Upgrading the Graphics Card
I'm not trying to be bother-some; it's just that I am.
I'm considering upgrading my graphics card. I have an MSI RTX 3050 8GB VRAM with 2574 CUDA cores, 1807 Mhz. The research I've done this week strongly suggests that I consider the MSI RTX 3060 12GB, which has 3584 CUDA cores and the same1807 Mhz. DAZ3D hardware recommendations bluntly state that VRAM determines only whether the GPU or the CPU will render the scene and has no influence at all on rendering time. Further, they state just as bluntly that CUDA cores do affect render times: the more CUDA cores, the shorter the render time. Importantly, RTX 3060 specifications recommend a 550 watt power supply, which is what I have. The alternative card is the MSI RTX 4060 8G, which has 3072 CUDA cores and seems to require a 650 watt power supply, an additional and unwelcome expense. (MSI cards are the only ones I consider because MSI offers gaming and studio drivers; naturally, I install the latter.) So I have concluded that I am actually better off with the older card. I'll get more GPU renders and better GPU render times.
Remembering that money is money and just to be on the safe side, I will happily consider cogent arguments to the contrary. Thanks to one and all.
Comments
I would recommend going for a GPU with 12 GB or more VRAM. If you exceed VRAM you will fall back to CPU and that will be much slower.
And exceeding 8 GB VRAM can be done fairly easy unless you optimize your scene.
I also recommend a GPU with more VRAM.
To begin with, if the scene doesn't fit in VRAM, no matter how good the GPU's score is, it won't be effectively utilized.
Thank you felis and CES3D. I do appreciate your comments.
"VRAM determines only whether the GPU or the CPU will render the scene and has no influence at all on rendering time"
This statement makes no sense. That will affect render time more than all the other factors combined.
No, the point is that if the cene fites either way then the card with more memory will not be faster by virtue of its extra memory. Enough is all that is required or beneficial.
@TheMysteryIsThePoint: The attached image is from the DAZ3D Help Center on the role of VRAM in rendering. VRAM plays no role in render times. Let's say your .duf file is 4GB and you have a card with 8GB VRAM. The scene fits the GPU and will render there. Now let's say you have a card with 12GB VRAM. Again the scene fits the GPU and will render there; but it won't render any faster.
Yes, guys, I know that.
From how you wrote it, it sounded like you weren't aware of how big a deal it is if your scene falls back to the CPU.
But in any case, it sounds like you could consider the "null" option, that is, to do nothing. The 50 series is going to be available soon and there will probably be a better option. November before a release year is never a good time to go shopping for a GPU.
Sorry if I sounded officious, I just saw something that sounded fishy.
Bear in mind that with previous geenrations it has been a while between release of the cards and release of a compatible version of Iray - which then has to be tested and integrated into DS. Having a card that wasn't supported for several months would be extremely frustrating.
Really?! Just one more reason to render in Blender as if one were needed.
But I actually meant what I thought that the 50x0 cards will do to the prices of the previous generation. I've heard that it's getting hard to find a 4090, but the other cards are easier, apparently.
If people don't like the new cards (feel they are not worth the price for the advances etc.) then the new versions can actually give the old ones a fillip. And of course, on past behaviour, the cards lower down the range will not be part of the initial roll-out anyway.
What Have I Done?
Geez!