I hope this topic hasn’t been asked about to death. I tried searching for answers, but I didn’t really find definitive ones.
As far as I understand. Rendering a scene in cycles on your GPU is fastest (if the GPU and CPU age are in about the same class-range), although you are limited by the GPU’s memory.
Another thing the GPU isn’t able to do, is handle any of the physics well (if at all?)
So if for example I have a short walk cycle scene of a woman with; softbody physics(breasts), cloth physics(dress), hair physics, smoke physics (cigarette). Would this be impossible to render with the GPU?
Or would it be possible to pre-bake a lot, or all of these things, before rendering with the GPU?
My current plan for a new PC would be a AMD Ryzen 7 2700X CPU, and a GTX 1060 GPU. Simply because as far as I’m aware, a GPU won’t help much with with rendering when there’s physics involved. But I’d love to hear what my options are!
Thank you for your time!
not sure if theres any physX plugins for blender, but GPU rendering will work just fine with or with out physics. i just doubt the gpu could compute any of the physics.
CPU will calculate the physics, positioning the scene, then the GPU will take that and make a picture/frame of it. then do that again with the next frame, and on and on.
Maybe you’re confusing simulation with rendering. The simulation is done on CPU. Rendering all of it can be done with GPU, as long as there’s enough memory
There are recent changes that allows to render with both GPU and CPU, and also memory management.
Those should be available in the development version https://builder.blender.org/download/ but those builds aren’t recommended for any real work.
As others have said, you can still totally use the GPU to render a scene with pretty much anything that cycles supports, providing you don’t run out of video RAM. All the physics/simulation is per-calculated and baked by the CPU and at the moment, much (possible all) is single threaded. In other words, it doesn’t matter how many cores that fancy new CPU has, it will only use 1 of them to basically calculate where all the vertices are each frame. Once that’s done, the geometry can be passed to the GPU to render, just like anything else.
With that in mind, if you are really doing a LOT of simulations, then it’s possible that an Intel CPU with the fastest possible single core speed would be better (tho that’s not to say that the new Ryzen 2700X wouldn’t still be fairly good at it). However, for final rendering, plus all of the 101 little test renders one is likely do due during a project, a good fast GPU is still very handy.
Another plus for a faster GPU is Blender 2.8 and Eevee, nothing like seeing near photo realistic images in real time in your viewport. As such, while you could stick with the Ryzen CPU, I’d really be looking at more GPU power then the 1060. To start with, the 1060 tops out at 6GB, compared to the 1070 at 8GB and 1080’s at 11GB. Now sure, future Blender may well have combined GPU/CPU rendering with shared memory, but I’ll bet that if it can still all fit in the GPU RAM, then that will be better and faster.
The only word of caution at the moment is that GPU prices are rather inflated (cryptocoin mining) tho minor signs of starting to drop again and current word has it that Nvidia may well release new GPU’s by the end of the year.
Thank you all for your replies!
It looks like I’ll have to rethink my strategy…
I’m probably gonna get me a GTX 1070 Ti, or a regular 1080. I’d have to reach a little too deep in my pockets for the 1080Ti.
So that leaves me with the decisions for the CPU. My initial worry was that I wouldn’t be able to render a scene with GPU if there were Phsx simulations in it. But with that outta the way, I’m most likely not gonna go crazy with the physics simulations anyway.
So, bearing that in mind, I think I’ll settle with a tier down on the CPU, the Ryzen 2600X.