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.
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.
Probably around 2–3x the overall speed.
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.
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.
EDIT: Uni provided computers with VM stuff installed, and let us remote into them during COVID.
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.
It's really nice to have something powerful to play with.
* 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.
Not sure I'd enjoy doing something like compiling LLVM on a Chromebook for a compiler class.
Not normal to work on a compiler during a compiler class?
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.
Took a couple weeks on his machine, took 30 minutes on mine (mostly because of memory).
Though he was studying AI and not compsci