horizon Event:
AMD-Radeon-Instinct-MI60-01-770x489.jpg

Its the same GPU, and it also has the same Video Out.
 
You are showing renders of the GPU. renders do not have Video outs. Actual GPU, and this is not a render, because you can see the markings on the PCB.
It wouldn't be hard to render PCB markings.... It's also possible that it has mDP ports for factory diagnostics - but those ports are disabled in normal mode.

Anyway, ATI doesn't list "Output Ports" on the datasheets for the Instinct series - so it's all guesses.
 
https://www.pcbuildersclub.com/en/2018/12/vega-2-what-amd-might-do-with-the-brand/

It can be a bit confusing with all the fuss about Navi, but I am pretty sure the mMP will have at least an option with a 7nm Vega, very likely two or three, given some combinations of 56 and 64 cores and the VRAM, as Apple already does with the iMP.

Coming up with a Navi-based mMP would seem a bit strange, given the first cards to arrive seem to be the mid- and low-specced that will be followed by the more powerful ones by the end of the year (if not later). On the other side, mounting an almost 2 years old Vega 14nm would just mean shooting on their foot (especially given how the stocks are doing), if they want to satisfy the anticipation they contributed feeding.

The question is: given it supports PCI-e 4, could it be a sign that Apple might move to the EPYC platform (and stick with the PCI slot) or that they want to implement PCI-e 4 on a proprietary daughterboard to allow it to work with a Cascade Lake Xeon?

I know they could just go with PCI-e 3, but wouldn't it be a wise marketing choice future-proofing the mMP, since we wouldn't expect them to upgrade it at least for another 18-24 months. (look at the iMP) By then, all pro GPUs will very likely have migrated to PCI-e 4.

Do you think that along with Vega 7nm, Apple will include a base model with a Navi card?

PCIe 3 versus 4 seems like a distinction without a difference for most people, given the performance of modern cards even in old v2 cheese grater systems. It's not a real bottleneck, the same way RAM speed doesn't make a huge difference.
 
Maybe with the tMP, but not with the MBP and iMP. They had AMD Vega 56/64 and 20 even prior to mass marketing.
 
Maybe with the tMP, but not with the MBP and iMP. They had AMD Vega 56/64 and 20 even prior to mass marketing.
There were never high-end GPU's for any Mac.
AMD Vega did not change that.
Apple can get them cheap, because Nvidia can sell all their chips at a price point they want.
AMD cannot, so they make supply contracts with Apple for the volume.
 
There were never high-end GPU's for any Mac.
AMD Vega did not change that.

That doesn't change what I said :) Whatever card you think is the best, they still came out with up to date cards and not 2 year old ones.

Vega cards have been out since May 2017.

Actually, they were released in August, but Apple was the first to disclose them when they previewed the iMP at WWDC.
 
That doesn't change what I said :) Whatever card you think is the best, they still came out with up to date cards and not 2 year old ones.



Actually, they were released in August, but Apple was the first to disclose them when they previewed the iMP at WWDC.

The Frontier Edition cards were out in May. The workstation equivalents were soon after. Not that it matters that much as there are only a few specific workloads where they are competitive.
 
AMD cannot, so they make supply contracts with Apple for the volume.
AMD charges 9000$ for MI60. And they are selling that silicon, in HPC space.
The Frontier Edition cards were out in May. The workstation equivalents were soon after. Not that it matters that much as there are only a few specific workloads where they are competitive.
Last time I checked, they are competitive in far more workloads, only not CUDA.
 
To get this thread back on track, if that is really even possible...

Lisa Su has said that 7nm Vega is not for the consumer market, & the new MI50 & MI60 cards are labeled as "datacenter GPUs" & have no external monitor connections...

A substantive issue is that those MI50 and MI60 cards don't even have Windows drivers ( last I looked. ), let alone macOS ones. [ Linux x86_64 only. https://www.amd.com/en/products/professional-graphics/instinct-mi60

"...
OS Support Linux x86_64 ...
... "

]

Apple would have to put in very substantive investment just to change the vector on that is a very simple way on the software support side. [ Let alone custom hardware. ]



But the iMac Pro is based on the Radeon Pro Vega 56 & 64, and these are custom packages for Apple, as are the MBP Vega GPUs...?

