You are not logged in.
I'm trying to get Blender to recognize my GPU (GTX 760) as a compute device. I have both the packages "cuda" and "opencl-nvidia" installed. I'm using the Nouveau driver right now.
I've tried installing the proprietary Nvidia driver, but I also have the "xf86-video-intel" driver installed. Ideally, I want to use the CPU-integrated graphics on my Intel i5 to display my desktop, and save the GPU just for use with Blender. When I tried to install the Nvidia proprietary driver, "nvidia-libgl" conflicted with "mesa-libgl". I issued "pacman -Rdd mesa-libgl" and proceeded with the installation. However, without "mesa-libgl", Blender no longer recognized my Intel graphics as a OpenGL capable device.
Is it possible to install both drivers at the same time? If not, could I enable GPU rendering with Nouvaeu? If so, how much of a performance hit would I experience with Cycles between CUDA with Nouveau vs with the proprietary Nvidia driver?
Offline
You need to install and configure bumblebee.
Offline
Wow. That worked really well.
Unfortunately, the BMW test scene renders at 1:22 on my Arch install, but only 0:54 on Windows. I tested Blender on Linux through optirun, primusrun, and also without either. All of those tests came out within a reasonable margin of error of each other, so I'm guessing it's a driver issue rather than a Bumblebee issue. Is there anything I can do to make up the difference?
Offline
FWIW, I noticed on one of my scenes that Windows using CUDA rendered ~10sec faster than Linux. It's the first time Windows beat Linux on a rendering. I have no clue what's going on with it.
Matt
"It is very difficult to educate the educated."
Offline