OTOY Forums

Hi,

Netrendering is due to the constant crashes of the Deamon slave
useless for me because a rendering overnight thus it is not possible.
Also 4 or 5 times restarting the slave means everytime 4 to 5 miutes
for sending scene data again, before the slave actually starts to render.

So another method would be using a GPU expansion box, but
with about 3.000,- Euros this solution is too expensive for me.

My question now:
is it somehow possible to use/connect a PC in the sense of an
Expansion box to my main master mashine?

Maybe someone has experience with it.

thanks
Mike

cheaper solution: PM to Abstrax this problems. He should fix it.

hi, Mike =)

there are solutions, from very simply & relatively cheap to some pretty expensive as You've probably figure out by now Yourself.

First of all You can use something like splitters to [urlhttp://amfeltec.com/products/gpu-oriented-cluster/]clusters[/url] (this one I really like as it takes care of housing GPUs & You don't have to think about how to put them!)

Downside of such solutions is that they are hooked via 1x or 4x & thus upload speed is cropped in way, but if You upload a scene once & then render for extended period, that's completely fine to live with (link speed influences only upload - rendering speed is the same as if the card would be hooked inside computer)

on the other hand there are "professional solution", like those netstor, magma, etc boxes that will be hooked from x16 - providing more speed, & they might look better..but are not as flexible & depending on cards You're using..-might not be very good in terms of cooling.

That Cluster solution if You'd be using good aftermarket coolers like Windforce, ACX, STRIX, etc.. is a good way to have not only good temps, but also relatively cheap build (a frame with all wires for 4xGPUs cos around 300-400 & two of them ~600$ =)plus PSU depending on power Your GPUs are going to take.

or..Simply use ORC (Guess Aoktar had this in mind! =)

Hi glimpse,

thanks for your comments!

These Clusters really looking strange, don't they?

I'm not sure about the upload-times you mentioned.
(With my last animation network rendering testing, data-sending to
the slave took longer than the actual render time of one frame,
so the slave never started to render.)
If these clusters supports 1x or 4x PCI speed, for animation, is the
upload-speed really not for every frame? Only once?

And sorry, whats ORC Octane Render C....?

thanks
Mike

miohn wrote:Hi glimpse,

thanks for your comments!

These Clusters really looking strange, don't they?

I'm not sure about the upload-times you mentioned.
(With my last animation network rendering testing, data-sending to
the slave took longer than the actual render time of one frame,
so the slave never started to render.)
If these clusters supports 1x or 4x PCI speed, for animation, is the
upload-speed really not for every frame? Only once?

And sorry, whats ORC Octane Render C....?

thanks
Mike

ORC Stands as OctaneRender Cloud that it's currently in beta stage (Guys polishing it out, but as I understand it works! & pretty good.) For now it's only for stills, but I guess they will make it animation capable (read somewhere about this..) - so if You don't want to invest into hardware that might be a solution. Though it depends on Your scenes..turnarounds, workflows (that influence file sizes) & Your internet connection..

overall tailor building something for Your needs would work the best I believe..whether that 1x or 4x works for You You need to verify Yourself by plugging a card into 4x slot to see how it goes in Your case with Your files etc..

what is Your current system & slave that You have (list parts) & maybe we'll give You some thoughts on how to improve or what to change.

I do agree these clusters look weird (it was my thoughts too), but with proper cards it actually makes a lot of sense - good airflows means You get more performance too, as cards are not throttled down because of heat =)

Just be avare You'll probably hit ceilings very fast as "only" 7-8 GPUs are going to work on mayority of motherboards
source: http://render.otoy.com/forum/viewtopic. ... 09#p217574

Hi Glimps,

I didn't know these clusters and think, that they will
be a good solution for me!
Not expensive and therefore worth a try!

With, for the moment, only "one" additional Titan-Z
I won't hit the ceiling I think!

thanks lot!
Mike

miohn wrote:Hi,

Netrendering is due to the constant crashes of the Deamon slave
useless for me because a rendering overnight thus it is not possible.
Also 4 or 5 times restarting the slave means everytime 4 to 5 miutes
for sending scene data again, before the slave actually starts to render.

So another method would be using a GPU expansion box, but
with about 3.000,- Euros this solution is too expensive for me.

My question now:
is it somehow possible to use/connect a PC in the sense of an
Expansion box to my main master mashine?

Maybe someone has experience with it.

thanks
Mike

Network rendering should be stable and from what we can see it's definitely used a lot. So which version are you using and what exactly is happening?

Hi abstrax,

Main maschine is dualcore xeon 64GB RAM 2 x Titan
Slave 2006K 16 GB RAM Titan-Z

Drivers, Octane Standalone, Plugin and C4D (16) are
the same on both computers. Connected via Ethernet.

Deamon starts automatically on slave and on main PC
the slave is listed correctly and activated.

Rendering starts on master and after (for this project)
5 min the slave is starting rendering. (green subline)
Then after 2 or 3 minutes Slave resufed/stopped rendering
(don't remember the exact wording in the CMD-Window)

After restarting the deamon-slave same procedure as before ...
After 5 min (sending data) rendering starts and 2 minutes later slave stops ...

hmmm?

thanks
Mike

Maybe also this post worth reading for a DIY approach.
http://render.otoy.com/forum/viewtopic. ... it=diy+gpu

interesting,

but I think you have to be a hobbyist
in building own cases etc. Not for me

regards
Mike