5 ms·

And why wouldn’t it? CS education is rarely taxing in terms of processor/memory requirements. I doubt the curriculum is all that different than 10 years ago or

And why wouldn’t it? CS education is rarely taxing in terms of processor/memory requirements. I doubt the curriculum is all that different than 10 years ago or more with far less specs. It’s about theory and writing algorithms. And if this is a program specializing in ML, the models are still likely to be toy, and anything more can use cloud computing like Google’s Collab.

ThinkBeat 5y ago
I must politely disagree with that. It may be that we have been to distinct types of CS programs. Some of my work as an under grad was taxing on my computer and much more so as a grad. Even simple things like compiling a system can quickly get taxing and slow. Some classes require models, algorithms that takes a lot of power to process or to prove. My master thesis was system would run for 48 hours on the desktop I had then, after some optimization got it down from 65 hours. Ideally though you have access to a lab / server that you can offload your hard work on. In my experience those resources were already under maximum load during most of the day and parts of the night. Being able to just run them locally was convenient.
- mbreese 5y ago
  > Ideally though you have access to a lab / server that you can offload your hard work on. And I think this is the key point here. Most CS things you can easily handle with the horse power of a Chromebook. And when you hit the 10% of the coursework that you can’t use a Chromebook for, you probably wouldn’t use a “normal” laptop for it either. You’re always going to have jobs that will take a bit more horsepower. Those should be farmed out to a cluster or server anyway. Where I think the OP will have issues is what resources they will need for their research, which again will probably need a server or cluster. Likely, the most taxing thing they will likely need to work on locally is their thesis. This is where I’d think they might run into issues, especially with figures. But even here, Overleaf should be up for the job.
- codeisawesome 5y ago
  Not to mention being on video calls with classmates / actual classes and also having a few other windows running (lets say Slack / Discord), and maybe driving a decent-resolution monitor.. these relatively basic sounding things can push the CPU/gfx memory/RAM of even higher-end hardware to their limits! Especially if you choose to / need to run all these together and multi-task. That might partially be an indictment of software performance, though.
gtm1260 5y ago
All my main CS coursework was done while SSH'ed into a class cluster, but I took a bunch of electives where you're like working w/ docker, doing android development, etc. where its nice to have a fast machine.
PNWChris 5y ago
Agreed. While doing my undergrad in CS a decade ago (wow - I can't believe it's been that long) a lot of coursework was relatively light-weight to execute, and when it wasn't it could be done by SSH-ing or VNC-ing into a university-managed RHEL environment. We could even RDP into well-equipped Windows machines that had licenses for all the software you could ever want if you happened to need Windows. This was at a large Midwestern research university. From talking with friends and colleagues over the years, it sounds like it was a pretty standard set-up. I did try to do local development on an ARM Chromebook I got as a curiosity back then, but it was just too constrained. Cruton wasn't perfect and only the XFCE desktop environment had acceptable performance. Plus, not a lot of software was ARM compatible out of the box back then and compiling from source was a pain on the low-power processor I had. These days, especially with the crazy fast ARM cores we're seeing and Android Apps working on ChromeOS, things look way more usable for a CS student's general computing needs. It's always cool to see stories like OP's to confirm what I hoped was the state of computing outside the Mac/Windows norm. Disclosure since this is about ChromeOS: I work at Google, but this is all my personal experience and opinion.
- edgyquant 5y ago
  I went to a community college and we had the ability to SSH into ridiculously powerful servers on campus + were given Azure credits, etc. I think it’s pretty standard
chrisseaton 5y ago
> CS education is rarely taxing in terms of processor/memory requirements Not sure I'd enjoy doing something like compiling LLVM on a Chromebook for a compiler class.
- thiht 5y ago
  Not sure what compiling LLVM has to do with a compiler class though. Yes it's a compiler, but compiling it won't give lots of insights on how a compiler works or how to write one?
  - chrisseaton 5y ago
    > Not sure what compiling LLVM has to do with a compiler class though. Not normal to work on a compiler during a compiler class?
    - thiht 5y ago
      On toy compilers to present the basics sure. On LLVM, maybe that's a bit much? Reading the code, why not, but compiling it? Don't see the point in a compiler course. I can't see what you could learn about compilation in a CS sense by compiling LLVM.
      - chrisseaton 5y ago
        > Reading the code, why not, but compiling it? Compiling it with changes you made to learn about how compilers work. Such as adding a new optimisation phase. > I can't see what you could learn about compilation in a CS sense by compiling LLVM. You don't learn by compiling it obviously. You learn by extending it and running it, and to do that you need to compile it.
rybosworld 5y ago
I think this is mostly right. I'm currently enrolled in same the GaTech CS Masters that the author will be starting. I have had just two courses that strained my machine out of the 7 I've completed. Reinforcement learning in particular had one project that really tested my machine. (I trained my agents on CPU, so that was definitely a factor). Even then, my agents were training in fairly reasonable times (5-6 hours max for a grid search). If I'd taken the time to vectorize every computation, I could have probably gotten the time down at least 10x. This is on AMD Ryzen 1700x - a pretty speedy chip but not top of the line - and 16 gb ram. Powerful hardware is a convenience more than anything in this program. It allows you to focus less on optimizing your solution.
- ylyn 5y ago
  A 1700X is quite a bit more powerful than the i5-8250U in the author's Chromebook. Probably around 2–3x the overall speed.
- sand500 5y ago
  Do a HPC course, the whole point is to tax your system.
  - rybosworld 5y ago
    I've heard it described as the best course in the program. I won't have room to take it unfortunately. I went full ML spec and my last two courses will be Video Game Design (Been dying to take that) and Graduate Algorithms (Required)
colinmhayes 5y ago
Compiling and training ML models can be pretty taxing. Hopefully OP can just ssh into Gtech's cluster or else I could see it being annoying.
x0 5y ago
During my B.CS, most power I needed was whilst doing Android and iOS development, and even then that's just for the VMs. Otherwise did it all on a beat-up 2014 MacBook Air. Specific programs we used like SPIM, the pan/spin verification thingo, gcc/g++/etc, you can run wherever you like. EDIT: Uni provided computers with VM stuff installed, and let us remote into them during COVID.
josefx 5y ago
I once bought an Eee PC for note taking, it was great execpt for two things: * The low end graphics chip was already 12 years old on release, the advertised OpenGL 2.0 was emulated in software while everything else that came out supported OpenGL 3. * Intel didn't do virtualization on low and mid range variants of its CPUs, while an Eee PC barely had any resources to it was the second time I got burned by that. Basically 10 year old requirements meet technology that was outdated 12 years ago. Even toy samples will fail if the hardware doesn't implement the features required.
xvector 5y ago
For my computer graphics class, being able to break out my overclocked and watercooled 20-thread behemoth in an era where most processors had far fewer, definitely helped me procrastinate a bit longer, because my renders took a fraction of the expected time :) The same machine’s GPUs helped me be the first to solve a password brute-force challenge in my computer security class (designed to teach us how to use password cracking tools), though that wasn’t for any credit, of course! For my machine learning class, I was able to train models much more quickly than other students. Generally speaking, being able to brute force certain things was just fun. But yeah, I would agree that there is no practical benefit to a powerful machine in a CS education. These few instances are not worth the cost - buy or build a powerful rig because you want to, not because you think it’ll help. I felt immense pride at watching my system whir to life.
- seventhtiger 5y ago
  Yeah, I used my machine to do a ton of NLP on relatively big datasets, and I had cool realtime 3D visualizations to show off. I lugged my full tower to class for that one. It's really nice to have something powerful to play with.
dijit 5y ago
Had a friend who needed to borrow my beast machine to run “word2vec”. Took a couple weeks on his machine, took 30 minutes on mine (mostly because of memory). Though he was studying AI and not compsci