Custom packages how? Binned at reduced clocks isn't a new package. Mac specific drivers? Again not really a package change.


So, what I am getting at is, do you think we might get "special to Apple" Radeon Pro Vega II 7nm GPUs as options in the forthcoming modular Mac Pro...?!?

No. Well at least it would be a dubious idea for Apple to chase down the rabbit hole. Three reasons.

First ....

And keep in mind we are getting Small Navi GPUs coming up, with Big Navi behind that...

So why? For a Mac Pro in 2020-2021 do a "big Navi" (if there is such a thing with that architecture) , but for the grossly late Mac Pro just do Vega without the clock binning and a perhaps a "mid' (really mid ... I think "smaller than the big" is closer to what AMD is aiming at. ). Perhaps use those as the entry-mid cards. That would have been a sketchy bet back in 2017 if they were shooting for a end of 2018 solution, but perhaps they were targeting 2019 all along.


Second, if have an open PCI-e x16 slot and some decent power supply capacity then the next Mac Pro could use the new MI50 and MI60 cards as compute cards. They don't have to be perfectly symmetrical to the primary display GPU subsystem. (e.g, could go Navi/Vega for the display GPU and then have a "cost insensitive" Vega II for top level Compute card option. ).

It doesn't make sense for Apple to try to spend gobs of money trying to pound a round peg into a square hole to make the Vega II into a single display GPU product when AMD is putting close to zero effort into doing that. If Apple wanted to throw megabucks at a highly custom display oriented GPU from AMD then just do a early custom "big" Navi for single display GPU. The volume of the Mac Pro though would make it highly dubious that would get anywhere near close to breakeven for a custom GPU that was purely restricted to the Mac Pro. Where a custom GPU is far more likely is in the significantly (like order of magnitude ) higher volume laptops. And Vega II makes no sense at all for those.

Third , an open slot would also be applicable to Mac Mini , iMac (and Pro) , MBP , etc. well a beefy external PCI-e expansion box ( some refer to as eGPU). So again the same slot cards from MI50/MI60 with Thunderbolt applicable driver updates could also work an empty slot inside the new Mac Pro. ( so not restricted to Mac Pro only volumes in the Mac system sales space. )


The Big Navi GPUs are what I could see Apple offering (but only thru them for the "proper" card...) as part of the "modular" aspect of the new Mac Pro...

Big Navi is probably more eventually important for iMacs than Mac Pros. But right now the Mac Pro is grossly late. If Big Navi means "close to 2020" it is largely the wrong option. Apple is chronically 2-3 tier systems on driver availability and anything coming late in 2019 is highly likely to slide into 2020 if software stack is delayed.


The last sentence presupposes that we actually get the new Mac Pro sooner rather than later come 2019...

The core issue with "Sooner rather than later" is that AMD's '12nm' stuff would be better for the Mac Pro. Not the 7hm stuff. Long term the next Mac Pro needs better, future GPUs, but the far more immediate problem for the next Mac Pro is that it isn't shipping period. Apple probably needs a custom main display GPU card like right now far more than than they heed a highly custom GPU package at some point late in 2019. They could work with a GPU package that just exists now (late 2018).


If the mMP is later in 2019, then we might just see them released with Big Navi GPU options...?!?

late 2019 gives the mid Navi cards a shot. It would help to keep some of the Mac Pro configs in the historical range first-second quadrant pricing to have some GDDR6 cards in the mix. HBM2 isn't going to keep the costs down and production scaling up.
[doublepost=1545879876][/doublepost]
YActual GPU, and this is not a render, because you can see the markings on the PCB:
mini-02_amd_radeon_instinct_mi60.jpg

Compare this to what David Wang holds in his hand during the Next Horizon Event:
AMD-Radeon-Instinct-MI60-01-770x489.jpg

Its the same GPU, and it also has the same Video Out.


One video output isn't going to cut it for a 2018 era Mac Pro card. There were at 2-4 stage back in 2010, so backsliding to one won't cut it. The other major problem is that physical outside the box probably doesn't cut it either ( due highly likely to integration with Thunderbolt issues. ). If the GPU still has multiple outs on the chip package then possible , but that doesn't mean the baseline driver infrastructure would have robust support for that though.

