SLI and Crossfire question?

X

Xenozx

2[H]4U
Joined
Jun 16, 2004
Messages
2,127
OK heres my question. I dont see why you "HAVE" to have 2 of the same cards for SLI and crossfire....

Why cant you use 1 7900 GTX, and your second card being a 7900 GT? Why exactly do they have to run at the same speed?


Also I have heard that even if you have 2 512 MB cards, thats with SLI/Crossfire you still end up with only a total of 512 VRAM. Well why cant they create a second card lets say the same card you have but with 32 mb of memory (will save quite a bit of cost) and that way its still processing and doing everything its needs to do and costs so much less? I think it would make SLI more available to the main stream based on pricing? Basically only the first card is being used memory wise so why waste the extra power on the second card?
 
You can use two different manufactuer's cards in SLI, no problem there, EDIT: as long as they are the same GPU die.

You only have 512MB total with 2 cards because each card is still rendering the scene (just not the whole scene), therefore it still sources all the textures, shaders, and geometry that it would need to render the whole scene. You don't get any more memory because all the scene data needs to be duplicated on both cards. The combination of output from the two cards only happens at scanout, there's not enough memory bandwidth for one card to look up textures on a different card.
 
eXzite said:
You can use two different cards in SLI, no problem there, the cards will load balance.

You only have 512MB total with 2 cards because each card is still rendering the scene (just not the whole scene), therefore it still sources all the textures, shaders, and geometry that it would need to render the whole scene. You don't get any more memory because all the scene data needs to be duplicated on both cards. The combination of output from the two cards only happens at scanout, there's not enough memory bandwidth for one card to look up textures on a different card.
Click to expand...

I C, thank you for the knowlegable explaination.

So I can use any 2 sli capable cards? If I wanted to run my 7900 GTX and a 6800 GS I can? I dont supose anyone has benchmarked like this, or is this not worth while?

Would you gain performance from using two different cards? what if one card had 256 mb of memory and the other had 512?
 
Sorry I should have been more clear, you can use to *different* cards as in two different manufacturers, clock speeds, etc. BUT you need to have the same actual GPU. So no, you can't do a 6800 with a 7800. The reason for this is that it's difficult to keep different cards in sync, so to avoid driver nightmares nvidia just forces the gpu's to be the same. Technically, SLI *would* work between, say, a 7800GT and a 7800GTX, however it's disabled in the driver to avoid confusion.

And yes, you can use two cards (same GPU) with different amounts of memory. Perf will take a hit though, unless you do something (i.e lower texture quality) so that the scene fits on the smaller card.
 
I C, so basically if your looking to do SLI it has to be the same GPU, but doesnt matter on manufacturer.

How does overclocking work with SLI? Do you have to overclock both cards separately or do they overclock together. Based on you saying that they can be different frequencies I would asume they overclock separately?

why are there games that dont get an advantage from SLI? I would think having 2 GPU's doing work instead of 1 would pretty much give SLI an advantage everywhere, just like having 2 people working on something instead of 1. Dont supose you know the reasoning behind this?
 
Yes if you overclock, it can be seperate frequencies on each card, they will auto load balance.

SLI scaling is a little complicated, but what it comes down to, are two main factors: 1) Rending algorithm (i.e supertiling, alternate frame render, split screen), and 2) the bottleneck of the game. For example, if a game is texture size/caching bound, then having 2 cards might not help as much as say, if the game was shader processor bound, since the other card will hit the same bottleneck in the texture case, where as having double the number of shaders scales well to increasing pixel throughput. If a game is geometry call bound, as in, bound by the time it takes for the cpu/driver to issue the geometry primitives to the hardware, then SLI will not help at all, since the primitives must be pushed to both cards anyway.
 
I see so if your game is GPU bound SLI will help, if its CPU/Texture memory/space bound then little no no benefit.

Thanks for the info.
 
You must log in or register to reply here.
Back
Top