News AMD points to multi-GPU advances in Mantle, DX12

This is marketing fluff. You can't share memory between the gpu's, so each card still needs all the resources in memory which for any game is basically the same stuff (textures, etc). Just cause you can theoretically put some stuff in only one gpu's memory and not the other doesn't make it useful most of the time.

Even if you could find some use for this and lets say it gave a 5% performance boost which games company is going to bother coding and maintaining this? Exactly what % of their customer base would benefit - the numbers using mantle enabled cards in Xfire is a tiny % of the number that could buy the game. Then you've got to think of that tiny group how many wouldn't have bothered buying the game but now will buy it cause it runs 5% faster on their Xfire setup? The extra money they'd generate by doing this is insignificant and certainly doesn't come close to the costs of implementing and maintaining it.

 

This reads to me like a return to 3dfx's original definition of SLI, where GPU A draws all the odd-numbered lines, and GPU B draws the even ones.

 

Indeed as I thought too. This halves the amount of pixels that are required per frame in essence halving the vram requirement and giving the same effect as doubling the vram available per pixel rendered.

Will be interesting to see the end result I think

 

Carmack has talked about doing this for Oculus Rift. From what he said its brilliant if your displays refresh rate is high enough.

 

Its funny because I missed that entire last bit off

 

Also had major flaws if the game was not coded for it from the ground up. High display refresh rate was also more cheaper and available in the time when 3dfx did it.

These days a 120hz display from a decent manufacture will cost a decent chunk of cash. Or you can pick up the same size screen with 60hz for half the cash in most cases from the same manufacture.

Also issues with 4k as it can bearly drive a 60hz panel let alone a 120hz one. Even 2 980s are not enough for 60fps in most modern games without sacrafices.

 

3Dfx's SLI technology actually duplicated the texture data for every card in the system. Only the frame-buffer content was unique and thus could render at higher resolution without performance drop. Data duplication is the default mode for every unmanaged remote/split-memory system out there.

 

In my defence, I claim lack of caffeine

 

It will be interesting to see what the guys from Valve, Unity, Frostbite and Epic will be highlighting regarding glNext at GDC 20105. And which tech (DX, GL) is going to provide the consumers what they want vs what the devs need.

http://schedule.gdconf.com/session/glnext-the-future-of-high-performance-graphics-presented-by-valve