GPU rendering via GTX/Quadro on TX1

Autonomous Machines Jetson & Embedded Systems Jetson TX1
1

I have seen a few people ask about adding extra GPUs via the PCIe bus but the answers seem to be vague and untested. I render with nVidia GPUs currently with Octane and would like to know if it’s at all possible to add GPUs to render from via CUDA. I don’t need them to display just number crunch CUDA style.

Can the PCIe bus be biforcated at all to accomodate multiple GPUs and is there enough PCIe resources to make such a thing possible with GTX9XX and GTX10XX based cards?

Leveraging dual XEONs just to get PCIe lanes is pretty inefficient and I would love to find new, better, solutions for GPU rendering.

2

“Can the PCIe bus be biforcated at all to accomodate multiple GPUs and is there enough PCIe resources to make such a thing possible with GTX9XX and GTX10XX based cards?”

TX1 has two PCIe controller and can be set as x4+x1 or x2+x1, both controllers support Gen1 (2.5GT/s/lane) and Gen2 (5.0GT/s/lane) speeds.