OTOY Forums

With only 1 video card (GTX580), Windows becomes quite unresponsive during CUDA rendering. I have added a 2nd video card (8800GT) to help with this, but I am not sure how to set the system to do the job correctly. The GTX580 has 3GB memory while the 8800GT has only 512Mb, so I can't enable CUDA in both at the same time, or I will level the VRAM by the lowest value. In the nVidia Control Panel I have set performance to a single monitor, since I have only one attached.

I have set the 580 as the primary card in Windows 7, and the 8800 as the secondary. The monitor, of course, is plugged into the 580. There is nothing plugged into the 8800. In the Octane plugin, I have unchecked the 8800 for CUDA processing, leaving only the 580. This leaves me the 3GB VRAM to work with, which it fine. Everything works fine, except that Windows is still kinda lagging during CUDA rendering. I was expecting more than this considering I had to add a 2nd video card.

Is there something else I should configure (nVidia Control panel?), or am I doing something wrong?

Plug the monitor into the 8800. Also, build 1.13 of the plugins gives you the ability to throttle off the rendering so that Windows remains responsive. Works really well.

Paul

face_off wrote:Plug the monitor into the 8800. Also, build 1.13 of the plugins gives you the ability to throttle off the rendering so that Windows remains responsive. Works really well.

Paul

Oh hi, Paul! Didn't know you were in this forum too. ^^

If I plug the monitor on the 8800, won't I have lousy performance when I play games or use applications that need more video kick like 3DSMAX and Poser?

Ken1171 wrote:

face_off wrote:Plug the monitor into the 8800. Also, build 1.13 of the plugins gives you the ability to throttle off the rendering so that Windows remains responsive. Works really well.

Paul

Oh hi, Paul! Didn't know you were in this forum too. ^^

If I plug the monitor on the 8800, won't I have lousy performance when I play games or use applications that need more video kick like 3DSMAX and Poser?

of course. but it's the only way to make use of the second card at all.
what could help is to connect both cards to the monitor (given it has 2 inputs) so you can switch there. but this has other drawbacks and it is still not guaranteed that switching the inputs will be noticed by the cards. so using the single 580 plus prioritization (soon also in the daz plugin) is you best bet so far...

Oh I see. So the solution would be to degrade CUDA performance so that it doesn't choke Windows. Sounds like none of the available options are quite ideal - either we degrade one thing or the other.

I could simply switch the monitor cable to the other video card when I am rendering, but I suspect those DB9 plugs are NOT hot swappable. I suspect I could damage a monitor or the video card by plugging it in when the computer is on. Could anybody here confirm or deny this?

Ken1171 wrote:Oh I see. So the solution would be to degrade CUDA performance so that it doesn't choke Windows. Sounds like none of the available options are quite ideal - either we degrade one thing or the other.

I could simply switch the monitor cable to the other video card when I am rendering, but I suspect those DB9 plugs are NOT hot swappable. I suspect I could damage a monitor or the video card by plugging it in when the computer is on. Could anybody here confirm or deny this?

well, there are a lot of kvm switches to handle 9pin vga and/or dvi, so this should be no problem; used one myself for a long time to switch between machines. drawbacks came with windows 7 as switching displays might mess up the desktop layout. in the end this is something not really recommended.

btw, the priority switch in octane won't lower rendering performance by margins and it does this not by a fixed amount - octane will still grab as much resources possible, but will allow shared processes to gain higher priority. as the system and hosts ogl viewport usually don't need much gpu power to work much better, you won't see much degrading, and if nothing happens there (i.e. when rendering final), octane will still go with nearly max. available speed. and you can even change this setting on the fly...

t_3 wrote:as the system and hosts ogl viewport usually don't need much gpu power to work much better, you won't see much degrading, and if nothing happens there (i.e. when rendering final), octane will still go with nearly max. available speed. and you can even change this setting on the fly...

Thanks for the feedback! And BTW, I still don't quite understand what the "final" button does. I have tested with and without it, and I couldn't see any differences in performance or rendering times. The same happened with the "smoothing" settings from the 1st tab. If there was any difference in performance, I couldn't see it.

Ken1171 wrote:

t_3 wrote:as the system and hosts ogl viewport usually don't need much gpu power to work much better, you won't see much degrading, and if nothing happens there (i.e. when rendering final), octane will still go with nearly max. available speed. and you can even change this setting on the fly...

