While it is true that DX12 will have that feature, I do not see that additional memory multi-GPU setups will scale the same way as carbon-copy ones.
I think what it means is that DX12 allows for a "tile" or "slice" multi-GPU rendering scheme (such as, one GPU renders the top half of the image and the other GPU renders the bottom half, or right/left, whatever) where most textures will only be needed on either card, for any given frame. If such is the case, this rendering scheme will cost more VRAM than the carbon-copy version.
So if a carbon-copy version costs 3.5gb, expect that independent VRAM rendering scheme to cost, say, 4gb, which is saying 2gb per card.
Which is still pretty awesome, but we'll have to see how it all works out in terms of bus bandwidth, performance (scaling) and VRAM consumption.
I think what it means is that DX12 allows for a "tile" or "slice" multi-GPU rendering scheme (such as, one GPU renders the top half of the image and the other GPU renders the bottom half, or right/left, whatever) where most textures will only be needed on either card, for any given frame. If such is the case, this rendering scheme will cost more VRAM than the carbon-copy version.
So if a carbon-copy version costs 3.5gb, expect that independent VRAM rendering scheme to cost, say, 4gb, which is saying 2gb per card.
Which is still pretty awesome, but we'll have to see how it all works out in terms of bus bandwidth, performance (scaling) and VRAM consumption.
Facebook
Twitter
Instagram