Last post was over a year ago, just curious if anyone is still using TK1 boards. I got a TK1 cheap, would like to learn cuda even if it is an older version. Wondering what kind of trouble Id run into with older jetpack and cuda
I bought one last month to try to use as a cheap SBC that has sata / mini pcie ports. Here’s what I have learnt so far:
Graphics drivers work up to Ubuntu 16.04 - so you can do one level of sudo full-upgrade. Make sure to lock the x.org file before doing this (look up for guides on upgrading to 16.04 via google or in this forum)
You can upgrade to Ubuntu 20.04 (which I have done) BUT you will lose functionality of your GPU, which basically makes it not worth your while to tinker with unless you know your way around compiling and using old drivers in Linux.
The TK1 will look to boot from the SD card slot first before using your internal flash storage. You should install a copy of the filesystem on an SD card and boot from it there first. This way you can tinker with whatever you like and just reboot afresh if you mess something up (wish I knew that before I tried updating all the way to Ubuntu 20.04).
The LAN is flaky - if your connection drops out, you may have to restart all your network services. Not sure exactly why it happens but it happened to me during times of high utilisation (basically when doing full upgrades)
My main goal is enabling the ASM1061 mini pcie card on it, but while it does detect the card, it does not detect any drive attached to it
I haven’t tried using Jetpack (since the version required apparently doesn’t work on newer Ubuntu versions) but again, if you unpack all the directories onto a flash card, and don’t damage whichever partition the bootloader (on the internal flash) is on, you can always reformat and recopy the folders on to the external flash card to start afresh
Were you able to run either pytorch or tensorflow on the GPU in your experiments?