Tweet
Hi Vlado,
I've been testing the GPU beta with a GTX 480, for the last couple of days. I wanted to see how it worked with bigger data sets and using it for lighting stuff interactively.
I noticed that the "SDTree" is rebuilt whenever you adjust a light source intensity or moving it, aswell as shuffling all the geometric data over to the graphics card when doing these kind of adjustments.
Since transferring that data for bigger scenens take even longer time on the GPU compared to the CPU version of RT(probably due to the system architecture in the computer I guess...?), but would it be possible to make the process more efficient when adjusting lightsources?
I'dont know, but I guess rebuilding the SDTree is necessary when moving a light source due to some optimization in Vray RT, or?
But "re"-sending all the geometry and rebuilding the SDTree when adjusting the intensity of a light source isn't, right?
At the moment it isn't very interactive when you have to wait 25 seconds before it updates when adjusting a light sources...
These RT GPU scenes eat up almost all of the GPU memory of the 1,5 gigs available on the GTX 480, mostly due to the polygon count. And they are spread out over 450 objects, I guess that's slowing down the SDTree compilation.
I'dont want to complain, Vray-RT has already saved us a lot of time and made our workflow better when working with smaller isolated parts of scenes. Just wondering if there are room for improvement for these kind of scenarios.
I've been testing the GPU beta with a GTX 480, for the last couple of days. I wanted to see how it worked with bigger data sets and using it for lighting stuff interactively.
I noticed that the "SDTree" is rebuilt whenever you adjust a light source intensity or moving it, aswell as shuffling all the geometric data over to the graphics card when doing these kind of adjustments.
Since transferring that data for bigger scenens take even longer time on the GPU compared to the CPU version of RT(probably due to the system architecture in the computer I guess...?), but would it be possible to make the process more efficient when adjusting lightsources?
I'dont know, but I guess rebuilding the SDTree is necessary when moving a light source due to some optimization in Vray RT, or?
But "re"-sending all the geometry and rebuilding the SDTree when adjusting the intensity of a light source isn't, right?
At the moment it isn't very interactive when you have to wait 25 seconds before it updates when adjusting a light sources...
These RT GPU scenes eat up almost all of the GPU memory of the 1,5 gigs available on the GTX 480, mostly due to the polygon count. And they are spread out over 450 objects, I guess that's slowing down the SDTree compilation.
I'dont want to complain, Vray-RT has already saved us a lot of time and made our workflow better when working with smaller isolated parts of scenes. Just wondering if there are room for improvement for these kind of scenarios.
Comment