Installing Ubuntu for Part 2 on an XPS 15 9550 notebook


(WG) #1

I have an XPS 15 9550 notebook that came with Windows 10 installed and I’m considering wiping everything and replacing the OS with Linux so I can do Part 2.

My machine currently is configured with 32GB RAM, 750GB SSD, and a GTX970 graphics card.

What would be the best approach for installing Ubuntu and configuring everything for machine learning?

I’m fairly comfortable working with Linux in AWS, but I’ve never actually installed and/or configured the OS and environment locally. Don’t know if there is any best practices or black magic I should know ahead of time or if I’m just overthinking the whole thing.


(nate) #2

Holy crap that’s a nice laptop.

Anyway, I have a dual-booted windows 10/ubuntu laptop I use as my personal laptop right now (but I’ve used Windows maybe once ever for like 5 minutes). If you haven’t had much experience using ubuntu, I would consider dual-booting unless you’re sure you want to wipe windows completely.

In any case, you’ll first want to create a recovery drive of your Windows OS (which you can do within Windows, just google it).

The other main thing you should do is try to find a recent tutorial about installing ubuntu on a windows 10 machine.

Another thing to know about is the bootloader- I’d recommend at least looking into it a bit, it’s basically what launches your OS when you start your machine. Usually when you turn your computer on, the bootloader starts Windows up automatically, but when you install linux, you have to trick your computer into not starting windows and instead starting up into the bootloader’s menu, which can be a pain with newer boxes.

The general gameplan for dual-booting linux is:

  1. burn a ubuntu installation usb (https://tutorials.ubuntu.com/tutorial/tutorial-create-a-usb-stick-on-windows#0)
  2. Clear some disk space using the disk management tool in Windows
  3. get your computer to startup in the bootloader menu instead of Windows
  4. from the bootloader menu, tell it “hey quit booting into Windows by default. Instead, when you startup, boot from my Ubuntu installation USB (or CD)”
  5. reboot the computer (which you can do from the bootloader menu, you don’t have to power off necessarily), at which point it will boot into your installation USB or CD and then the rest is downhill from there

If you’re wiping your whole system clean and starting fresh, then the gameplan is going to be similar once you get into the bootloader.

The only other thing I’d mention is that just be forewarned- linux doesn’t always play nice and sometimes you have to spend an hour here or there figuring out what you think would be a simple problem. For example, on my first ubuntu laptop, the wifi didn’t work so I had to link up with cable to the internet and then download a special driver. The first time I ever installed ubuntu, it took me about a day (I was also pretty careful). Last time I did it (yesterday as a matter of fact), it took me 30 minutes while watching a soccer match because I knew exactly what to do. Stuff doesn’t always work 100% out of the box, so just be aware of that.

That said, I think everyone should be on ubuntu. Since I switched form Windows to ubuntu about 1.5 years ago, I haven’t looked back. Ubuntu is so much better and if you get a working system, you’re going to love it much more than Windows or OSX, at least I did.

Sorry if this is too much too fast, I have no idea how much you already know about the topic


(WG) #3

Good advice … thanks!

Yah, perhaps I’ll dual boot and keep Windows around just in case. I got a 750GB SSD so maybe I’ll just allocate 250GB to Windows and the rest to Linux to create a reasonable deep learning notebook (at least for testing).

I know the notebook I have comes with an option to have Ubuntu installed on it when purchased, so I’ll check with Dell and see if they got the appropriate drivers on their site.


(Eric Perbos-Brinck) #4

I agree with @natedl98: why not try the dual-boot path first ?

I went that way with a desktop PC-Gaming rig from mid-2015, you can see its specs here:

Some tips, based on my own experience, if you go that way:


And I agree -again- with @natedl98: once I experienced Ubuntu, I barely looked back.

I only reboot to Windows for gaming, as the best games (I’m hardcore on Blizzard, was a No-Life multi-boxing on WoW a decade ago ^!^) and their ecosystem (add-ons) only run on Windows.

E.


(WG) #5

Did you or @natedl98 use Jeremy’s bash script to install everything or did you do it all manually?

btw, check my graphics card again and it is only a 960 (not 970). Anyhow, was able to get Ubuntu dual booting on the machine; it was pretty trivial with the tutorials online. I actually had to do the install a few times as my machine kept booting into Windows automatically instead of GRUB. Not sure what I did, but eventually GRUB came up and I was able to get into Ubuntu.

Thanks.


(Eric Perbos-Brinck) #6

I think on my last setup, I edited Jeremy’s inital script to install Python 3.6 over 2.7, and Keras 2.0.

The nice thing with Ubuntu over Windows: if your Deep Learning setup starts getting unstable/weird (it may at some point, especially with Nvidia drivers) , you just say “Screw it, wipe and reinstall” because it’s so fast to do.

Make sure you back-up your notebooks, that’s your knowledge archive. I zip the key ones and Gmail them to myself regularly.


(nate) #7

I didn’t run the script on my own laptop- I’m running it in the cloud on a paperspace machine (which has been my top solution so far that I have found, although I haven’t tried the new and improved floydhub environments they just released like a week ago)


(WG) #8

Paperspace?

Never heard of it.

How do you like it? What kind of GPU do you have access too?

I tried floydhub but never really clicked for me. I just didn’t really enjoy the development experience and it seemed a bit more “black magic” than I wanted.

-wg


(nate) #9

Paperspace is a cloud desktop service (like you can run an ubuntu desktop virtual machine in your browser) that I believe was created for video gamers, so rather than investing thousands of dollars in a gaming computer, you could go to paperspace and spin up a desktop with awesome specs and then play the games on that.

However, they’ve expanded into the ML space now. I am not sure what kind of GPU access you get, but when I tested the various solutions, paperspace was 2x faster than the Amazon AWS setup when it came to running CNNs. Paperspace is also much easier to setup, and for me, cheaper (since I got their $0.65/hr instance whereas AWS is $0.90/hr).

However, paperspace also requires a $5/month flat rate in addition to the hourly billing, as well as a $3/month fee for setting up ssh access to your machine. But like I said, it is much, much easier than AWS and also results in faster training of your models, at least it did when I tested it


(Aditya) #10

Is the dual booting procedure remains same if we begin from a dos based laptop ->install win 10 -> install Ubuntu?

I have a 1 TB HDD.

My doubt is that should I leave the free space(512 GB)while installing windows as unallocated space so that I can install Ubuntu there?

Can someone shed some light in the partitioning part of win and Ubuntu both?

Thanks in Advance…


Building home GPU server
(WG) #11

I don’t see why not.

I think you’re good leaving 512GB unallocated for Ubuntu. See the previous comments for links to tutorials on how to partition your drive to do so.

-wg