OTOY Forums

Hi there,

I co-run a small studio and we currently have 3 workstation computers. Up to this point we've done all our rendering with dual 980ti's in each of our workstations, or used an online render farm, but we just upgraded to a single 1080ti in each workstation, and we would like to build a dedicated 1080ti render station for a bigger project that is just about to start.

We don't necessarily need to use this render station for anything but rendering, and we would like to build something with the option of expanding this down the road for future projects. I'm not sure the best approach at this point.

I'm assuming using Octane network rendering is the idea here, but does anyone have any advice on what type of rig(s) would make the most sense for dedicated rendering and not for actual animation? Or how many 1080ti's would the most make sense in each tower? Would really appreciate the advice, Thanks!

Maybe something like this:

viewtopic.php?f=11&t=62530

Or this:

viewtopic.php?f=11&t=58045

Thank you for the links!

or this: viewtopic.php?f=11&t=61106

rappet wrote:or this: viewtopic.php?f=11&t=61106

Nice looking build rappet, great work!

If I was using Octane network rendering and the budget was available, would you say it would be more ideal to have 2 rigs similar to the build you made with 3-4x 1080ti's in each, or a single 7x 1080ti rig similar to the builds of Tom Glimps or smicha?

I would assume 2 rigs could have more costs with the extra license and componentry, but if there is a tight schedule on a project would it be safer to have 2 rigs running in case one might have issues? Or are there any other things to consider?

Thanks

colekush wrote:

rappet wrote:or this: viewtopic.php?f=11&t=61106

Nice looking build rappet, great work!

If I was using Octane network rendering and the budget was available, would you say it would be more ideal to have 2 rigs similar to the build you made with 3-4x 1080ti's in each, or a single 7x 1080ti rig similar to the builds of Tom Glimps or smicha?

I would assume 2 rigs could have more costs with the extra license and componentry, but if there is a tight schedule on a project would it be safer to have 2 rigs running in case one might have issues? Or are there any other things to consider?

Thanks

The difference of one extra license is neglectable I think... when the price is not more difference the couple hundred dollars/euros the I would prefer building two gpu rigs (with 7 gpu) instead of one rig (with 7 gpu).. easier in future to adjust config, and also having backup just in case.

When using the rigs only for networkrendering as slaves, and not as a pc to work on projects or use for other purpose then just slaves.. then you could build 1 or 2 rigs with less specs.. less ram, less cpu.. building two rigs with 4 gpu and less specs might be chepaer then building one rig with 7 gpu and more specs.
I build mine as a renderslave, but also as a second working station, therefor I have chosen higher specs then if I would build a renderslave only.

i.e. I had build a renderslave once by using an old PC that I was not using anymore with old mobo few ram and less cpu then these newer ones, and stuffed it with new gpu's.
You have to make sure you enough pci lanes to run the number of gpu ofcourse.

I think you have to do some math by comparing the configs you need,
after first making choice building working pC or slaves.
Maybe then building one 11 gpu rig (like Smicha is doing) might be the relative cheap way to go.. I do not know.
Ofcourse it can be much more convenient having 'just one'slave instaed of 3.. you have to monitor more PC's..
I did not go that route though.

That is my two cents.

greetz,

edited/added: Oh, and having at least one slave on air and not on water has benefit of swapping newer GPU in few years easily.. I have done that a few times now.. going from 580 to 780 to 1080... On water has benefit of lower temps on card and you might be able to stick more cards in one config... but just a thought to makae you think about future plans as well

And a four hybrid gpu beast is an option too. As we say in about a dozen threads, the Thermaltake Core X9 case will easily accommodate four gpu rads and a 240mm cpu rad, all at top of case.

https://www.evga.com/products/product.a ... P4-6598-KR

http://www.thermaltake.com/Chassis/Cube ... design.htm

rappet wrote:then you could build 1 or 2 rigs with less specs.. less ram, less cpu..

Thanks for the insight rappet, this all makes sense. I'm thinking 2 rigs might be the right path here, and these will be just slaves.

FrankPooleFloating wrote:And a four hybrid gpu beast is an option too

This Thermaltake Core X9 case looks like a great functional solution for what I'm looking for. I love the stackable feature too for expansion!


Rappet, your point about air-cooling is good, and this coincides nicely with FrankPooleFloating's idea for the 1080ti Hybrids in the X9.
So if I were to build a 4x 1080ti Hybrid using the Core X9 to be used as a dedicated slave (not a workstation), would anyone have any advice for components while keeping costs as minimal as possible? I can't seem to add 4x 1080ti on pcpartspicker for compatibility checking, so I'm still just researching.

Reading some other threads on Slave build here it seems 16GB might be a healthy minimum for ram, but 32GB is probably better? Is there a cheaper motherboard and particular 40 lane cpu that would be ideal here? Along with the 240mm cpu rad mentioned, I would also assume 1000W minimum for a psu?

Seeing as this is my first dedicated slave, would you recommend any monitoring units? I see some images of the X9 with a display built into the case. I could install the OS on a small crucial SSD, and have plenty of cheap seagate storage linked to the shared dropbox for renders. I feel this could work quite well.

Thanks

I run a slave with just 12 gb ram and bought a small square cheap monitor to monitor the slave with speccy and msi burner

rappet, what about out of core textures, voxelization, etc?.. I was under the impression that folks should have at least double the RAM as their GPU VRAM...

32GB would mean you could send just about anything to render, with confidence, and not have it poop out during a deadline jam... Yeah, I'd personally spend that $150 and get another 16GB, for 32 total.