Building a Deep Learning / Cryptocurrency PC (#2): The OS

Among the buzzwords in the tech world of 2017, two tower above the rest: deep learning and cryptocurrencies. It seems that everyone I know (in tech) wants to learn these things. And guess what — so do I! So much so that I'm building my own computer in order to facilitate that learning.

What follows are my notes-to-self as I build a computer to learn about deep learning and cryptocurrency mining. In the previous installment we discussed assembling the hardware. In this installment I'll talk about choosing and installing the OS, configuring it and making sure the hardware is working, and some basic OS security.


The operating system

Last time we left off immediately after assembling our computer. Our computer came without an operating system, so the next step is to choose and install one.

I chose Ubuntu Desktop. Lots of the tutorials I see for deep learning are written for Ubuntu, and I'm comfortable with Linux. Windows also seems to have plenty of support but I'm not nearly as familiar with the MS ecosystem.

I don't think it matters a ton which OS you choose; most of what you'll be using it for is platform-agnostic.

The big thing to keep an eye out for is CUDA support. Luckily, most of the operating systems (Windows, Linux and Mac) are supported. So don't go choosing some crazy new-fangled OS nobody's heard of!

So, the conclusion: Pick a Linux distro, Ubuntu unless you're familiar with Linux and comfortable with the command line, or pick Windows and go find yourself another tutorial.

Installing

Once you've chosen your OS, the first step is to burn an image of the distro to a USB stick. Ubuntu provides a very straightforward tutorial. Once you've got that, you'll restart your computer and reboot from USB. Instructions on doing this differ depending on the motherboard so read your manual; for me, I had to restart holding down F10.

You then will run through the OS installation, of which plenty of guides exist on the internet.

Now at some point, this being a computer you assembled from scratch, it's likely you'll run into problems. For me, I had a hell of a time getting my display to come on and stay on. I'm still not 100% sure why. Assuming your hardware differs from mine, your problems are likely to be unique to you. Don't panic. Google is your friend, and it's likely that any problem you run into, others have run into too.

The problems I struggled with:

Just keep at it, and don't hesitate to ask the community if you're wrestling with a particularly thorny issue. They're plenty friendly, so long as you've done your homework beforehand!

System Diagnosis

Now we're booting into Linux and we've got the basics installed. As my next step, I want to check that the hardware is all running correctly and there's nothing out of the ordinary. To check that, I learned about a few good tools on Linux. You may have to install a few of these, if something is missing try installing it with:

sudo apt-get install xxxxx

System-wide commands

lshw is used to list hardware. This will dump out a ton of information in your system.

inxi is another good one. Here's my inxi -Fx output:

inxi -Fx

Let's step through component by component and run a sanity check to make sure everything is working.

CPU

There's a few commands you can use to get more information on your CPU:

cat /proc/cpuinfoprints out information on your CPU. Here's a good Stack Overflow thread going into more detail on what this output means.

cat /proc/cpuinfo

lshw will list out all hardware, and the CPU section has some relevant information:

lshw

lscpu will give you some more CPU-specific information, including MHz and cache information:

lscpu

These commands are great if you're familiar with the ins and outs of hardware, but I barely have a clue what I'm looking at. I just want to know if the CPU is working or not.

For that, top and htop will do the trick.

top:

Output from top

And to get more graphical, use htop:

Output from htop

This article does a pretty deep dive into htop. To ensure I'm getting 100% throughput on each of the cores, this article at peteris.rocks does a good job explaining how:

If you run cat /dev/urandom > /dev/null which repeatedly generates random bytes and writes them to a special file that is never read from, you will see that there are now two running process.

Running this gets me to 100.0% on all cores. I'm not 100% sure this is a valid test of the CPU but it seems good enough for me.

Hard Drive

Running lsblk prints information about hard drive partitions and other storage devices. Running that gives me:

lsblk

Which is roughly the size I expect. df is another one that gives you information about hard drive space:

df

RAM

free -m will check the amount of RAM:

free -m

That doesn't look so good — I expected 16 GBs! Let's check lshw:

lshw

This tells me I installed one of the RAM chips wrong. Time to break out the screwdriver again.

GPUs

lspci will give you information on PCI buses and devices in your system, including your GPUs:

lspci

Another command you can use is nvidia-smi:

nvidia-smi

This gives you information like fan usage, temperature, memory-usage, and more goodies.

Cooling

Check CPU temperature with sensors :

sensors

Between this and nvidia-smi above you should be able to get a good idea of whether your cooling supply is adequate or not.

Power

Is the system getting enough power? Short of measuring the power between the computer and the electrical, I don't think there's a 100% software solution to find this out.

powertop is a utility that measures consumption. It seems like it's mostly useful for laptops, but there's some good info for desktops as Wayne points out here.

Ultimately, making sure you have adequate power needs is something you should do before assembling your system, either by checking pcpartpicker.com when choosing your parts or checking something like this link (thanks Sean!). And if you can afford it leave yourself plenty of breathing room in the watts department.

Security

Having satisfied my worrywart side that everything is running smoothly, the next step is to make sure we've configured our system following standard Linux security procedures.

I'm going to be using this machine to practice deep learning (not too worried about someone hacking this) and for mining cryptocurrencies (more concerned about something hacking this!)

A good place to start is the Basic Security guide hosted on Ubuntu Wiki:

I assume you're familiar with the general Linux security procedures — strong passwords, limited root access, that sort of thing. If not, spend some time catching up on that.

Automatic Security Updates

I'm lazy as sin and if there's one thing I hate it's manual security updates. Luckily Ubuntu provides an automatic update mechanism:

I followed the “unattended-upgrades” package and it was a straightforward process.

This will likely cause issues if I'm running deep learning simulations overnight, but I can tackle that issue in the future. For now I'm erring on the side of caution.

Firewall

A firewall will allow you control over which ports are accessible publicly. Generally you want the most restrictive policy you can stomach.

Here's the Ubuntu wiki for firewalls (if you're noticing a trend, it's that the Ubuntu wiki has some pretty darn good information!):

I'm going to use ufw because I'm on the terminal. I went with the recommendations the guide laid out — DHCP, web and mail access — but left out the torrent ports since I don't need them.

This is an absolutely massive security guide by bodhi.zazen. It's well worth a read and implementation.


If you want to hear about how I tackled the crypto mining and deep learning setups, drop your email below and I'll let you know when I publish those sections.



Kevin Scott

After ten years of experience building apps for brands like Venmo and GE Healthcare, I'm building bleeding-edge AI software. Join me as I go through the latest in machine learning, web development, and design.

Tensorflow.js and Machine Learning with Javascript