RT problem / Building OpenCL trace program taking forever

[muoto]

Hello,

on max2016, vray 3.2, when using RT on my quadro 4000, on a simple teapot scene, the builing "openCL trace program" is forever and ever...(like 172 seconds...)

Is this normal ?

Thanks

[super gnu]

yes.. it will take sometimes 1/2 hr + usually less. it will also eat a massive amount of ram..

however, it only has to do this once (until you update gfx drivers unless im mistaken)

any reason you are not using Cuda? it doesnt have this behaviour.

its the recommended choice for nvidia cards.

[savage309]

Use CUDA, it will be much faster on nVidia GPUs, has more features and it is more stable.

However if you have to use OpenCL - the time needed does not depend on us. We do have valid OpenCL kernel and it is up to the OpenCL implementation vendors on how (and if) the kernel is going to compile. With nVidia GPUs it takes a few minutes with the latest driver, Intel CPU and AMD CPU implementations usually manage to compile it in less than 10 seconds. AMD GPUs also are using a few minutes as well, however the chance for them to fail to compile the OpenCL kernel is huge. If it takes more than that, you should either update your driver (check here http://docs.chaosgroup.com/display/VRAY3/GPU+Rendering#) or fire up a bug to the OpenCL vendor.
The good thing is that the compiled kernel is cached, so the next time you start RT GPU it will start up instantly.
There are stuff that we can do to make it easier for the vendors in theory, and we are working on them.