AMD announces Radeon Pro Vega II and Pro Vega II Duo GPUs

[HEXUS]

These 7nm Vega GPU cards utilise high-speed HBM2 memory and AMD Infinity Fabric Link.

Read more.

[Xlucine]

So another outing for the V7 chip? I like the look of the extra power contacts on the PCIe slot, shame it's unsuitable for consumer cases

[Tabbykatze]

So another outing for the V7 chip? I like the look of the extra power contacts on the PCIe slot, shame it's unsuitable for consumer cases

No, this is the full fat MI60 V20 chip whereas the V7 was a cut down MI50 chip.

Basically, this Radeon Vega 2 is the top end Vega for consumers.

What I'm verrrry interested in is the duo, this could be a sign of a return of AMD dual GPU on a single card designsbif the IF does it's job well...

[LSG501]

Yay for proprietary connectors....

DUO is essentially a return to the dual cards of old, nothing particularly wrong with that approach if you can get enough bandwidth, which in the case of the mac pro seems to require an extra socket...

I wonder if AMD will bring out something similar on pcie4, that should have enough bandwidth I would think. Although as much as I want to support AMD and opencl on the gpu side of things, Nvidia's cuda is just so much more practical for my usage.

[darcotech]

Well it is Apple commanded part. They probably gave a lot off money to AMD so they just did what they asked for.
So Apple MacPro has world's fastest graphic card?!!

[Luke7]

AMD is getting some huge wins through contracts and Ryzen a the moment... I'm not foolish enough to say "Intel is dead" or in trouble - I'm just here hoping AMD can invest in themselves and R&D to REMAIN competitive long-term.

[SylvanSagacious]

Considering crossfire support has all but withered away to nothing, I wouldn't be too eager to see a return of dual GPU cards.

Unless you're thinking of using it for home heating (like I did in uni with my old 4870x2)

[3dcandy]

As they use Infinity Fabric there should be no real issues for speed.

As Apple sales are so low it can remain a niche product

*not serious about above statement*

[Xlucine]

Ooh, it is infinity fabric. Doesn't hold a candle to the HBM2 bandwidth (84 GB/s compared to 1 TB/s)

Yay for proprietary connectors....

DUO is essentially a return to the dual cards of old, nothing particularly wrong with that approach if you can get enough bandwidth, which in the case of the mac pro seems to require an extra socket...

I wonder if AMD will bring out something similar on pcie4, that should have enough bandwidth I would think. Although as much as I want to support AMD and opencl on the gpu side of things, Nvidia's cuda is just so much more practical for my usage.

I believe that's for extra power, not bandwidth

[LSG501]

Ooh, it is infinity fabric. Doesn't hold a candle to the HBM2 bandwidth (84 GB/s compared to 1 TB/s)



I believe that's for extra power, not bandwidth

taken from the apple mac pro page:

A second connector. An industry first.

The MPX Module starts with an industry-standard PCI Express connector. Then, for the first time in a graphics card, additional PCIe lanes were created to integrate Thunderbolt and provide increased capability. With up to 500 watts, the MPX Module has power capacity equivalent to that of the entire previous-generation Mac Pro.

IMO that reads as extra bandwidth for the gpu and thunderbolt ports... plus extra power instead of a power cable like 'normal'.. but seeing as it's proprietary it could literally just be there to physically support the card lol

As to infinity fabric, while it might not be up to HMB2, it's faster than pcie to pcie, supposedly 5x faster... not sure how that compares with nvidia's new rtx nvlink bridge though, that can do 100 GB/s according to nvidia.

[Xlucine]

I was going off this hexus story, that includes an image from AMD calling the first connector "standard PCI Express 16x" and only referring to extra power from the second one:


So there might be more lanes through the second connector? I'm kind of curious why the second connector looks different to the first (less contact area), and the things that look like connectors on the other side of the card are also a mystery.

[acekorneya]

So another outing for the V7 chip? I like the look of the extra power contacts on the PCIe slot, shame it's unsuitable for consumer cases

No, this is the full fat MI60 V20 chip whereas the V7 was a cut down MI50 chip.

Basically, this Radeon Vega 2 is the top end Vega for consumers.

What I'm verrrry interested in is the duo, this could be a sign of a return of AMD dual GPU on a single card designsbif the IF does it's job well...

If im correct the Pro Vega II Duo will only be available on the new Mac Pro

[acekorneya]

I was going off this hexus story, that includes an image from AMD calling the first connector "standard PCI Express 16x" and only referring to extra power from the second one:


So there might be more lanes through the second connector? I'm kind of curious why the second connector looks different to the first (less contact area), and the things that look like connectors on the other side of the card are also a mystery.

Well that is because this connector is a Mac Proprietary Connector so this card won't be sold anywhere else.. only on the new Mac Pro tower..

[aidanjt]

I was going off this hexus story, that includes an image from AMD calling the first connector "standard PCI Express 16x" and only referring to extra power from the second one:


So there might be more lanes through the second connector? I'm kind of curious why the second connector looks different to the first (less contact area), and the things that look like connectors on the other side of the card are also a mystery.

As best as I can tell, it'll be the same amount of PCIe lanes, just added connector pins for video out, and PCIe power, only without the PCIe power cables.

[LSG501]

I was going off this hexus story, that includes an image from AMD calling the first connector "standard PCI Express 16x" and only referring to extra power from the second one:


So there might be more lanes through the second connector? I'm kind of curious why the second connector looks different to the first (less contact area), and the things that look like connectors on the other side of the card are also a mystery.

That image is from the mac pro webpage. That second connector is likely being used to supply data for the thunderbolt ports, a single pcie wouldn't really be enough for 40Gb/s data let alone 2x gpu's while giving them enough individual capacity. It might also be being used for additional bandwidth for the second gpu, without a hands on it's a bit hard to really tell in all honesty, it's more speculation from what we can see.

As to the second connectors (on bottom of your image), it seems like they're some sort of interconnect between the cards because you can see a 'bridging' card in some of the other images. So maybe it's an infinity fabric connector for when it has as second set of the duo gpu's.

[DanceswithUnix]

Apple's page says the second connector has PCIe lanes on it, but that square chip next to the standard x16 connector looks rather like a PCIe bridge to me which I presume will feed both GPUs off the main slot. So yeah, it does seem like the extra lanes are for wiring up Thunderbolt.

Man that board is so damned proprietary, I wonder how much you would lose vs a Threadripper board and 4 Vega 7 cards (which with "only" 16GB of VRAM each might slow some rendering).