Thanks for the really informative replies in here guys. I THINK I’m going with a duo for longevity, but still debating.
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.
MP 7,1 Two Vegas vs Vega Duo
- Thread starter Shadow Puppets
- Start date
- Sort by reaction score
So for clarification...for things like gaming, is the duo really going to “double” the performance of the single?
Short answer is that it really depends on how the software is written.
Slightly longer answer starts with the short answer and adds:
"But most games in particular would probably have close to linear scaling, so almost double in a lot of cases, yes - at least with Bootcamp"
And the proper long answer would go something like this:
For gaming workloads, the past few years have seen a trend going away from multi-GPU set-ups, so there's less priority on it. With APIs like DX12 and Vulkan (and Metal), the developer has more control of spreading the workload across GPUs, which means they can optimise more for multi-GPU, but it also means that if they don't bother scaling doesn't come as freely. With traditional Cross-fire/SLI solutions, the operations where often just split by the driver between the cards, which only really worked properly with a sufficiently large chunk of VRAM on both dies. (For the Vegas that's no concern). So for anything that is written to utilise multi-GPU, you should see pretty good performance scaling, though it'll never be 100% linear, and the exact scaling will vary between every operation/game.
thanks for the detailed response. Out of curiosity, reading up on this subject, a lot of people are saying that if the game doesn’t support crosslink (I believe that’s what they said from memory) it won’t make use of the duo?
CrossFire is AMD's term. And that is... Sort of true. The way games typically are written for multi-GPU is with CrossFire, but there are technically other ways of writing code to scale across multiple GPUs, though I haven't really seen it much in games. Think just once actually. But with DX12/Vulkan/Metal there are new ways of scaling across multiple GPUs as well that also are not CrossFire, but again, the developer has to write it like that. For games, CrossFire/SLI is the most common solution for multi-GPU scaling, to the point you might call it the only (almost)
The game has to support multiple GPUs on macOS. I don’t think I know of any that do.
Exactly same situation as the 2013 Mac Pro. Even with the Duo card.
Edit: Realized we could be talking Windows too. Under Windows if the game supports multiple GPUs (either through Crossfire or DX12 GPU pools) it should work. Again, same thing as the 2013.
The only real game with support for Explicit Multi-GPU is Ashes of Singularity.
Curious to know if anyone here has any experience with this? (no answer from Apple).
I'm considering adding a second MPX Vega II solo to my existing config (one other Vega II solo). The second module comes with an ' Infinity Fabric Link bridge', however the Apple support site does not add any specific documentation beyond:
Connect the Infinity Fabric Link
If you install two Radeon Pro Vega II MPX Modules, you can connect them using Infinity Fabric Link for increased performance and faster data transfer between the GPUs:
As far as I can see, my existing MPX module is sealed with only a release catch visible. Perhaps the MPX module is meant to be removed, then the whole shroud taken off, who knows ....
I'm considering adding a second MPX Vega II solo to my existing config (one other Vega II solo). The second module comes with an ' Infinity Fabric Link bridge', however the Apple support site does not add any specific documentation beyond:
Connect the Infinity Fabric Link
If you install two Radeon Pro Vega II MPX Modules, you can connect them using Infinity Fabric Link for increased performance and faster data transfer between the GPUs:
- Remove the caps from each MPX module.
- Align the connectors on the Infinity Fabric Link bridge with the connectors on the MPX modules, then slowly press it in.
- Use a Phillips Head screwdriver to turn both screws into the locked position.
As far as I can see, my existing MPX module is sealed with only a release catch visible. Perhaps the MPX module is meant to be removed, then the whole shroud taken off, who knows ....
This image from the product page illustrates where the connectors are.
The connectors at the top. There is a usually something covering those ( in the top view image you don't look "down" on those. So I suspect the view above has the covering removed because there is a 'break' in the top round edge around the connector. ). That covering is probably what Apple is referring to here as the 'cap'. Apple has probably worked pretty hard to camouflage the seem where the cap fits in . The non 'blue' end of the lever, but more on the rounded edge. (if zoomed in close on the top view image there are some artifacts there to suggest a seam. )
The caps refer to the black caps covering the infinity on the GPU itself. They are capped from factory. So you would need to remove gpu and take caps off. It is possible to remove them while installed in Mac but its easier to take GPU out of system to remove them.
Edit: Snapped a picture of my AMD Vega II
Last edited:
2nd Vega II on the way, thanks to this forum for advice. The Infinity Fabric link will have no benefit to my NLEs, but will hook it up nonetheless.
Shall be interesting to see how much this might boost times for rendering, noise reduction, plugins in DaVinci Resolve Studio (either in Catalina 10.15.6 or in BootCamp Win10 for Workstations 2004). Or not. My gut feeling is this will be only moderate at best, we'll see. Ditto for FCPX, the new 10.4.9 might have some more grunt with this, depending on codec & workflow. Otherwise, 2 weeks to return if underwhelming for the money.
Shall be interesting to see how much this might boost times for rendering, noise reduction, plugins in DaVinci Resolve Studio (either in Catalina 10.15.6 or in BootCamp Win10 for Workstations 2004). Or not. My gut feeling is this will be only moderate at best, we'll see. Ditto for FCPX, the new 10.4.9 might have some more grunt with this, depending on codec & workflow. Otherwise, 2 weeks to return if underwhelming for the money.
How is the setup so far? I am also on the fence between getting another solo Vega II or just replacing with a duo version.
Runs very well - better than I expected & was prepared to return the second MPX. I don't use for gaming at all, the main reference point was how well it boosted performance in DaVinci Resolve Studio caching, rendering & plug-in use (& FCPX to some degree). Generally using 4k H264 or all-Intra on a 1080p timeline. Tried all that & monitored on Sensi & iStats - yes, at times both GPUs can be running up to 80%, 100% & bounce around a little. One GPU has zero I/O attached (like Resolve likes it), the other for all displays & some thunderbolt. Neat Video plug for demanding noise reduction - I let it test & optimise its settings & it came up with a combo of aprox. half the CPU cores and a half dozen cores each from the two Vegas - again, improved the NR performance by a useful margin.
So I kept the GPU (& surprised myself a little). The biggest downside: two MPX modules means there are no slots left in this Mac Pro (also have installed Sonnet 4x4, UAD-2, BM mini-mon 4k). & so optioning for a Vega Duo in the first place would have left those extra slots & x16 available for other possibilities in the future. Still, the duo cost way too much up up front (& especially in comparison to an RTX 2080Ti I have in a Win workstation). This way I guess the advantage is that one can spread the cost over time.