Apple doesn't need a GPU targeted at being a "Compute" GPU tasked with being a primary Display GPU. Overly entangling those two roles is one of the things they screwed up on that Mac Pro 2013 design. Repeating that again is an extremely dubious move.
[doublepost=1545880589][/doublepost]
....

Actually, they were released in August, but Apple was the first to disclose them when they previewed the iMP at WWDC.

The iMac Pro wasn't released until December (with real substantive volume in 2018). Showing engineering demo units isn't "releasing". AMD had leaking to the press long before WWDC 2017.

May 2017
"... AMD has once again confirmed that they will be launching both the Naples server CPU and the Vega GPU architecture in this quarter. ..."
https://www.anandtech.com/show/11319/amd-releases-q1-2017-earnings-vega-and-naples-in-q2

End of May 2017 Computex

"... Radeon Vega Frontier Edition, Launching June 27th ... "
https://www.anandtech.com/show/11476/computex-2017-amd-press-event-live-blog-starts-10pm-et


Also even earlier in January 2017

" ...
To that end, AMD is looking to address all of these factors with Vega. Which is not to say that this is everything – this is a teaser, after all – but this is where AMD is starting. Where they are going to be with their next generation architecture and how they believe it will address the changes in the market. So without further ado, let’s take a teasing look at what the future has in store for AMD’s GPUs.
..."
https://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser


The notion that Apple's WWDC Vega announced was a undiscussed , low visibility GPU is a huge stretch. Yes it wasn't "1-2 years behind" but they were never out in front ( more than 6+ months ) either.
 
Last edited:
So what’s our next potential announcement, reveal, sneak peak date? Anything realistic before WWDC?
 
Custom packages how? Binned at reduced clocks isn't a new package. Mac specific drivers? Again not really a package change.

I knew when I typed "package" someone was going to take it literally..

Look at the logic board for the iMac Pro, look at where the GPU is integrated... That is not something one can buy off of the shelf & plug into a PC, hence a "custom package"...

And the iMac Pro offerings are the Radeon Pro Vega 56 (w/8GB HBM2) & the Radeon Pro Vega 64 (w/16Gb HBM2)...

Neither of those are off the shelf stick it in my PC items either...

These are custom items for Apple, most likely based off of the WX 8200 (Radeon Pro Vega 56) & the WX 9100 (Radeon Pro Vega 64) workstation GPUs...

So why? For a Mac Pro in 2020-2021 do a "big Navi" (if there is such a thing with that architecture), but for the grossly late Mac Pro just do Vega without the clock binning and a perhaps a "mid' (really mid ... I think "smaller than the big" is closer to what AMD is aiming at). Perhaps use those as the entry-mid cards. That would have been a sketchy bet back in 2017 if they were shooting for a end of 2018 solution, but perhaps they were targeting 2019 all along.

Seeing as how the iMac Pro is using WX 8200 / WX 9100 GPU derivatives, it sets a precedent for what Apple could/might/should place in the new Mac Pros...

But as I have been reminded, Mac Pro users are not all 3d/video pros... The folks wanting Mac Pros for audio work can get by just fine with less GPU horsepower...

So this is where the mid-range Small Navi GPUs can come in handy; I would hope for the Radeon RX 3080 GPU as the baseline for the new Mac Pros, with workstation-oriented versions of 7nm Vega GPUs for the high-end...

The Big Navi GPUs would be a replacement / aff-on option (modular Mac Pros) down the road, as would the Next Gen AMD GPUs that come after Small & Big Navi...

Second, if have an open PCI-e x16 slot and some decent power supply capacity then the next Mac Pro could use the new MI50 and MI60 cards as compute cards. They don't have to be perfectly symmetrical to the primary display GPU subsystem. (e.g, could go Navi/Vega for the display GPU and then have a "cost insensitive" Vega II for top level Compute card option).

Compute cards as an option are a no brainer, just NOT as a required built-in like in the current stagnating Mac Pro...

It doesn't make sense for Apple to try to spend gobs of money trying to pound a round peg into a square hole to make the Vega II into a single display GPU product when AMD is putting close to zero effort into doing that. If Apple wanted to throw megabucks at a highly custom display oriented GPU from AMD then just do a early custom "big" Navi for single display GPU. The volume of the Mac Pro though would make it highly dubious that would get anywhere near close to breakeven for a custom GPU that was purely restricted to the Mac Pro. Where a custom GPU is far more likely is in the significantly (like order of magnitude) higher volume laptops. And Vega II makes no sense at all for those.

