OctaneRender™ Standalone 3.00 alpha 2

[Notiusweb]

Maybe a separate phenomenon, but maybe related to my above post about laggy viewport UI. Please note photos were with Alpha 1, but same occurs when using Alpha 2.

Same scene:
(1) V2.23.2 at default loads at ~58 memory usage

(2) V3 at default loads at ~388 memory usage

(3) V3 loads at ~729 memory usage if Max tile and parallel samples and are set to max (used to match render speed of V2.23.2)

What might be going on here...

[smicha]

Reduce parallel samples to 1 to save vram memory

[coilbook]

Thanks!
Will Octane 3 have Procedural Gradient map like in 3ds max

[Notiusweb]

So a trade off for environment medium is speed, but will it also be a permanently clunkier UI with multi-GPU?
Is that the sacrifice for the awesome new effects? WAAAAAAAAAAHHHHHHHHHH!!!!!!

[glimpse]

In a nutshell, the more GPUs you use, the worse the Viewport lag (panning, rotating, zooming) gets, even if a very low memory project. Again, interesting that this bad lag is not present in V2.23.2.

I'm not a dev, so this is just a personal opinion based on basically nothing..but my guess that Your system is not capable to handle that much of GPUs.. in order not to bottle neck them ideally You would have to put them on fast lanes (not x1) & then have enough CPU power to feed them all. If that's true, that might change a bit the way we build systems, as now we do not care too much 'bout speeds etc..but maybe we should start to? =)

[grimm]

There might be a lot more traffic on the PCI bus than before? I haven't seen any lag, although I just have two GPUs and they both have 8 lanes.

[Notiusweb]

Post by glimpse » Sun Dec 27, 2015 12:21 pm
I'm not a dev, so this is just a personal opinion based on basically nothing..but my guess that Your system is not capable to handle that much of GPUs.. in order not to bottle neck them ideally You would have to put them on fast lanes (not x1) & then have enough CPU power to feed them all. If that's true, that might change a bit the way we build systems, as now we do not care too much 'bout speeds etc..but maybe we should start to? =)

Tom, I defer to you as a Guru of Octane. I think you are right, I would have increased performance at all x16 PCI bandwidth in this scenario. The software is here extending perhaps beyond the limits of hardware. I myself was not to keen on going from X79 to X99 because the PCI lanes were at best only equal to X79, perhaps a future build with the newer Pascal or Volta GPUs will breeze right through all features of V3 with no lag, slowdown, etc.

The developers do note:

Speed
It's hard to quantify the performance impact, but what we have seen during testing is that in simple scenes (like the chess set or Cornell boxes etc.) the old system was hard to beat.

and

To solve these issues we moved the film buffer into host memory. Doesn't sound exciting, but has some major consequences. The biggest one is that now Octane has to deal with a huge amount of data the GPUs produce. Especially in multi-GPU setups or when network rendering is used.

To me, what they are alluding to is the V3 is slower by default and more taxing on GPU in order to provide the new effects, and we can try to best finagle settings to meet our needs.
All I can do is note that V2 did it, so if V3 Alpha can be optimized to "do it" too, great. if not,
"WAAAAAAHHHHHHHHHHH!!!!!!!!!!"
12-GPU-4-Life

[Notiusweb]

Post by grimm » Sun Dec 27, 2015 5:25 pm
There might be a lot more traffic on the PCI bus than before? I haven't seen any lag, although I just have two GPUs and they both have 8 lanes.

I will note that if only using 2-3 GPUs, for example, the lag is not noticeably there. As such, a 1x PCI speed is not an issue unless you have 8, 9, or 10 GPU, so it may not be that 1x is a problem, by default per-say.
I would be interested to hear what someone with 8+ GPU at x4, 8, or 16 is experiencing as far as this goes. It may in fact be the case that for any PCI bandwidth speed in V3, as # GPU increases, the UI Viewport response time will be linearly slowed down, so it is worth noting to the developers that this occurs. Especially if the CPU is tied into it somehow.

Someone run a 12-GPU test with all Titan Xs at x16 speed and let us know what happens.

[Notiusweb]

Abstrax, I thought of something that could be useful to someone with a multi-GPU lag.

Basically it would be some sort of hotkey which could toggle an 'active GPU profile' without having to pause the render, go to File>Preferences>Devices, and then select GPUs by clicking the little check boxes.

In my example, I am experiencing serious lag in V3 with 12 GPU in the viewport when I pan rotate or zoom, but none at 1 GPU. It occurs regardless of which Titan Z is selected, as long as just 1, smooth.
If I could set a 'Profile A' to have 1 GPU active, and then 'Profile B' to have 12 GPU active, and then enable a keyboard command or function key (ie F5 or whatever) to instantly toggle between them, it would make the workflow easier, as I could snap back and forth between posing scene with 1 GPU and then letting it render with 12.

Of course, it would be just as useful for the V3 Alpha to have the UI performance of V2, but if this is not the direction V3 will be moving into, an active GPU profile snap could be a very handy feature to have.

Thanks.

[gardeler]

Some things I have run into during tests:
-Looks like displacement is killing the motionblur. I have a scene with some debris flying, and decided to ad some displacement to the break edge. But that somehow removed the motionblur from that edge.. (but not the rest of the shape) Only tested in alpha1

-Alpha two cannot read my vdb's, but alpha 1 can. (only on linux, windows cannot in alpha 1 either.) I will try to send you one for testing.

/Martin