GPU rendering with modern cards on a 3,1: bottlenecked?

[nognome]

First off: I'm in the same boat as a lot of you folks here waiting to figure out Apple's next move with the Pro. I'm an animator that does a decent amount of 3D work, and i've been looking into the Octane render engine for Cinema 4D. It's currently only CUDA but they claim it'll run on OpenCL by next year – so I figure that if I do end up on the next Mac Pro, whatever it is, I can't go wrong if I end up liking the engine.

With the lack of new info after WWDC, I want to pull the trigger on a GPU update in the meantime and start playing around with Octane and getting a feel for it. I've done a certain amount of research on using the latest generation of PC GPUs with a 3,1 and read that they're bottlenecked by the CPU for the most part, which is understandable.

Here's my question: most of these folks seem to be using the PC cards just for games. I can get how it's bottlenecked since they'd need that CPU power to feed other stuff than just graphics. With GPU rendering, though, i'd imagine it's really more about the raw GPU power rather than the rest of the computer's architecture having to keep up.

Would this sort of usage produce the same bottlenecks that games do, or would I be able to get the same level of render power out of a good GPU that a more modern machine would? It would be pretty great to wait until Pascal drivers get released and then pop in a 1070 if that sort of thing would run in a 3,1.

Thanks in advance for any advice.

 

[Synchro3]

AFAIK related to GPGPU you will not bottleneck on 3,1. Here are lot's people with capable GPU's running on 3,1.

Note that capable GPU's often need an external PSU.

 

[goMac]

AFAIK related to GPGPU you will not bottleneck on 3,1. Here are lot's people with capable GPU's running on 3,1.

Note that capable GPU's often need an external PSU.

There are a lot of possible bottlenecks for GPGPU on a 3,1, depending on how you are using it. RAM speed is slow, so streaming data to the card is going to be affected. I'm pretty sure there are also significant limitations to how much data the PCIe backplane can handle at once, which isn't an issue for the older cards, but causes problems with large amounts of data on newer cards.

You can run a newer card in a 3,1 Mac Pro, but it's going to be slower than if you were running it on a 5,1. There is a performance ceiling even for GPGPU. If your data sets are small, you might not be affected too much.