Low Priority Main GPU

[Moriah]

Is there any way to put the main GPU (used for display) on low priority? So we can still use it for rendering but it doesn't lag like crazy in every operation... Right now i have to switch it off for rendering. Other GPU renderers have this feature, would be nice to have in Vray too.

[benb32]

Suppose you can always try and reduce the ray bundle size - know it's not the same as changing the priority, but it will reduce the lag with the small price of reducing the rendering throughput.

[savage309]

Suppose you can always try and reduce the ray bundle size - know it's not the same as changing the priority, but it will reduce the lag with the small price of reducing the rendering throughput.

Actually, it is the same thing.
The best thing one can do is to reduce the chunks of work each GPU gets.
We call it "reduce bundle size and rays per pixel", other probably are calling it "lower priority". It is not lower priority however.

Best,
Blago.

[Moriah]

Actually, it is the same thing.
The best thing one can do is to reduce the chunks of work each GPU gets.
We call it "reduce bundle size and rays per pixel", other probably are calling it "lower priority". It is not lower priority however.

Best,
Blago.

What's a good value for those? I just tried bundle size = 32 and rays = 4 and it's still very laggy... I have 3x 980 Ti

[Raph4]

It's the reason why, in certain circumstance, I prefer using OpenCL for keeps one GPU free and I use my Xeon instead.

[savage309]

What's a good value for those? I just tried bundle size = 32 and rays = 4 and it's still very laggy... I have 3x 980 Ti

Try 1 ray per pixel and 32 bundle size.
We are using all CPU threads to feed those GPUs and probably we need to add option for fewer CPU threads to do that.

Anyway, if you have 3 GPUs, I think it actually might be faster to use only 2 (but fully utilized) for rendering and leave one for monitor, instead of having 3 not-well utilized.

Best,
Blago.

[Nicinus]

Any news on using less priority?

It just feel like such a waste to have something like two Titans just for the RT part, and then maybe a 1070 for the viewport. I know that if I did this and then used them all for RT, the memory availability would be capped at the amount of RAM of the smallest card, but how would the speed of the cards affect this? Would each card compute at its own pace or are they all synchronous?

[Raph4]

Any news on using less priority?

It just feel like such a waste to have something like two Titans just for the RT part, and then maybe a 1070 for the viewport. I know that if I did this and then used them all for RT, the memory availability would be capped at the amount of RAM of the smallest card, but how would the speed of the cards affect this? Would each card compute at its own pace or are they all synchronous?

yes, on nightly 3.45.01

[savage309]

yes, on nightly 3.45.01

[ATTACH=CONFIG]32671[/ATTACH]

Still work in progress though. There are some improvements that I hope we would manage to implement before the SP ...

Best,
Blago.

[Nicinus]

Great!

But how would this work if I for example have a 780 and then adds a Titan X to it. Would the 780 somehow determine maximum speed in the same the way it would limit how much memory is accessible if I want to use both cards in low priority mode?

[savage309]

Great!

But how would this work if I for example have a 780 and then adds a Titan X to it. Would the 780 somehow determine maximum speed in the same the way it would limit how much memory is accessible if I want to use both cards in low priority mode?

I still highly suggest to have a separate, slower GPU for display only. I see this option for people that want to do GPU rendering and have one GPU only (thus in the pre-viz stage that get sluggish OS UI).
Otherwise - when Low Thread Priority option is "on" we utilize the GPUs with attached displays less and we try to feed the other one as usual. The problems we get are that sometimes this can actually slow down the whole render process, because of some internal waits that occur, so we will see.

Best,
Blago.

[Nicinus]

I hope you can get it to work, there is something inherently frustrating with not being able to use all your fire power not to mention the irony of having for example dual 1080s or Titans and then driving all your viewport graphics with a 780. :/

[maxwood]

I was just about to post regarding this! Happy to see it will be coming soon, I'm running a 980 Ti but it's a struggle to use for look dev because of the UI sluggishness.

Might this be available in 3.5?

[savage309]

I was just about to post regarding this! Happy to see it will be coming soon, I'm running a 980 Ti but it's a struggle to use for look dev because of the UI sluggishness.

Might this be available in 3.5?

It is in the 3.5 Beta now, will be in the 3.5 final as well. Feedback will be helpful.

Best,
Blago.

[eyepiz]

It is in the 3.5 Beta now, will be in the 3.5 final as well. Feedback will be helpful.

Best,
Blago.

Does the low priority affect all gpu's? I enabled it and it didn't seen to make much of a difference compared to when I had the main off.