I found a strange post on the Nvidia forum, it seems like someone named "Venturi" found a way to make 4 way SLI scale well, using dual CPUs. If AMD could encourage dual configurations for Rizen and make motherboard manufacturers make mainstream AMD dual socket systems, I believe we can beat the green team for good. I pasted the original post below, please read.
I know that you don't support 3 or 4 way SLI any more, but you are aware that it works more efficiently now than ever before, a user at Guru3d named Venturi discovered this and taught others (he taught Youtuber ThirtyIR). Have you ever seen ThirtyIR's videos (https://www.youtube.com/user/ThirtyIR/videos)?
"Venturi" discovered that you allow games to be ran in custom SLI profiles enabling 4-way SLI and he discovered ways which show that SLI scales more effectively across 4 GPUs now than ever before.
Please check out his thread on Guru3d (https://forums.guru3d.com/threads/functional-4-way-sli-pascal-titan-x.409468/).
Also Venturi found that dual CPU socket motherboards allow 4 way SLI to function more effectively, when he divided each GPU to a corresponding CPU effectively in the BIOS. It's like having twin 2-way SLI systems work together.
Nvidia, you are aware that 2 GPUs work great with one CPU, it is clear now that 4 GPU's need 2 CPUs (please put pressure on the industry to make mainstream dual socket systems). The time of 1 CPU for an enthusiast rig has come to an end.
Also Nvidia, you could design cards in SLI from the beginning and sell them as SLI unit blocks. With each block interconnected and using the fastest method of data transmission between GPUs possible with current technology (with dual CPUs required for 4-way SLI). This could be done even if you have to disgard current SLI and start something new, like Ultimate SLI or something (a completely new kind of technology with an insanely fast bandwidth).
Everyone who wants to buy GPUs separately could still do that but suffer a slight performance decrease (compared to the SLI blocks), or just save up money to buy the SLI block which they want (like a Duel, Tri or Quad SLI block). These blocks need to be open to Third party enhancement (with aftermarket coolers, etc.) and you should still sell the cards individually with an improved kind of SLI.
Each SLI block would be spaced to fit in any motherboard with enough available PCI-E slots, and register as one GPU in the system with each (of 2 to 4 components) manageable like a single card (just fully synchronized). And dual CPU motherboards should be able to manage all GPUs in the most efficient way you find (even if you need to run it as two separate local systems, linked perfectly in all operations).
So if developers want to manage resources between each card they can, though all VRAM would be shared across cards in a RAID0-like configuration (so that we are capable of higher resolutions).
Nvidia, you're the best GPU company in the world and can have a huge impact on the future. I ask that you consider these options.
If AMD GPUs remain affordable and stay in pace with Nvidia the hyper-connected "Crossfire block" idea with RAID0 between cards and dual CPUs could destroy the green team for good. When do any of you think this will be possible?
I see a future where Nvidia is defeated, where more than 2 CPU sockets are on a normal motherboard, photo-realistic graphics have been achieved and multiple high-performance GPUs are part of a normal PC.