Every 16 GB machine I've used for work in the last three years (until current 32 GB one) I'd occasionally run out of memory. Culprit? Web browsers... On Mac, Linux, and Windows. (Though occasionally other stuff needed significant memory too.)

For me the Android emulator in conjunction with a fairly large react native dev server and an ide is enough to fill 16gb. I've had to resort to closing the ide before starting the emulator, and waiting for the first build to complete or things get OOM killed. If I run the test suite whilst the emulator is open then there's a good chance something will get killed as well.

This looks like it could be a compelling choice for my next laptop but will have to pass if it's not possible to go to 32gb

This only happens if you’re the type to keep 70 tabs (or whatever) open. I run 16GB and have never, ever had a browser cause me to start heavily paging.

>This only happens if you’re the type to keep 70 tabs (or whatever) open.

not really, you can just leave a leaky or script-laden page open for awhile to see the same behavior, depending on how things are put together.

example: blue iris is a cctv package that has a webui module. That web-ui client-side browser tab will inflate to system-memory size if allowed within a few days.

Not much of a point other than there are other (probably pretty esoteric) things that one can do to a browser that'll make them fall apart even on modern hardware.

Also keep an eye out on all those extensions...

Machine learning will generally swallow as much RAM as you can throw at it, assuming you're not just using a nvidia coprocessor with its own memory (and even in that case, you have to keep it fed).

Does it really make sense to do serious ML in a laptop though? my ML/DL training rig runs pretty hot all the time, I could use it as a space heater in the winter. Even if I could have all its processing power in a notebook I definitely wouldn't want it on my lap.

There are a lot of people using virtualized development environments, which can make it much simpler to run a production-like system locally.

I’m pretty sure 16gb machines have been the baseline for developers everywhere I’ve worked for the past decade.

>There are a lot of people using virtualized development environments, which can make it much simpler to run a production-like system locally.

When running Linux, you do not incur the VM overhead for containers (Docker) that you do on Mac or Windows. This allows for isolated production like environments, without carving out multi GB chunks of RAM. Additionally, if you are targeting Linux for production, you often don't even need to containerize, unless you need to run three different versions of node/python/ruby or something else that is just library conflicts all over the place.

> I’m pretty sure 16gb machines have been the baseline for developers everywhere I’ve worked for the past decade.

My 2012 daily driver was a well spec'd HP, and it had 8GB of RAM. It did not support more - I tried and had to take the RAM back because the machine would only address 8GB. A look at Macbook specs in 2012: 4GB or 8GB models. I think the 16GB becoming the norm was a few years later.

And you don't have to incur the container overheads if you switched to Nix

Let me introduce you to my friend Java.

Java sets upper memory limits, so that shouldn't be an issue.

I was replying to the whole 8GB is enough for typical development. Java is an extremely popular language to develop in and the IDEs alone eat up 2-4GB on their own.

By default a Java app will set its max heap size to 25% of RAM (Java 11 and above). While it may not use all of that, it’s likely to. So, by default, running two Java apps on Java 11 on 8GB of RAM has the likely potential of eating 4GB RAM. Then there’s the OS needs, browser needs, etc.

Clearly 8GB isn’t enough for such a developer. I’ve had this discussion countless times with people over the past two decades. It always ends the same — with people from both sides of the discussion shaking their heads.

The only claim that I feel anyone can make, given enough experience, is that RAM needs vary. So making a claim that N amount of RAM is enough for many or most will always be patently wrong in my view.

Completely agree.

Unless your stack involves 5-10 JVM apps (with the smallest ones generally working at -Xmx512m but some expecting around 2-4 GiB), and your IDE is also a resource-hungry Java app, in which case the bare minimum is 32 GiB RAM + about 10 GiB of swap.

True, if you are trying to emulate an entire cluster of servers on your laptop you are in for a bad time.

I’ve done that! But I also have 32GB RAM in my machine :)

I want 64 GB but I will say for my current needs, 32 GB is the sweet spot. My normal daily use is around 20GB. And it is a MacBook Pro. One day I’ll probably go back to a laptop running Linux. It is the fastest for code compilation in my experience.

Using Linux on a desktop workstation, I'm normally able to get by with half the amount of RAM that my Macbook requires for the same activity. Probably because I'm not having to run MacOS and Linux at the same time to pretend that I'm not running Linux.

But it's not just memory, I find running code on MacOS generates a lot more heat causing the CPU to spike and harder to run in the first place.

I disagree. When I have to use a system that isn't Linux which seems to be the corporate rule, the requirements go through the ceiling as it always seems to need to use Docker to prevent to be Linux anyway.

Running multiple OS' is always demanding. They could fix it by just using Linux for the dev environment in the first place since all their tools are always electron or Java based.

What’s your full tech stack and OS, and how much swap/pagefile do you have? My personal Windows machine has 8 GiB, and it’s quite easy to fill it up by just running Firefox (with far too many tabs) and WSL, let alone an IDE.

AMD Ryzen 3, running Pop!_OS, 8gb ram, 16 GB swap.

Perhaps it's a windows thing?