Network performance question

[Alteris3D]

My company put together a network render box for me with the goal of using Octane for our day to day animation work. I have a fairly complex scene I put together when we first started using Octane last November. I was rendering locally, with a Quadro 4000 and a GTX 1070. The scene I'm using for my benchmark scene would takes 18 minutes to render a frame using those two GPU's.

Anyway, my company put together a network box with four GTX 1080's. The Octane network render found the new device, and shows all four GPU's. While the render speed improved, the performance boost wasn't as dramatic as we had hoped. The same frame that took 18 minutes to render, now takes 7 minutes Also, while using "Open Hardware Monitor" to monitor GPU performance during rendering, I notice the measurement of performance on each of the four 1080's bounce back and forth between 0 and 80% at one second intervals. 0-80-0-80 etc. etc.,.

My question is, is this normal? And if not, what would be the best configuration to maximize the performance of this setup: 4 x GTX 1080, 1x GTX 1070? My bosses are asking why the GPU's aren't being used to 100%. Our IT guy has asked if it could be related to network speed and is wondering if we should install Lightwave on the new device so it sees the 4 1080's as local and I would work from my graphics suite over the network. Does that make sense?

Id appreciate any advice or suggestions the board could offer. If it makes sense to post this with OTOY support, let me know.

Primary workstation: Windows 7 Pro 64, 2 x Intel Zeon @ 2.5 GHz, 128 GB ram, GTX 1070 6 GB, Quador 4000 4 GB
Network server: Windows Server 2016, 2x Intel Zeon @ 2.4 Ghz, 64 GB ram, 4x GTX 1080 8 GB

[juanjgon]

Hi,

What are the scene kernel and imager node settings? There are a couple of parameters, like the "parallel samples", "max tile samples" and "minimize net traffic" in the kernel, or the "tonemap interval" in the imager node, that can affect the network rendering performance. Try to tune these parameters to see what happens.

Thanks,
-Juanjo

[Alteris3D]

Thanks Juanjo,

I was using a PMC render kernal before, and when I switched to direct lighting to test those settings you mentioned, the remote GPU's are running a steady 90-98% performance load, and after tweeking the direct lighting settings to get the the image quality I had before, I have my render times down to under 3 minutes a frame.

One other issue, I set up a render to run overnight using this new remote server we have, and it shut down about an hour after I left the office. I'm attaching a screen grab of the slave dameon window that was on screen when I came in this morning. Any idea on what may have caused this? I'm going to try to run another test tonight.

[3dreamstudios]

While I can't speak to the network rendering performance...as I too have my own thread for this before I upgrade some GPU's, I can say that switching from PMC to direct lighting will dramatically increase your performance no matter what you tweak from the other variables mentioned. So you may not have found a true answer other than now rendering with a different kind of look. Direct Lighting is very simple and great for exteriors. I too use it alot on interiors but fake a lot of the lighting for that bounced light look you'd normally get from PMC or other true global illumination kernel.

Good luck..and let us know what you find out about getting those 1080's rocking!

[asher]

Main causes of network failure are as follows:

1.- Overheating on the GPUS. try MSI afterburner to tweeak performance Vs Temperature. IF you have Watercooling, the performance improovement is dramatic.
2.- Using to much Vram, creates Cuda Errors.

Other alternative is that you have a bottleneck on your network, we have the Slaves connected directly to the main CPU via a standard Switch, but no other network trafic, no internet.

Possible issues with motherboard not handling well the IO of four GPUS in one box. Motherboards are very strange in that regard.

[asher]

Juanjo, Same issue here, upgraded from 3.03 to 3.05 today, and now my Titan X wont go 100% or overclocking!!!! it stays under 1500 Mhz, and with 3.03 upt to 2 Ghz! That is a a nono.
( Despite Afterburner set to overclock!!!! )
Looks like a bug to me.
If i cant fix this, i will go back to 3.03.

Thank you,

Asher