We have no clue as to what Apple is doing / going to do with Vega II... We have no clue (beyond the "datacenter GPUs") what AMD is doing / going to do with Vega II...

For all we know, the reason we have heard little regarding Vega II aside from the MIxx products is because Apple & AMD are keeping things a secret until announcement...?

Third , an open slot would also be applicable to Mac mini, iMac (and Pro), MBP, etc. well a beefy external PCI-e expansion box ( some refer to as eGPU). So again the same slot cards from MI50/MI60 with Thunderbolt applicable driver updates could also work an empty slot inside the new Mac Pro. (so not restricted to Mac Pro only volumes in the Mac system sales space.)

If I am paying 9000 bucks for a compute card, I am going to want to get every penny of performance out of it; so it will be in a PCIe slot in a Mac Pro, not in an eGPU box...

Big Navi is probably more eventually important for iMacs than Mac Pros. But right now the Mac Pro is grossly late. If Big Navi means "close to 2020" it is largely the wrong option. Apple is chronically 2-3 tier systems on driver availability and anything coming late in 2019 is highly likely to slide into 2020 if software stack is delayed.

The core issue with "Sooner rather than later" is that AMD's '12nm' stuff would be better for the Mac Pro. Not the 7nm stuff. Long term the next Mac Pro needs better, future GPUs, but the far more immediate problem for the next Mac Pro is that it isn't shipping period. Apple probably needs a custom main display GPU card like right now far more than than they heed a highly custom GPU package at some point late in 2019. They could work with a GPU package that just exists now (late 2018).[/QUOTE]

Late 2019 gives the mid Navi cards a shot. It would help to keep some of the Mac Pro configs in the historical range first-second quadrant pricing to have some GDDR6 cards in the mix. HBM2 isn't going to keep the costs down and production scaling up.[/QUOTE]

All of my GPU speculation was contingent upon when specific GPUs were available...

All of my GPU choices were supposing Apple wants to announce / release a very brand new, VERY late to the party Mac Pro that has cutting edge GPU(s), not something from a year or more ago...

Apple NEEDS to make sure this new Mac Pro is shipping with the very best & the very latest cutting edge hardware & software; otherwise they are right back to selling the Pros old gear in a shiny new wrapper...

To summarize:

Early to Mid 2019 Release of the new modular Mac Pro = Small Navi for budget GPUs & 7nm Vega for Pro-class GPUs

Mid to Late 2019 Release of the new modular Mac Pro = Small Navi for budget GPUs & Big Navi for Pro-class GPUs

2020 will have Next Gen GPUs from AMD & options for the same from the Apple Store for upgrading the modular Mac Pro...

My two cents...
 
So what’s our next potential announcement, reveal, sneak peak date? Anything realistic before WWDC?

Apple doesn't need a fixed in stone date. When they are "done" ( or fixed the design but prepping to ramp production) they can simply gather folks in their multimillion dollar theater and do a small 'show'. Attaching the Mac Pro to one of the two "fixed in stone" large dog and pony shows that Apple does is next to useless in their grossly late context. There is zero advantage in having finished the next Mac Pro and then sitting on it to maximize the dog and pony show. The simple fact that they are specularly late means it will get coverage in the media markets/venues that matter.

Apple should have something more substantive to say by the April anniversary date of their established "dog ate my homework" Mac Pro information updates. That is very realistic. If only to more firmly ground the depth of their SNAFU situation. Those small scale media/blogger gatherings got word out about the Mac Pro just fine the last two years. Zero reason to think it couldn't do the same in 2019. There is zero requirement for a large production dog and pony show.

( I'm sure some folks will want to tab that as purely NAB driven. That's probably a contributing factor, but not the primary one. NAB 2019 is April 6-11. )


If grossly fixated on a multiple device event, the next is probably in March-April .

1. iPad ( the mainstream normal more edu focused) version has updated in last two Marches .
There is a substantive uptick in iPad mini rumors. One or both will be due by then.

