Why doesn't your VRAM increase in crossfire/SLI?

[KamiQueso]

Sorry if this is a noob question, but after watching Linus' videos for quite some time I've learned that your VRAM stays the same as one videocard in Crossfire or SLI. Why does this occur? If you run two 3gig cards in SLI,  how come you  don't have 6gigs of VRAM and what happens to the 3 other gigs of memory?

 

Again, sorry if this is a stupid question but I've wondered about this for a long time.

[M-ursu]

Both GPU's need to have theyr own memory, they cant share it.

[mineblaster]

because they are like painters all working on the same picture the but when you add video cards its like hire more workers but the picture dose not get bigger

[killcomic]

What the above post said. Both cards are being used to render the same scene. So for example, the exact same textures are loaded in duplicate on both cards, so 2x3Gb cards will still be holding 3Gb worth of info because it is duplicated.

[Thunderjolt]

Because for the GPU to do work it needs to have the files loaded onto it. Since they can't share resources they both need the same files loaded onto them.

[Panda of the Shadow]

They don't fuse they are still two cards. They work together but they aren't the same card. One does one frame one progresses another.

[Giggitygoebbels]

The VRAM acts like "RAID 1" when in sli/cfx,not "RAID 0".

[lufihengr]

They process the same amount of data as one card would, but faster and thus more FPS.

[KamiQueso]

Thanks everyone!

[R4STABAN]

The communication overhead involved in moving data between the vram of multiple cards (among other things) would result in less performance than a single card.

[kevinheinrich]

If a programmer finds out how to develop a 3rd party utility/driver to multiply your vram in sli/cfire  Over the course of 24hrs he/she will become richer than the person who invented angry birds.

[Panda of the Shadow]

If a programmer finds out how to develop a 3rd party utility/driver to multiply your vram in sli/cfire  Over the course of 24hrs he/she will become richer than the person who invented angry birds.

Haha no. 

[Aniallation]

It would be a lot of work for one GPU to have to pull info from the VRAM of the other, which would require processing through the slots, PCIe bridge, controller, other card, etc. so in the end it's just faster for the card to deal with its own VRAM if it's possible.

[MightyUnit]

OP the answer to your question is an understanding of AFR (Alternate Frame Rendering). Each GPU in a SLI or Xire setup takes turns rendering frames. ie: GPU 0 renders frame 1, meanwhile GPU 1 renders frame 2, GPU 0 renders frame 3 and so one. Each frame in stored in the vram of the card that rendered it. So you can see that both cards are not working together on each frame and are NOT sharing vram. Does that make sense??