And another GPU question ... RTX 3060Ti (8GB) versus RTX3060 (12GB)
After buying a new PC (Intel i7 12700K, 32GB RAM, M.2 SSD module - OMG - ist it fast ;-), I'm also thinking about a GPU. Given the current prices, the RTX 3060 variants are the most interesting for me. And I have a simple question and hope fore a simple answer ;-)
Which one is the better choice for rendering (in DAZ Studio of course)?
- The 3060Ti has more raytracing and tensor cores, but "only" 8GB VRAM
- The other one is approximately 100 Euros cheaper and has 12GB VRAM, but only 28 instaed of 39 RT-Cores.
Both are connected over PCIe 4.0 x16. I think, thats important and therfore the RTX2060 is out (although I'm no expert).
What do you think? Currently the prices are falling a bit and that's why I'll wait a little longer ...
Sideinfo: Currently I don't have a card. The integrated UHD-GPU (or the CPU itself - I don't know) is very much faster than my old GT710 *LOL*
I'm using 3Delight and the render-time improved dramaticaly but unfortunately dForce is still not possible.
Update: dForce works after I installed the latest UHD graphic drivers from the intel-site. It takes about 4 minutes for a simple skirt and default settings. I have no idea, if this is typical (probably not).
Comments
The 12GB variant is the best value in my opinion.
Yes it has fewer RT cores, but you can fit more on the card so it falls back to CPU less.
I bought the 12GB version myself.
I'll suggest you want more speed than RAM. 8GB on a video card in DAZ Studio is more than most DAZ users can throw at it.
For IRay, I would take the RTX 3060 12GB over the RTX 3060ti 8GB (and maybe even my RTX 3070ti 8GB), especially if you are rendering D-Force Strand-Based hair, which can use a lot of VRam. I have exceed 8GB on my RTX 3070ti rendering two characters clad with Strand-Based Hair. When you exceed 8GB, rendering reverts to CPU-based, which is terribly slow.
At this point, I would really just wait for Hopper.
Here's a little info on the Nvidia "Hopper" tech.
https://www.tomshardware.com/news/nvidia-hopper-h100-gpu-revealed-gtc-2022
incase you are wondering how much a Volta V100 costs :)
https://www.amazon.com/NVIDIA-Tesla-Volta-Accelerator-Graphics/dp/B07JVNHFFX
No info for the new consumer "Ada" card architecture yet.
Depends on what you are doing in DS.
For animations, the 3060ti may give you faster results, but for still images the difference in speed is negligible => You want the 3060 for the 12GB VRAM.
Yes, you can manage and do a lot with an 8GB card but you can't just throw everything and the kitchen sink in to the scene and expect it to render, you need to understand what to use and what not to use, or know how to fix items that are not optimized.
For the last 1.5+ years the items have started using more and more VRAM, partly because increased texture sizes and number of related maps, but also due to bad UV management.
I have a RTX 2070 Super (8GB) and it typically renders complete scenes with 2-3 clothed figures, architecture and environment in 10-20 minutes, single figures in a few minutes - As far as rendering speed is concerned, I would not be looking for an update, but I would like to get more VRAM.
In practice a 12GB card will have almost twice as much available VRAM for Iray rendering than an 8GB card due to the base load by the OS, DS and the scene.
Wow - so many answers in short time and all are very helpful. Thank you very much.
I understand now that both variants has their advantages and disadvantages (as expected). I think, 12GB RAM will be the more benefit for me, especially because this card is also cheaper. I don't do much animation at the moment - mainly still images. And of course - two or three figures at the same time plus backgrounds should be possible, when I spend about 500 Euros for a card ...
@Nvidia Hooper: Of course that sounds interesting. However, the second link posted by IceCrMn is interessting, too ... *LOL*
( but good to know, that my CPU/chipset is able to connect those cards over PCIe 5.0. So I'm well prepared for future scenarios - when I get rich :-)