2. The iMac, which is more strategic, is strangely lacking an update also. The mini has 6 cores and
the iMac is stuck on 4 cores. There are few "good" reasons for that other than lack of attention by
Apple. They may be waiting on a GPU bump to go along with it so it could be sliding into 2019.
(MBP updating with a Vega option a couple of months after "refresh" upgrades is indicative that the
GPU upgrade availability is off by a several months. )

Apple could toss a Mac Pro ( and iMac Pro ) update into that pile if they wanted. Not necessary, but they could.

The next bump to Intel W should be out by April/May and were probably roadmapped ~2 years ago as being "late 2018" products. ( Intel's 14nm production logjam should ease by April ). Similar ( old roadmap late 2018 ...reality 2019 ) issues for some AMD options.

It would have been sensible for Apple to pick some "late 2018" parts back in 2017. If those slide into 2019 then the Mac Pro would slide into 2019. April should be sufficient to catch up with the 'slide'.


Perhaps some new info will surface at CES 2019 next month but WWDC isn't particularly good for anything Apple might be considering as far as components. By WWDC the current Mac Pro would have hit the 2000 days since last update mark (only 5.5 months to go). [ less than 3.5 month to 2,500 mark if measure from the 2012 intro. ] Apple will be clowned long before WWDC starts. There won't be any arm flapping they'll be able to do at that point to avoid those rounds of ridicule.

If Apple bet the farm on some 2nd half 2019 parts then WWDC is bit too early for a ship/release date. There is zero upside in waiting until WWDC to just "talk don't touch" a new Mac Pro.
 
I feel like since they had Panzarino back in 2018 to give a pro update and make clear the Mac Pro was a 2019 product it seems more likely than not that would be the timeframe for release if the chips are there, or at the very least we'd get another update.

I do agree that it's more likely as shown by recent activities that they'll do a NY campus/spaceship campus invitation to journos rather than waiting until Dub-Dub.
 
....

I do agree that it's more likely as shown by recent activities that they'll do a NY campus/spaceship campus invitation to journos rather than waiting until Dub-Dub.

If ready by March they could invade SXSW in Austin (or just before/after ) as that is another billion $ campus location at this point too (and could squeeze in some 'in person' meetings with some substantive internal groups also) . SXSW is a entertainment+tech mash up kind of event that probably play well with a portion of the Mac Pro target audience. They will definitely be tech porn bloggers and media folks there in spades.

If only for the Mac Pro though, they don't really need a splashy venue. The only thing Apple may have to do with these other events and venues it get the timing right so most of the folks they want to invite aren't off heading to location B when they want them in location A.
 
I'm gonna say sneak peek at WWDC with an order date of 12/31/19 and delivery in mid 2020

If Apple's real shipment dates are in 2020 then they should basically say so in April 2019 ... not June. In April 2018, they announced that Mac Pro would not be in 2018 purchasing calendar year so that people could plan accordingly. If they know in April they are extremely likely to miss 2019 then they should explicitly say so for exactly the same reasons they already established a baseline to admit to.

If the situation is extremely SNAFU they should just confess earlier rather than later. And confessing to a SNAFU at WWDC is dubious. That's where should be talking about accomplishments; not screw ups. Getting all the grumbling about screw ups out of the way in April-May so that can get back to putting the message back onto accomplishments in June would be far better. Kicking that can into June will only make it worse.

Apple isn't going to shot for dates that late in December because it would be hard for many customers realize them a completed transaction for their financial reporting. The MP 2013 Dec 20-21 date (technical end of Fall in Northern Hemisphere is a more pragmatic "drop dead" deadline. )

With the Mac Pro 2013 WWDC and iMac Pro 2017 WWDC sneak peak it was about just as much an announcement of a brand new form factor for that level of performance. In 2013 case "the King is dead, long live the King". For the iMac Pro is was also a big of a change for many targeted. " King literal desktop MP 2103 is mostly dead , long live the new literal desktop king, iMac Pro". If what Apple is going to be doing here is a large chunk of backsliding to some of the baselines of the 2009-2012 era Mac Pro isn't a more innovation update. It is more of a SNAFU that are fixing ... which is more so a screw up. Like said above, WWDC is not where Apple is likely to do that.

[ edit fixed April 2018 date from 'typo' June 2018 for last Mac Pro 'update' session. ]
 
Last edited:
  • Like
Reactions: fuchsdh
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.