Hello everyone, in what way is GPU memory used in rendering? I’m thinking of replacing my RTX 3050 and would like to know if the KS uses all available memory. Will a 12Gb GPU perform better than an 8Gb one, speed gain?
I’m not sure specifically how the memory works with Keyshot, but I know that the larger your scene is the more memory it will need. If you work with big or complicated scenes you will probably run out of memory pretty quickly with an 8GB card.
My CAD station has a 12GB card, which albeit isn’t really optimized for renderings, but when I got a new dedicated rendering computer with a 16GB RTX card the difference was like night and day for me. I do work with very big scenes though, and you might not notice as much of a difference if you mainly work with smaller projects.
I’m not sure how much difference it does for plain speed though, since I can’t really compare that aspect between a slower CAD card and a faster RTX card.
The speed gain you get will be caused by the amount of CUDA cores in combination with the clock speed of your GPU. I’m not sure what you are aiming for but instead of a 12GB of a 40xx series you could also consider a 24GB 3090 second handed.
For a global idea you just count the CUDA cores, a 3050 has 2304 and a 3090 has 10496. So roughly a 3090 would be more than 4x as fast as a 3050. It’s also higher clocked but that is not the biggest thing for speed.
A 4090 would have 16384 cuda cores and because it’s also a new generation it’s close to 2x as fast as a 3090 (I’ve tested both). It has also some extra cores which accelerate ray-tracing so that will make it that it performs better than just 1.6x looking at the cuda cores.
Looking at the prices of a second hand 3090 or a new 4090 the 3090 is the better pick. If you have a real heavy PSU and enough space in your case you could run two identical 3090s and use a NVLink in between them and have a hopping 48GB of VRAM. And a great render speed.
The NVLink option is not available anymore on 40xx consumer cards. So the 3090(ti) is the last one where you could pull this off and get 48GB.
The memory doesn’t mean nothing for speed but all your textures have to fit the memory of your GPU to be able to render a scene. I went from 8GB to 11GB to 24GB and a lot of my scenes wouldn’t fit in 8GB unless I used lower resolution textures. So it was a bit more of a puzzle. And if you just put 8k textures on everything 24GB won’t be enough as well, so you just need to consider it all even when you have a lot of VRAM available.
I’m actually planning on having both the 3090 and 4090 in my PC and while I have a big case I need a bigger PSU (have 1000W now) and need to move some stuff around so it will be quite an operation. Also because at my motherboard the PCIe slots are closer to each other while on some board there’s an empty spot extra between them.
Anyway, hope you’ve enough info and else just ask.