"RM 9.1 Crashes when its idle for some time"

varunm1

Hi,

I see that there are crash reports from other members on the community related to 9.1. I am also facing crash issues frequently when working with deep learning. Most of the times its related to Java and Microsoft closes the application. Recently my software started to close when its idle for some time stating windows explorer java issue. I had this both in laptop as well as my PC. But when using 9.0, I never had any software crash issues.

Is there any issue with java or required jdk versions?

Thanks
Varun

hughesfleming68

Hi Varun, could this be related to cuda when using the GPU or is this happening with the CPU as well? I have an issue with LSTM's on 9.1 that I don't on 9.0. At the moment is it not a problem because I am testing DL4J with CNN's which so far are working fine. I have had cuda related problems in tensorflow and pytorch where Windows would seem to lockup. Everything has been stable since going back to using only the CPU. I think people using Linux have a better experience overall.

varunm1

Hi @hughesfleming68. It happened with GPU. Then I thought it might be out of memory. But couple of times it crashed without any process that is what confused me.
 
Thanks
Varun

hughesfleming68

I think it is Windows that is the problem. I even have to use something called Process Lasso to keep everything working smoothly but 9.1 does work differently than 9.0 with the new Deeplearning extension. I would just use the CPU and see if that solves the problem. A few months ago the only way that I could get tensorflow to work on windows with the GPU was to use a non-standard build of tensorflow. That might be fixed now but there was a time that it didn't work at all.

varunm1

Thanks @hughesfleming68 I will look at that.

sgenzer

so be sure to cc @pschlunder as he's is your guy for all DL/GPU questions on our side

Scott