When Quad-Scaling Isn't Enough

[koolkat574]

So, all things being equal, could one go beyond quad-scaling solutions? I mean like four Gtx 590s; that's 8 effective gpus. I realize that there aren't really any motherboards or drivers that are capable of this. However, if there were, would this be even slightly feasible? I can't think of anything that would require that much graphical computing power, except scientific calculations and modeling...but then one would probably pick a supercomputing solution- if one were a scientist and needed that much computing power 😱

Yeahhhhhhhh

 

[starravier]

Well for starters. no its not possible to have 4 590's. Hence quad sli.
Two. You couldnt find a power outlet to pump out enough juice to power that
Three. There is only one Sli slot for a reason

 

[koolkat574]

I do realize all those things lol

I just had the thought and I decided to post it.

 

[trogdor796]

Do you have a specific case or real life experience where quad scaling isn't enough? Or is this just a theoretical question? Because I don't think there's anything out there where 2x590's won't be enough...

 

[bystander]

Do you have a specific case or real life experience where quad scaling isn't enough? Or is this just a theoretical question? Because I don't think there's anything out there where 2x590's won't be enough...

I've seen 2D surround/eyefinity tests that show that it's not always enough, but still gives pretty good results, just not maxed out results. Now imagine what 3D surround would require (2 images per screen, for 6 images to be rendered per frame).

 

[megamanx00]

Not really. For one if you're doing alternate frame rendering you're talking about rendering more than 4 frames in advance. Even if you did a combination of AF and scissor or tile rendering (like having 8 GPUs rendering 4 frames with 2GPUs to a frame) the driver overhead would be huge and thus you would be waiting on the CPU.

 

[megamanx00]

If you're talking about CUDA though I suppose you could have two GTX590s and some Tesla cards in some kind of funky server case. Tesla cards are essentially Quadro cards without video outputs. Assuming the drivers would allow it you could use all those CUDA GPUs for some kind of calculation.

 

[koolkat574]

No, I can't think of anything that would require more than 2 590s or 6990s, other than what bystander said about 3d surround on several monitors...or things that use CUDA. Twas a hypothetical.