Hello everyone. I am trying to paint my roadmap for production upgrades in the future. I was curious if I could get your input on what direction would be best for GPU acceleration. I currently own a Dell T5610 with the 685W PSU and a k2000.
While the PSU sounds relatively powerful, it has isolated power rails. This is both nice and annoying. It’s nice because in theory it means that Dell Engineers can distribute power independently from other components. It’s annoying because I only have 300W of power for the GPU total. I’ve tested this on a massive Asus OC 780ti. I can upgrade to the larger PSU, but haven’t looked into that yet. So this is a little limiting when we’re talking about multiple GPUs.
Currently, I render everyone on CPU, which is OK, but as we all know time=money.
The next option would be to use a Tesla or GRID compute card. This would have the benefit of having everything in one package. But I don’t know what my performance would be looking like compared to gaming cards. Currently, this doesn’t seem very cost effective.
I could likewise have an external networked render node, which would have the flexibility to render as a cluster with my workstation. I could use consumer GPUs and a gaming motherboard, install into a 3U package and rack it with my audio equipment. Using four GTX 970s, this would cost about as much as a lower-end GRID or TESLA K20, and, unless I’m overlooking something, would probably perform better also.
But dreaming about it is one thing, and actually implementing is another. Obviously if I’m going to invest another $2-3000 on rendering, I’d want it to easily implement into everything I use: Blender, Premier, AE, Maya and eventually (hopefully) Nuke.