Adding CUDA cores via GTX 1030

Seems like a fiscally cheap way to increase CUDA count the 1030 also happens to natively use pcie x4 (even though it’s pinned for 16x), are there any significant hurdles to adding extra cores into the mix?

I don’t know on the software side, but for any external power connection I would advise use an external power supply…don’t supply any auxiliary power on the card with power from the Jetson itself. The power requirements of even a “low power” video card would completely dwarf the power requirements of a Jetson running all by itself.

One speculation on software…the existing video driver (which is used by CUDA/GPU) is for this particular hardware…the GPU is directly connected to the memory controller and is not PCIe. This means you’d have to get a second video driver for video on PCIe…however, the NVIDIA-supplied hardware-accelerated drivers normally found for a PC won’t work…they are not aarch64/arm64. If there is a flaw in using a PCIe video card, then this is probably it…finding a PCIe-based driver under aarch64/arm64 since Nouveau drivers do not access the GPU directly.

Linuxdev is correct, the GPU drivers included in JetPack for Jetson are userspace drivers for integrated GPU, not the kernel PCIe driver you would typically see on x86 systems with discrete GPU. This is not officially supported at this time.

I had assumed arm would be an issue, was hoping some work had been made on drivers at this point. NBD.

What about a tv tuner card? Really interested in using some of the visionworks demos on some ATSC feeds.

Sure it’s possible, you would need to make sure the video input card had a Linux-compliant driver available that preferably had been built for ARM relatively recently.