Network Rendering - Performance observations

[glimpse]

Hi, Guys.

so I took a spin with new feature for OctaneRender called network rendering =)
setup was straight forward without too much of issues (just my own stupidity =)
& after few minutes I started rendering..

I gues it's fine to see the scene being exported & sent via LAN to other PC.
(at the bottom line Octane informs 'bout the progress on the process).

However I noticed one thing that I'd like to share & ask developers, is it normal?
('cos I believe sooner rather than latter we'll see this question asked anyway).

take a look here:

That's a printScreen from GPUZ of Render slave rendering. Now the question is:
Why it isn't working all the time & we see drops (on GPU load)?
I started lookin' more at this as I noticed samples/second changing.

If it's normal & that should be this way - it's ok, just want to hear that.
If it's not normal, just sharing if You haven't observed this issue so far =)

[suvakas]

Sry for intruding into your thread like this, but since you've done it I think you know the answer.
Do you need a separate Octane license for slave? I looked everywhere, but I found no info about it.
If I want to setup 2 workstations for rendering (one master and one slave), then do I need to have 2 licensed copies of Octane too? (one for each Pc)

Suv

[bepeg4d]

Hi suv,
Yes you need a standaole licence for every slave node.
ciao beppe

[glimpse]

Sry for intruding into your thread like this, but since you've done it I think you know the answer.
Do you need a separate Octane license for slave? I looked everywhere, but I found no info about it.
If I want to setup 2 workstations for rendering (one master and one slave), then do I need to have 2 licensed copies of Octane too? (one for each Pc)

Suv

no problem, Suvakas =) Yeah You need separate licence for every PC

(it's the same regular one at least untill now. Maybe, they will release some sort of discount for bunch of licences - though I'm not sure how to manage that these wouldn't be used as normal licences. So let's wait & see how OTOY is going to deal with that)

[whersmy]

Maybe it has to do with bandwith being interrupted?

Do the gpus scale nicely?

[abstrax]

Hi, Guys.

so I took a spin with new feature for OctaneRender called network rendering =)
setup was straight forward without too much of issues (just my own stupidity =)
& after few minutes I started rendering..

I gues it's fine to see the scene being exported & sent via LAN to other PC.
(at the bottom line Octane informs 'bout the progress on the process).

However I noticed one thing that I'd like to share & ask developers, is it normal?
('cos I believe sooner rather than latter we'll see this question asked anyway).

take a look here:

1.png

That's a printScreen from GPUZ of Render slave rendering. Now the question is:
Why it isn't working all the time & we see drops (on GPU load)?
I started lookin' more at this as I noticed samples/second changing.

If it's normal & that should be this way - it's ok, just want to hear that.
If it's not normal, just sharing if You haven't observed this issue so far =)

Assuming there is no change in the scene that restarts rendering: During the initial samples the utilization of the slaves may fluctuate a bit and with that the samples per second. In the beginning Octane distributes small work chunks to the slave and then ramps it up. As a result you may see some fluctuations in the slave GPU usage and the samples/s. But after a few viewport updates (seconds) it should have stabilized and the slaves should always have enough work to stay busy while they send back results and wait for the next work load. If that's the case, then there is some kind of bug/issue/bottleneck.

What is your configuration? How many slaves do you use, what is their GPU configuration? Which operating systems are you using and which network connection (100Mbit/s or 1Gbit/s)?

[glimpse]

What is your configuration?
How many slaves do you use, what is their GPU configuration?
Which operating systems are you using and which network connection (100Mbit/s or 1Gbit/s)?

for the sake of just testing the feature
(as some guys might want to leverage GPU on home lan)
I've used 2x casual PCs, nothing special:

Z77, i5 3570K + GTX670 (wireless USB adapter 300mbps)
Z97, i5 4670K + GTX660 (directly wired to 1Gigabit router)

I've tried to render from both systems (changing slave/master order).
Observed issue was more or less the same in both ways.

& Yes, You can see from printscreen it seems those 0% load gaps get
less frequent (just noticed that after reading what You've wrote).

Maybe testing longer would provide way better insight
(as performance will even out on a long run). At the time of testing
I've simply took it as 'normal operation'


Please, feel free to add any info to this thread if someone is testing this,
(as at the very moment only one system by my side =)

if needed GPUZ could be downloaded here: http://www.techpowerup.com/gpuz/

[bepeg4d]

Hi tom,
here is my snapshot from gpu-z, the master is a mac and the slave the pc via wifi:

the two gaps are because i have rotated the model in the master viewport, but as you can see the load is always at 99% when there are no changes.
maybe your scene is too quick to render and the network too slow?
ciao beppe

[whersmy]

Can`t wait to test this feature out.

[glimpse]

Hi tom,
here is my snapshot from gpu-z,
...
the two gaps are because i have rotated the model in the master viewport, but as you can see the load is always at 99% when there are no changes.
maybe your scene is too quick to render and the network too slow?
ciao beppe

thanks, Beppe,

Most likelly this was caused by something on a my lan

Here I've used scene that I'm sharing in my signature &
while rendering I haven't touched anything in Octane.

again, thanks for info (good to know this works perfectly),
I will try to look for the cause on my side =)