I would like to beef up my ability to render my own animations at home. I currently have an i7 notebook with NVIDIA GT 330M graphics card. I am considering two paths:
Path 1 - Four i5 fast clock cpus, with network card, one master hard drive with the other three cpus booting through PXE using Loki.
Path 2 - Same as Path 1, however, I would go with the normal clock i5’s and put a high-end gpu in each cpu and render with Octane or another CUDA or OpenCL renderer.
Cost plus Blender compatibility would be of prime importance. For instance I have read that Octane will not support all of the wonderful textures, materials and other goodies I may be able to apply in Blender since it is independent of the software.
Can anyone please add their two cents? I am hoping to land this small, paying animation project and I would like to be able to render at home relatively quickly.
Blender Internal, Yafaray, Luxrender and many more… all pure CPU
Octane… pure GPU
Luxray… GPU+CPU
So with path 1 you got the most freedom of choise. I´d go with the i5 and only use boards with onboard graphics.
GPU rendering is just not too fit yet. Octane is nice, but not really production ready for animation IMO and Luxray isn´t either.
Also bear in mind that you need 1 license for each octane workstation.
So go with path 1, render with brute CPU power for now and you´ll be happy and in the future you can put GPUs inside them.
I´d just make sure you get mainboards that have 2*PCI-E electrical, then you can just put 2 cards in each i5.
And there is path 3, try to grab a used bladecenter from ebay.
There are often used ones for sale, ~50-90GHz total power, usually with RAM and PSU without HDDs for around 800-1600 Euro.
For instance here: 56 Ghz for 800 Euro with 4GB memory each, onboard graphics adn 4*1.8KW PSU and no hdd.
For path 3 you can always buy a tesla blade or a attach a machine with a 416xPCI-E board and 4GTX560 for instance lateron on demand.
Thank you, arexma for your comprehensive and helpful reply. I am going to try and put together a list based upon path1 and I am going to also look at the blade. Cool.
Thanks again, Rob.
Another thing, you might want to look into virtualisation/cloud server having server blades, setup one virtual system or use ubuntu cloud server and this way have one system that uses the resources at hand on demand, or on how you set it. And maybe setup a VPN gateway, this way you can hook remote machines into your local net and use as rendernodes, as loki has the disadvantage of only working within the LAN and even if you masquerade and forward, who wants his rendernodes connected to the outside =)
Loki is really easy to use, Dr.Queue is more powerful but tough to setup but personally I feel that B2.5s netrender is grand. It isn´t 100% working yet, but I already tried to use it for a smaller production. It is especially great because you can make a single frame testrender and distribute it automatically, it´s basically you press F12, the machines render the tiles and send it back to you, this way you work as usual but got the power of all machines at hand.
I have to say I haven’t ever used B2.5s netrender. I’ll look it up in the help and give it a try. I like staying in Blender, and for now my needs are projected. I have not hit a bottleneck yet. Let’s see…