JetPack 3.2 issue with DIGITS

Hey there,

I used to train Caffe models using DIGITS with my Jetson TX1 running with:
JetPack 3.0
Caffe 0.15.14
DIGITS 6.1.0

I had to work with quite low batch sizes to make it run, sometimes even with “1”, as otherwise the error with the code -9 occurred before the training started.

But now as I installed JetPack 3.2, even with batch size “1” I’m constantly receiving this error. I tried to use DIGITS 6.1.1 (using the archive) and then simply the master branch - none of them helped.

Any clue what could cause this issue?

Here’s the output log from Caffe:

https://github.com/NVIDIA/DIGITS/files/1924265/caffe_output.log

I posted this as well in the git repository of DIGITS, but as this error is describing a lack of memory, which doesn’t exist, I thought it could be an issue beyond DIGITS, which is why I post this one here, too.

Cheers,
Moritz

Hi Moritz, DIGITS isn’t officially supported to run on Jetson TX1/TX2, sorry about that.

There are some anecdotal reports on the forum from TX2 users who were able to get it running, presumably since TX2 has 8GB memory.