DirectX 12 to make two GTX 960s more powerful than one 980??

soldier5637

Distinguished
May 7, 2013
729
0
19,160
Hey guys. I have something going on in my head and i wanted to confirm if its right or not. With the new DX12 coming in a few short months, its now known that this new version will allow two GPU's to be treated as one card all together whereas before there was a "master" and "slave" card. Here is my thought: With this being the case, would two DX12 ready GTX 960's not be at least as powerful, if not a little more powerful in some cases, than a GTX 980?? Since theyll now be allowed to be stacked, so example you can have 8 gigs of vRAM now from both of your SLI'd 4 gig cards rather than being limited to still only being able to use four. both buses (128 bit) put together equal the same as a 980 bus, and it will have exactly the same amount of CUDA cores. And actually, like in the case of the EVGA GTX 960 SSC, it'd have higher clock speeds.... and for something like $200 less than a 980.. Tell me if im wrong, but this is really seeming like it will be the case.
 

chenw

Honorable
Unfortunately, just like all other DX related features, games have to be specifically be programmed to take advantage of it.

So by the time we see games rolling out with that feature as a common feature, 960 may already be obsolete, even in SLI.
 

soldier5637

Distinguished
May 7, 2013
729
0
19,160
Well I see where youre coming from as the Tech world is by far the fastest advancing in the world, especially GPU's, But I think its ill-perceived to think a GTX 960 will be obsolete by the time supported games come out in force. The computer gaming industry is a growing one and gamers are more demanding than ever. I think DX12 games will roll out faster than we may think. Just an opinion.
 
Split frame rendering, while having some advantages in smoothness, does not scale nearly as well as alternate frame rendering. Mantle has done this a couple times, and the scaling is abysmal, although the smoothness is good. AFR is almost exclusively used today because it results in much higher FPS and is easier to do.

You may start to see some SFR in the future, but you won't see good scaling from it, and won't likely be as good as getting 1 more powerful GPU.
 
Personally i think most developer will stick with DX11 (up to 11.3). I see many people make assumption that when DX12 arrive all future games will get benefit from it and try to make their decision based on that assumption. Also i haven't much about stack VRAM in DX12. But DX12 will have multi adapter; so you can mix two different gpu together (like IGP assisting discrete gpu). Not sure if that will also allow the mix between nvidia and AMD gpu. But it is worth mentioning it depends on game dev side to implement the feature which rise the question how far this multi adapter feature can go.
 

chenw

Honorable
Actually, come to speak of it, the mixed GPU rendering is definitely the biggest uncertainty in DX12. You can always let AMD and nVidia cards work together, assuming that neither side programs their drivers to stop working if it detects cards from another side.

nVidia already has done it with PhysX, while I don't see them going any further, I would not be surprised if they do that, which completely throws the idea of mixed GPU's out the window.
 


actually nvidia doesn't have to do anything to let the tech die. it is not that mixing nvidia and radeon haven't been tried before. also from the looks of it multi adapter will depends heavily on developer side. imagine if you want 980 to work with 290X, 290, 285, 280X, 280, 270X, 265, 260X, 250X, 250 and 240. and then repeat the process for nvidia 700, 600, 500 and 400 series. that is something you can't leave to developer when not nvidia nor AMD have the willingness to make it work.