Thanks for the feedback! And BTW, I still don't quite understand what the "final" button does. I have tested with and without it, and I couldn't see any differences in performance or rendering times. The same happened with the "smoothing" settings from the 1st tab. If there was any difference in performance, I couldn't see it.

final: you will see a difference as the node count gets higher in a scene, and the material setup gets more complex. try it with a stonemason scene if you happen to have one - usually 30-50 different objects. to provide fast geometry updates, in "normal" rendering each one is a separated geometry node in octane, so the needed geometry updates are very fast; pos/rot are directly executed, things like morphs need additional mesh building, but since done on a per object base, they are still very fast. when using "final" all meshes will be combined into one single mesh. i.e. the city ruins 01 scene renders with ~17ms/s (using auto mats and pmc) in normal mode on my machine (using 2 580s), and 24ms/s in final mode.

smoothing? do you mean the "use base mesh"? it again depends on. a single genesis, one time subdivided, needs around 0.5secs (again on my machine) to rebuild (and this is even needed if you lift on eyebrow only). having it subdivided twice, it'll need ~2secs. using subd level 0 (or base mesh), it'll need about 0.1 seconds to update. you may of course control the speed by changing the subd level, but in crowded scenes ticking one checkbox is of course easier than changing 10 modifiers. and it depends on the machine speed of course; if you have a fast intel quad- or six-core you will barely notice the difference, so you might want to leave it unticked. if a machine is a tad older using a 2-core amd or so, it might still be fine for octane, as you can stuff quite some gpu power into it without the need to build a whole new machine, but geometry updates are a lot slower, so this setting might come in handy...

t_3 wrote:final: you will see a difference as the node count gets higher in a scene, and the material setup gets more complex. try it with a stonemason scene if you happen to have one - usually 30-50 different objects. to provide fast geometry updates, in "normal" rendering each one is a separated geometry node in octane, so the needed geometry updates are very fast; pos/rot are directly executed, things like morphs need additional mesh building, but since done on a per object base, they are still very fast. when using "final" all meshes will be combined into one single mesh. i.e. the city ruins 01 scene renders with ~17ms/s (using auto mats and pmc) in normal mode on my machine (using 2 580s), and 24ms/s in final mode.

Oh I only have a single Genesis on scene, so perhaps that's why I couldn't see any visible differences in performance. I have avoided Stonemason models because my GTX580 maxes out at just 32 grayscale maps, so I have trouble rendering larger scenes.

I noticed controls get locked when I activate "final", so my question is if I can (or should) use it with animations to speed it up too? I know animation rendering is broken at the moment, but that's the first thing I tried here (before I knew it was broken). Once "final" is on, I cannot click on any other tab to switch to animation mode until I disable it.

t_3 wrote:smoothing? do you mean the "use base mesh"? it again depends on. a single genesis, one time subdivided, needs around 0.5secs (again on my machine) to rebuild (and this is even needed if you lift on eyebrow only). having it subdivided twice, it'll need ~2secs. using subd level 0 (or base mesh), it'll need about 0.1 seconds to update. you may of course control the speed by changing the subd level, but in crowded scenes ticking one checkbox is of course easier than changing 10 modifiers. and it depends on the machine speed of course; if you have a fast intel quad- or six-core you will barely notice the difference, so you might want to leave it unticked. if a machine is a tad older using a 2-core amd or so, it might still be fine for octane, as you can stuff quite some gpu power into it without the need to build a whole new machine, but geometry updates are a lot slower, so this setting might come in handy...

Here again, I only have a single figure on stage, so I didn't notice the above. I do have an i7 hyper-threaded quad core with triple channel memory, so I have been leaving the "use base mesh" unchecked and didn't notice any difference in performance. But now I know how to use it.

Ken1171 wrote:I have avoided Stonemason models because my GTX580 maxes out at just 32 grayscale maps, so I have trouble rendering larger scenes.

btw, when creating auto mats, the plugin does some tricks to utilize the available slots - have described it here: http://render.otoy.com/forum/viewtopic. ... 75#p133875 means you should be able to load i.e. soa2 without getting into troubles (but not much more )...