Boot Loops

CUDA, it seems, is much more complex than sudo apt-get install. You have to use nVidia closed-source/binary-blob/call-it-what-you-want/non-open-source driver package. It's supposed to be specifically designed for CUDA.

Okay, well, fine, I'm not picky, whatever makes it work. 
Turns out CUDA documentation and drivers are....involved.

Really really involved.
If you want, take a look here
http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

Those of you who read that, you will see the magic words "sudo apt-get install cuda" and think, "Ah ha! This guy is wrong, it really is that easy!"

No, no it's not. Note all the prior stuff you need to install first, and some of it isn't trivial, and if you don't do it in exactly the way the documentation says, it will send you into an endless boot loop from login.

Which means, time to start over.
Well, for the really advanced Linux user, I'm sure there is a way to recover from that, but despite my searching, dropping into CLI-only mode, and reading many many pages, I couldn't recover.

So, back to square one, wipe and re-install. Didn't seem to matter which order I installed CUDA in either. I tried with a fully patched system, and then CUDA (boot loop), I tried with a naked fresh system (ditto) I tried patched and Karpathy's Char-RNN install (same) and on

and on
and on
and on...

Frustrated, I read more of Char-RNN, and discovered it would also work with an ATI card.

But did it? Read on to find out.

Comments

Post a Comment

Popular posts from this blog

Introduction

Greetings and Salutations, This journal is going to document my experiments with Recurrent Neural Networks (aka Machine Intelligence) and the rather long way we have to go with this field. A quick overview on RNNs can be found here https://karpathy.github.io/2015/05/21/rnn-effectiveness/ and here https://medium.com/@camrongodbout/recurrent-neural-networks-for-beginners-7aca4e933b82 These are reasonably high level topics, but to break it down it is a program that can create new connections (data, sound, images) from a given set of data. MI goes by a number of different terms these days, Deep Learning, Recurrent Neural Networks (which is the term I use) and of course, the overused and incorrect Artificial Intelligence. I will be specifically talking about my experiences with Karpathy RNN, which can be found here https://github.com/karpathy/char-rnn Ok, that's all for this post, keep reading for the trials and tribulations.
Read more

Fun with Distros

Right, so, Ubuntu 16.04 LTS. Seems simple right? Well the basic install was a walk in the park, drop the CD in, and let it go to town on the PC. Then came the first set of instructions (taken from the Karpathy site I linked to earlier) $ curl -s https://raw.githubusercontent.com/torch/ezinstall/master/install-deps | bash $ git clone https://github.com/torch/distro.git ~ /torch --recursive $ cd ~ /torch ; $ ./install.sh # and enter "yes" at the end to modify your bashrc $ source ~ /.bashrc So right away, problems arise. Curl it seems, wasn't installed by default on 16.04. Ok, no issues, I'll just sudo apt-get install curl and go to town. Yeah, errors get thrown. Scrolling through them I realize there are more dependencies required. Sigh. Ok, so make note, do a sudo apt-get install of each dependency and move on to the next step. Oh, one of the important things from the Karpathy site? Yeah, download the actual program and unpack it. That hel...
Read more

How did this begin?

I read a tech site called "Ars Technica" and one day they ran an article about an RNN creating names for paint colours. You can read about it here https://arstechnica.com/information-technology/2017/05/an-ai-invented-a-bunch-of-new-paint-colors-that-are-hilariously-wrong/ and I was instantly hooked on the idea of creating my own Neural Network. So I followed the link from Ars, back to the original  site, which is here http://lewisandquark.tumblr.com/ and started reading from the beginning. The lovely and talented Ms. Shane linked to the Karpathy site where the program could be obtained from. Seemed simple enough, it need a Linux OS, and I was already familiar with Linux, so this would be a walk in the park. Right? Oh so very wrong... Read on.
Read more