Some SLI VRAM confusion.

C

CousinVin

Limp Gawd
Joined
Jan 11, 2009
Messages
393
I think i understand that putting two 3gb cards still only limits you to 3gb of usable vram.. right? If that is wrong please correct me.

Now my confusion comes in with the GTX 590. It is labeled as a 3gb card, but from the assumption above, and considering that it is 1.5gb per core, is it really only 1.5 gb usable vram?
 
Yeah. Multi-GPU does not have an additive effect on VRAM.

I also hate that they market multi-GPU single cards as having 2x the amount of usable RAM. Technically it's correct that the 590 has 3 GB of VRAM total but only 1.5 GB of that is "usable".
 
That is pretty lame... and is chasing me away from the 590. Effectively it's the 1.5 580 in sli with less shaders/ equipment. Ugh, this wait has got me by the balls here. I guess if i can wait like 10 days I'll probably be satisfied.
 
It's actually more like 2x 570s but with more shaders.

But yeah, buying one at this point would probably not be the best idea.
 
CousinVin said:
...Ugh, this wait has got me by the balls here. I guess if i can wait like 10 days I'll probably be satisfied.
Click to expand...
Man, you can say that again. I'm so on edge because of this wait that I probably could have used the time I spent looking up info on Kepler/7xxx to be more productive at work, and maybe I might have actually earned more money than I'm now trying to save by waiting :S
 
Not to threadjack, but I have a question regarding vram in sli/xfire....

I can't seem to find the answer (I have looked).

Assuming 2x 1.5gb gpus are used, the total available vram is 1.5gb.
WHY? Can't nVidia and AMD just enable full usage in the driver? Or is it limited by the OS?
 
CousinVin said:
I think i understand that putting two 3gb cards still only limits you to 3gb of usable vram.. right? If that is wrong please correct me.
Click to expand...

You are correct.
Now my confusion comes in with the GTX 590. It is labeled as a 3gb card, but from the assumption above, and considering that it is 1.5gb per core, is it really only 1.5 gb usable vram?
Click to expand...

It's marketing. Joe Average can't tell the difference between total memory and dedicated memory.
 
Shared video memory and other resources are the holy grail of dual GPU cards, hopefully some advancement has been made in this area with dual kepler or 7990.. even if not shared memory, better communication between GPU's on the same pcb would be great, making it operate more like a single GPU with less CFX/SLI issues (scaling, profiles ect).. until then I will not touch dual GPU cards.
 
There was something about keplar being able to use system ram or something...(so possibly other cards ram?) Probably just was another one of those rumors. Still would be interestinng.
 
E^vol said:
Not to threadjack, but I have a question regarding vram in sli/xfire....

I can't seem to find the answer (I have looked).

Assuming 2x 1.5gb gpus are used, the total available vram is 1.5gb.
WHY? Can't nVidia and AMD just enable full usage in the driver? Or is it limited by the OS?
Click to expand...

because thats not the way SLI/CFX are designed. it duplicates the frame buffer across both cards. (nevermind tsumi explained it better then i did so i removed part of my explination)

demowhc said:
Shared video memory and other resources are the holy grail of dual GPU cards, hopefully some advancement has been made in this area with dual kepler or 7990.. even if not shared memory, better communication between GPU's on the same pcb would be great, making it operate more like a single GPU with less CFX/SLI issues (scaling, profiles ect).. until then I will not touch dual GPU cards.
Click to expand...

AMD tried the side port memory concept but gave up on it pretty quickly just because theres to much required to make it realistic and because AMD/Intel no longer use external north bridges it has to be directly implemented into the cpu which takes up to much space (AMD talked about this and why they didn't include memory sideport support in phenom II a few years back).
 
DeathPrincess said:
There was something about keplar being able to use system ram or something...(so possibly other cards ram?) Probably just was another one of those rumors. Still would be interestinng.
Click to expand...

I doubt it can use another card's vram if that card is in SLI with it. Here's why:

SLI and crossfire both work by alternating frames between the GPUs. Therefore, each frame has to be buffered in the memory for each card. So each core is already using its vRAM.

The vRAM is not being wasted, as some are making it out to be in this thread. It is being utilized, but because each GPU needs to use its own dedicated RAM, it cannot share the RAM with the other card.
 
...and they won't ever attempt to share the VRAM between multiple cards because the resulting traffic on the PCI-E bus would be extreme.
 
I wasn't thinking so much for games. For lots of CUDA based stuff the vram is used heavily, but the GPUs lightly. Being able to have a single GPU with an effective 12GBs of vRAM would have lots of possibilities.

Then again, if you had a 2GB card, then added a SLIed 4GB card into the mix, on normal SLI you lose the extra 2GB vRAM. In that situation you might be able to regain the extra ram from the other card and give it to the first, so would have an effective 2x3GB (though there would be a few problems with latency between the cards...but PCIe 3.0 might have some bonus). Then down the line you could boost your current GPUs vRAM by adding surrogate 6/8GB vRAM cards, or even add-in vRAM cards, like those old system ram cards.

In games where SLI wasn't supported the second card would be able to use extra ram, making multimonitor things a little easier without the second GPU. SLI/Crossfire don't just use AFR, they use SFR/extra AA modes occasionally (like the 32x+ stupid AA modes where having a load of extra ram would be helpful).
 
You must log in or register to reply here.
Back
Top