I am not sure where I should post this question, so I will post it here. I hear that cycles does not perform well with 2x graphics cards as the second card only speeds up the renders by about 40% (is this still true btw?).
My question specifically is regarding a high end computer with 2x cpu’s. Will cycles perform the same way with dual CPUs? Will blender even run cycles on 2x CPU’s in terms of creating a single picture or is it just for animations?
it would be fast but not worth the money… if you don’t do research it could even be slower than a single i7 4930
gpus scale good its not like sli its cuda pretty much 95 - 100% scaling bvh building is still on cpu
you also have to remember tiles when rendering on gpu overhead can ruin your render times
heres a benchmark the top one is dual cpu (my 2 mid range gpus 300 dollars for both now are 2x faster than that) its 3000 for the cpus alone
if you look below it you can see i7s which are about 300 dollars and high end i7s are about 600 - 1000
8 core 16 thread i7s will be coming out supporting ddr4 soon it would be smart to wait
when rendering its better to go midrange and save money
If you use dual gpu’s, does it combine the ram of both cards? I ask because most of my scenes lately have exceeded the ram on the Titan cards used on Renderstreet and it switches to CPU rendering. @LordOdin, what gpu’s do you have?
no, ram is limited by the lowest amount of ram on a single card. if you have 2 6gb titans, you only get 6gb to work with. Downscale some textures or find some way to reduce detail if you want it to fit in a limited amount of ram.
From my experience, Cycles scales well both on two CPUs and on two GPUs. And this is valid for still images as well - both resources will be used for the render. There are a couple of things you need to consider though:
for Cycles, a high end GPU will be faster than a high end CPU in the majority of cases.
there are features in Cycles that don’t run on GPU (SSS is not officially supported, OSL doesn’t work)
if you have two GPUs, you need to make sure they are both selected in Blender’s settings
up to v.2.69, when rendering on GPU, the render also occupies a CPU core per GPU. I see that in 2.70 this has been optimized and the CPU usage is significantly lower
your project needs to be set so it can use the resources in an optimal way. If you set the tile size so that your entire image is one tile, the second CPU or GPU won’t be used. That’s an extreme example of course, but the reasoning behind it applies in various scenarios
I have been rendering some pretty heavy nature scenes with a large amount of instanced trees, shrubs, rocks and grass. Stuff like arch viz isnt an issue.
For what it’s worth, I’ve found that Blender scales beautifully across CPUs. The more cores, the better.
And I also must say, 6GB isn’t much at all. I’m working on a Tron light cycle as they appeared in Tron Legacy; to get the light trail to have the same fluidic distortion as seen in the movie required an incredibly high-poly mesh. The light cycle alone was 3GB, and that didn’t include the rest of the scene.
It’s for that reason that I prefer CPUs. I can add memory to each processor when need be, easily upgrade chips, and have access to all renderers, many of which are exclusively optimized for CPU work.