I’ve a PC build with RTX 3080 with 10GB VRAM. 10GBs aren’t enough virtual memory for me when I have to render out a heavy file, as it fails to render by giving Error: Out Of Memory.
I want to upgrade my build. Now the question is, if I upgrade to having dual RTX 3080s, will that help in optimising the memory? I am confused as 3080s don’t have NVlink in their hardware. But then, would memory still be split through the motherboard and fix the memory error?
Or should I get an RTX A5000 with 24GB VRAM? 24GBs should be fine for me. But will RTX A5000 work together with my RTX 3080 to give faster renders?
My PC’s specs below, if needed -
Ryzen 9 3950X
MSI MEG X570 ACE
MSI Geforce RTX 3080 Gaming X Trio 10GB
The memory of multiple cards will not be multiplyed by the count of cards you have, because they all have to load the same content into the VRAM. So for you it would be better to buy yourself a card with more memory!
And as far as I know (but I’m only 75% sure) two different cards would work together, but the card with the smaller RAM would become the bottle neck referring to the memory. But I would do some recherche to be sure!
If you need more VRAM you will have to not use the 3080 and get a card with more memory. VRAM is memory is not cumulative.
Check posts here in the forum about use of dissimilar models. But if you do you will be limited to the worse card VRAM. For heavier scenes you have to tick off the 3080.
VRAM cumulation in consumer segment is currently only possible with nVLink, which is supported by RTX 3090.
But still, it depends on the software if it is supported. Not sure, if Blender can handle this correctly.