OTOY Forums

Hi,

Forgive me if this has been answered elsewhere in the forum.

We're looking into setting up a standalone GPU rack connected to our office network (say a machine with 4-6 cards?) that can be accessed using the network render settings. There are around 4 people in the office who use Octane to render large interior scenes.

Is this possible to set up and if so can multiple users use the same cards at the same time?

If not would each user need to set up the scene locally and then notify the others that they were using the GPU rack and make sure no one else tried to access it while the render was running?

Only one person can use each GPU at a time.
If someone else wants to use the rack and someone else is already using them, they GPUs would not show up. If they are available, it will say so.

I havent tried it, but you might be able to setup a slave for each GPU on that computer and then they can choose how many of them to use at once. Otherwise, it is all or nothing.

On the plus side, once connected to that computer, the images will go very fast so it may not be much of a problem unless each person has multiple images they need to render out.

Thanks for the reply Kavorka.

It looks like the easiest option will be to set up a slave machine on the network with 4 cards that we can simply select through network render settings when we need the extra power. If they are already in use they should show up as occupied.

I don't suppose you know what you'd recommend regarding cards. Are the new GTX970 cards better bang for buck then the Titan Black?

Tipperz wrote:Is this possible to set up and if so can multiple users use the same cards at the same time?

as far as I know there is a possibility to use one GPU for multiple users..Renegatt Software is working for solution how to virtualise GPUs & deliver them to any user (these could not only be Your own GPUs in a server, but also in the cloud instances like those from Amazon servers). This solution is pretty complicated & it's far more tricky to setup compared to OTOY's own network rendering solution.

Only thing to keep in mind is that using virtualised GPUs is against EULA (if they haven't changed something).

So what would be the best option?

Option A: I would build a rig (or two) with 4GPU in a PLX in a Z97 motherboard (as You don't need so much CPU power) . Even if You have like 4x GPUs inside Your main workstation You still could utilise those two 4GPU nodes (reminder: total amount of GPUs is to be used with network rendering is no more than 12).

Option B (even more elegant) 4x TitanZ (8 GPUs as seen in Octane) in the same ATX format board& case, modified to be liquid cooled with single external MoRa radiator (see this thread for explanation: http://render.otoy.com/forum/viewtopic.php?f=21&t=43101)

* & buy extra licence for every node (as You need them in order to use slaves).