ThinkBeat 5y ago

I must politely disagree with that. It may be that we have been to distinct types of CS programs. Some of my work as an under grad was taxing on my computer and much more so as a grad. Even simple things like compiling a system can quickly get taxing and slow. Some classes require models, algorithms that takes a lot of power to process or to prove. My master thesis was system would run for 48 hours on the desktop I had then, after some optimization got it down from 65 hours. Ideally though you have access to a lab / server that you can offload your hard work on. In my experience those resources were already under maximum load during most of the day and parts of the night. Being able to just run them locally was convenient.

mbreese 5y ago
> Ideally though you have access to a lab / server that you can offload your hard work on. And I think this is the key point here. Most CS things you can easily handle with the horse power of a Chromebook. And when you hit the 10% of the coursework that you can’t use a Chromebook for, you probably wouldn’t use a “normal” laptop for it either. You’re always going to have jobs that will take a bit more horsepower. Those should be farmed out to a cluster or server anyway. Where I think the OP will have issues is what resources they will need for their research, which again will probably need a server or cluster. Likely, the most taxing thing they will likely need to work on locally is their thesis. This is where I’d think they might run into issues, especially with figures. But even here, Overleaf should be up for the job.
codeisawesome 5y ago
Not to mention being on video calls with classmates / actual classes and also having a few other windows running (lets say Slack / Discord), and maybe driving a decent-resolution monitor.. these relatively basic sounding things can push the CPU/gfx memory/RAM of even higher-end hardware to their limits! Especially if you choose to / need to run all these together and multi-task. That might partially be an indictment of software performance, though.

gtm1260 5y ago

All my main CS coursework was done while SSH'ed into a class cluster, but I took a bunch of electives where you're like working w/ docker, doing android development, etc. where its nice to have a fast machine.

