Wish: DR, but not for LC pass

[Micha_cg]

Hi,

I ask me, could it be possible to disable DR for the LC pass? An option at the DR options like "disable for LC" could be nice.
I observe, that mering and filtering the LC passes is done per one of the four Quad CPUs at the master engine. It need more time, if more LC phases must be combined.
At the moment I render a turntable animation and the LC is quite fast calculated. The LC pass is ready befor the last of the four slave engines starts the LC calculation. So, why start the LC pass at the slaves, if it bring no shorter render time? Merging all the phases need more time than the LC calculation - LC pass 15s and merging 20s.

Could be great, if the slow "merging LC passes" could be speed up. A single CPU usage ... but I suppose so, that could be done by Vlado only.

[dalomar_cg]

Re: Wish: DR, but not for LC pass

Try setting the number of phases to something fewer than the number that the host machine has. LC will only be able to use the number of threads in Phases, thus leaving the slaves idle. I have noticed that for faster LC passes the merging will take a little longer than desired. Thats why I typically only use 3 out of the 4 threads on my big machine for LC passes. I found that that actually gives the quickest result. I do make sure that the phases are set to 4 (or however many the num of threads would be) for the final pass, since the longer LC passes seam less effected by having more threads to merge.

[Micha_cg]

Re: Wish: DR, but not for LC pass

Hmm, the slaves are used allways, independent from the phase setting. It seems to be, that the phase setting will be used for each slave too - at phase 2 all engines run at 50%..75%.
I can't avoid the slave usage. Maybe the slave LC calculation is ignored, if the phases count is not above the CPU count of the master.

I suppose so, only Vlado know what happens.

[dalomar_cg]

Re: Wish: DR, but not for LC pass

Hmm...Looks like I'll have to take some time and look a little more closely at DR. May take me a few days to get set up though...

So at the moment, you're saying that whatever is set for phases is what will be used *at each machine* not as the set amount of phases. In that case, for the time being, just set the phases at 1 and that will use the minimum number of phases per DR render. As long as you're not hooked up to 20 machines (wouldn't we all like that..;D) you shouldn't have too much of an issue.

To be honest, I don't know if this is standard behavior or not...I'll have to check out the max spawner as well.

[vlado]

Re: Wish: DR, but not for LC pass

For the time being, in DR mode each slave calculates its own light cache, with the specified number of passes as in the scene settings.

Best regards,
Vlado

[Micha_cg]

Re: Wish: DR, but not for LC pass

Vlado, if I understand right, so the LC calculation at the slaves dosn't help anything for the image calcualtion at the moment and in the future will be there an optimization for DR?

So, if I understand right, the DR dosn't help for LC, it works like a single machine is used. The number of phases can be the same or less than the master CPU count. Less could be faster some times.

[dalomar_cg]

Re: Wish: DR, but not for LC pass

Yes, it looks like you're on the same page as Vlado...apparently if you've got 4 machines for DR your going to get 4 *extra* LC passes. Vray won't really be able to use any of that information because its redundant anyway. Right now it seams like QMC is better for secondaries when using DR...Its been my experience that QMC for secondaries doesn't take as long as QMC for primaries, but its still not as fast as LC in most cases. For DR though, thats a little less of an issue.