rt gpu pflow

[super gnu]

im doing a relatively simple pflow sim, but using sprite ( camera facing flat) particles,with a vraylightmtl, soft edge opacity map, many particles to give a soft look. nothing else in the scene really, no lights, gi etc.

i was interested how the gpu would handle it, but to my surprise, it renders barely any faster than the production renderer, if not a touch slower (ive not let frames finish to compare accurately.)

this is comparing a titan x to a single 3930k 6 core.

im excluding the pflow prep time from my assesment, talking about the actual rendering.

is this just one of those cases? or is there anything i can do to optimise.?




edit: ok so my test was actually done before i tried using "facing particles" with a soft edge, and before i tried using cache to disk to speed up pflow rendering... 2 things:

a) pflow "cache to disk" works fine in production renderer, but renders blank with rt gpu.

b) with cache disabled, i get a render, but the facing particles render without opacity mapping. i dont know yet if this is due to the particle type, or unsupported opacity mapping in the lightmtl (although it states lightmtl is supported apart from "direct" in the help files)

edit- its the opacity mapping on the lightmtl that is not supported.

[super gnu]

ah , also, seems exclude/include lists for vraydirt dont work either.

wrt the speed issue, i note now that rt gpu always does GI, even if you untick "use gi" in the interface. this is a bit misleading.
seems i can sidestep this issue by setting the gi bounce depth to 0 anyway, so it would be nice if the interface did this internally when you choose "no gi"

on a related note, it seems that setting the secondary to "lightcache" it renders faster than setting the secondary to "none" this suggests that the setting "none" is ignored too and it falls back to brute force.

[Zdravko.Keremidchiev]

is this just one of those cases? or is there anything i can do to optimise.?

Probably the difference is that the GPU always use GI. You can try enabling the GI and compare the render times then.

pflow "cache to disk" works fine in production renderer, but renders blank with rt gpu.

We can not reproduce this in here, would it be possible to attach an example scene with that issue.

its the opacity mapping on the lightmtl that is not supported.

This is in the "to do" list.

seems exclude/include lists for vraydirt dont work either.

I made a note about that.

the setting "none" is ignored too and it falls back to brute force.

Note that RT GPU will render progressively with BF, there is no control over the second bounces.