PNWChris 5y ago

Agreed. While doing my undergrad in CS a decade ago (wow - I can't believe it's been that long) a lot of coursework was relatively light-weight to execute, and when it wasn't it could be done by SSH-ing or VNC-ing into a university-managed RHEL environment. We could even RDP into well-equipped Windows machines that had licenses for all the software you could ever want if you happened to need Windows. This was at a large Midwestern research university. From talking with friends and colleagues over the years, it sounds like it was a pretty standard set-up. I did try to do local development on an ARM Chromebook I got as a curiosity back then, but it was just too constrained. Cruton wasn't perfect and only the XFCE desktop environment had acceptable performance. Plus, not a lot of software was ARM compatible out of the box back then and compiling from source was a pain on the low-power processor I had. These days, especially with the crazy fast ARM cores we're seeing and Android Apps working on ChromeOS, things look way more usable for a CS student's general computing needs. It's always cool to see stories like OP's to confirm what I hoped was the state of computing outside the Mac/Windows norm. Disclosure since this is about ChromeOS: I work at Google, but this is all my personal experience and opinion.

edgyquant 5y ago
I went to a community college and we had the ability to SSH into ridiculously powerful servers on campus + were given Azure credits, etc. I think it’s pretty standard

chrisseaton 5y ago

> CS education is rarely taxing in terms of processor/memory requirements Not sure I'd enjoy doing something like compiling LLVM on a Chromebook for a compiler class.

thiht 5y ago
Not sure what compiling LLVM has to do with a compiler class though. Yes it's a compiler, but compiling it won't give lots of insights on how a compiler works or how to write one?
- chrisseaton 5y ago
  > Not sure what compiling LLVM has to do with a compiler class though. Not normal to work on a compiler during a compiler class?
  - thiht 5y ago
    On toy compilers to present the basics sure. On LLVM, maybe that's a bit much? Reading the code, why not, but compiling it? Don't see the point in a compiler course. I can't see what you could learn about compilation in a CS sense by compiling LLVM.
    - chrisseaton 5y ago
      > Reading the code, why not, but compiling it? Compiling it with changes you made to learn about how compilers work. Such as adding a new optimisation phase. > I can't see what you could learn about compilation in a CS sense by compiling LLVM. You don't learn by compiling it obviously. You learn by extending it and running it, and to do that you need to compile it.

rybosworld 5y ago

I think this is mostly right. I'm currently enrolled in same the GaTech CS Masters that the author will be starting. I have had just two courses that strained my machine out of the 7 I've completed. Reinforcement learning in particular had one project that really tested my machine. (I trained my agents on CPU, so that was definitely a factor). Even then, my agents were training in fairly reasonable times (5-6 hours max for a grid search). If I'd taken the time to vectorize every computation, I could have probably gotten the time down at least 10x. This is on AMD Ryzen 1700x - a pretty speedy chip but not top of the line - and 16 gb ram. Powerful hardware is a convenience more than anything in this program. It allows you to focus less on optimizing your solution.

ylyn 5y ago
A 1700X is quite a bit more powerful than the i5-8250U in the author's Chromebook. Probably around 2–3x the overall speed.
sand500 5y ago
Do a HPC course, the whole point is to tax your system.
- rybosworld 5y ago
  I've heard it described as the best course in the program. I won't have room to take it unfortunately. I went full ML spec and my last two courses will be Video Game Design (Been dying to take that) and Graduate Algorithms (Required)

colinmhayes 5y ago

Compiling and training ML models can be pretty taxing. Hopefully OP can just ssh into Gtech's cluster or else I could see it being annoying.

x0 5y ago

During my B.CS, most power I needed was whilst doing Android and iOS development, and even then that's just for the VMs. Otherwise did it all on a beat-up 2014 MacBook Air. Specific programs we used like SPIM, the pan/spin verification thingo, gcc/g++/etc, you can run wherever you like. EDIT: Uni provided computers with VM stuff installed, and let us remote into them during COVID.

josefx 5y ago

I once bought an Eee PC for note taking, it was great execpt for two things: * The low end graphics chip was already 12 years old on release, the advertised OpenGL 2.0 was emulated in software while everything else that came out supported OpenGL 3. * Intel didn't do virtualization on low and mid range variants of its CPUs, while an Eee PC barely had any resources to it was the second time I got burned by that. Basically 10 year old requirements meet technology that was outdated 12 years ago. Even toy samples will fail if the hardware doesn't implement the features required.

xvector 5y ago

For my computer graphics class, being able to break out my overclocked and watercooled 20-thread behemoth in an era where most processors had far fewer, definitely helped me procrastinate a bit longer, because my renders took a fraction of the expected time :) The same machine’s GPUs helped me be the first to solve a password brute-force challenge in my computer security class (designed to teach us how to use password cracking tools), though that wasn’t for any credit, of course! For my machine learning class, I was able to train models much more quickly than other students. Generally speaking, being able to brute force certain things was just fun. But yeah, I would agree that there is no practical benefit to a powerful machine in a CS education. These few instances are not worth the cost - buy or build a powerful rig because you want to, not because you think it’ll help. I felt immense pride at watching my system whir to life.

seventhtiger 5y ago
Yeah, I used my machine to do a ton of NLP on relatively big datasets, and I had cool realtime 3D visualizations to show off. I lugged my full tower to class for that one. It's really nice to have something powerful to play with.

dijit 5y ago

Had a friend who needed to borrow my beast machine to run “word2vec”. Took a couple weeks on his machine, took 30 minutes on mine (mostly because of memory). Though he was studying AI and not compsci