A large chunk of the VEGA GPU is a package. The GPU chip and the HBM memory are all on a single interposer ( essentially another chip die). By in large a substantive chunk of the GPU subsystem comes in just one "chunk".
That Apple makes their own embedded GPU .... wel err so does about every other laptop vendor out there with a discrete GPU models. Putting a GPU on the main logicboard isn't rocket science. Apple isn't buying pre-build logic boards from Intel either and it isn't a huge hoo-hah for them to put the CPU into a socket on a board they designed. It isn't like AMD isn't also handing them reference board specs. And reference boards if they want them.
errr.... a quick search at www.newegg.com for 'Vega 56' brings up several boards.
As some on else pointed out the first Vega shipped, the Frontier edition, is a Vega 64 16GB. So yeah you can put on in a generic PC.
The PCI-e standard slot card form factor is a minor 'form" issue not a function issue.
Apple has "Pro" prefix on them. That's more driver different than hardware difference. Regardless they need macOS drivers instead of the Windows drivers the generic market cards come bundled with.
No, not for Vega II. As I noted before, the Vega II models only have Linux drivers. The WX variants actually come with Windows drivers (and Linux) .
Actually. "handy" would be something that is shipping. The next Mac Pro is grossly late. Handy would be something Apple could ship. Off chasing the pot o' gold at the end of the next gen GPU rainbow isn't "handy".
For the vast majority of audio DAW workload there will next to now material different between what a Polaris, Vega, or Navi does. For the most part, it just isn't on the primary critical path(s). So again, not particularly "handy" on that dimension either.
if the rest of the components were available should Apple be waiting around for Navi to finish up and reasonably viable/mature macOS drivers to also finish up? If AMD told them in 2017 that the timeline was mid-late 2019 ... Not really, that would be relatively bad product management for this product in this particular context ( being grossly late).
I think the rumors about the RX 3080 being around $249 are in the "too good to be true" zone. I doubt it is that though and by the time Apple/AMD pushes a $200+ mark up on that ( for 'Pro' badging and Apple market up) that isn't really going to be an suitable entry card for the Mac Pro unless trying to inflate the overall system price.... which is a dubious move.
What would make the most sense is to pick which Navi that will get used in perhaps the iMac (or MBP) so there was some more volume and the price doesn't creep up too high.
Big Navi is even worse in terms of timeliness. Some reports put it in 2020.... which is ridiculous for a Mac Pro getting to market in a timely fashion.
The issue really isn't "Big Navi". The primary issue should be that Apple develops ( and possibly promotes) a long term longitudinal defacto standard socket to put their main dispaly GPUs into. Someone with a 2019 Mac Pro could pick up a future card. And that Apple doesn't disappear down a rabbit hole for 3-4 years on new upgrade cards.
Errr, AMD has talked a bit about what/where they are heading. In the datacenter/compute space they are thinking of multiple chip packages. In the mainstream/gamer space they aren't. AMD has hand two tracks Polaris and Vega for a while. There was lots of hand waving that Vega was going to be a "top to bottom" solution for "everything in AMD stack. That didn't pan out. Nor do these recent comments pan out for one baseline microarchtiecdture to do everything.
AMD isn't going to be able to remain compelive in the upper space if they don't slightly fork off the HPC version from the more graphics focused mainstream one.
Vega II pumped up the number of external Infinity Fabric links and also the number of HBM links. A pretty straightforward way of doing that with a reasonable transistor budget would be trading off space for 2-6 displayport output subsystem for additional external I/O. Since not targeting folks with with multiple monitors just dump that stuff.
Vega II expanded in two direction on instruction set focus. Down int INT4 for super coarse gain MI stuff and up into double floats. I don't think Metal is particularly geared to either one of those two. Nor do I think Apple's primary display graphics stack particularly interested in them either. So that is an apparent large mismatch.
The other thing that AMD talked about with Vega II is mapping of GPU to Virtual machines ( on GPU to many VMs or one VM to many GPU coupled by Inf Fabric. ). Again missing the DisplayPort/HDMI output. The Raster aspects may still be there but it doesn't seem like physical output is a focus point.
I think trying to vaccum with what want to hear because Vega II is shipping so doesn't look as bad as "Big Navi" 2020 timeline. As I said before it would be a huge tactical shift for Apple even to get engaged with making Vega II avaialble just as compute card ( its primary purpose. ). Even that is basically blank right now. Let alone being a primary display card ( which is grossly not its primary focus point. One single physical output port ... that is not an single end user's display driving focused card at all. )
Mac Pro is highly unlikely to have PCI-e v4 so not going to completely saturate to theoretical max it out anyway.
The problem with chasing the "super uber ultimate" cutting edge is that just keep pushing back the possible ship dates. Apple put on that "can't innovate may a**" bluster at the Mac Pro 2013 intro. That was alot of hooey. Innovation involves actually doing something. They crawled into a rabbit hole and went to sleep for 5-6 years. That's not innovative in the slightest. It would be helpful if they put any the dog and pony bluster hocus pocus and actually just deliver on a regular , timely basis. Utlra , uber , mega GPU/CPUs aren't going to dig them out the hole they are in at this point. Putting in consistent real work might; both on their own Mac Pro product and with reasonable number of partners and suppliers. ( they keep digging a deeper hole so perhaps not.)
The current hype on "small" Navi is that it is in spitting distance of Vega.... so not sure what 7nm Vega is going to do since they moved the 7hm Vega design to focus more so on datacenter stuff. That change in focus is not a mystery it is what they have openly stated for that implementation.
That there is another Vega design that is sucking up resources from the "Big Navi" .... I would be very surprised. AMD isn't bleeding cash anymore but they also don't have huge numbers of parallel teams either. AMD is "fighting" a two front war. Intel and Nvidia. Neither one of those fronts is going to get all of the resources.
AMD taking the low-upper mid range with "small Navi" and grabbing a respectable chunk of the datacenter MI/HPC growth would be a huge win for them. That upper market super gamer market ... that isn't particularly necessary over next year or so [ Nor is it particularly necessary for Apple either for that same time period. ]
Apple just needs three "good enough" cards at different level at not too painful price points and an open x16 slot with reasonable high range of power 300w (maybe 350W ). Folks you have to have to super duper can and put it in. Apple could may sell some of those but they aren't probably critical for Apple to sell directly. They need to help faclitate that there are reasonable driver for them that are highly compatible with the rest of the system (and default display card). But Apple doesn't have to supply every possible card. That isn't necessary at all. In fact, a large group of folks don't want that ( the Nvidia fanclub for example) .
Not clear at this point that "Big Navi" isn't in 2020.
What AMD has be slow and lacking is on the smaller and mobile scale. Few indications they are going to unwind all of that problem before the end of 2019. The fanboy wars between AMD and Nvidia would like to enagage in smack talking in that space but there are other places where AMD is in worse position but have better fixes possible sooner rather than later.
Longer term I don't think Apple would have anything in the Apple Store that wasn't also concurrently part of standard configurations. So if there are current good , better, range of cards on the standard configuration then the cards from those would be in the Apple Store. When Apple shifts to another set of 2-3 cards then those are what are in the store.
There wouldn't be everything for everybody set of cards. Apple would just sell incrementally more of what they were already selling.
If Apple makes the card so it fits in as a compute card in a standard slot they could incrementally sell a few more that way also.
This would necessary have to be via the Apple store. It could be through the authorized service network to. Folks order and get it shipped via same logistics Apple uses to more around parts for repair. Stocked on the shelves in small inventories is probably the wrong model for these cards. They aren't going to fly off the shelves in "high volume" at any time. They would be an option for some people.
