I’m as excited about VrayRT on GPUs as anyone, and I’m aware there are still unsupported features, but what I found today is pretty glaring. I set up a test scene to investigate and what I found is that the only thing that VrayRT will recognize in a Mat’s diffuse or reflect slot is a color or a bitmap. It renders as black when using falloff, Mix, or any combination of the above. I read that VrayBlend was unsupported but never anything about good old falloff and mix.
Is this a know issue and if so is there a workaround?
Are falloff an mix something inherently impossible with OpenCL or is this simply a temporary situation until the software matures?
Like the blend map, both Falloff and Mix maps use multiple maps which requires a pre-render set-up time which can slow down the GPU considerably. Hopefully we’ll soon get a build that will use the first GPU render to cache a bitmap that will be used for subsequent GPU renderings using these maps.
Oh yeah, you asked about a workaround. Don’t know if it would work in your particular case (especially the Falloff), but when using some of these complex maps with RT/GPU I’ve been rendering out flat-shaded bitmaps using the standard renderer(s), and then applying that map back onto the object using a basic Vray material. I’ve gotten it to work pretty good in many cases…
Unfortunately my stainless steel shader is very dependent on falloff - which as you hinted at can’t be baked in very readily. But since I only use RT/GPU for pre-viz, I think the most sensible thing to do is to bypass the fancy falloff/mix map portion of the shader and settle for a straight bitmap in the reflect slot - that gets me to 85%. Then, rewire it back to how it was for final rendering.
I think resolving this blend map pre-render issue is by far the most important thing Chaos Group could do to improve RT/GPU.
Haha, tiles is the first thing i noticed that was missing! I rarely use that map so i didn’t think too much of it.
I think all of these things stem from the same issue that Alan enlightened me to: pre-flight computation of mutliple map operators. I really have no idea what’s involved in getting those to compute in OpenCL or what’s involved in having the CPU do part of the work and the GPU the rest. But I would trade a performance hit for being able to use basic operators like falloff/blend/mix/tiles and more.
