Windows NT on 600MHz machine opens apps instantly. What happened?

I am always frustrated with the usual answer to these kinds of demonstrations: “Yes, but these new apps are doing so much more. Also, security.”

Except, that they are not, not at the time they are launched at least. And even if they were, we have a hundred-fold more compute power, with a hundredth of the latency for memory and storage.

Regarding security, it should have negligible effect in most cases. At least, effects should not be perceptible to the human mind.

It really is just a consequence of the way we develop software nowadays. We do not need to optimize programs to make them work at all, so we just do not. We work on new features, and we hire people who can churn new features.

And we decided to optimize for developer time, instead of user time. So, instead of painstakingly developing a Web site, a native application, an Android app, and an iOS app, we just push Web apps everywhere.

> And we decided to optimize for developer time, instead of user time

That’s exactly it, and there’s no shame in that. I can, as a solo developer, build a fully featured app with a responsive UI and produce artifacts that run on Windows, Linux, and Mac. I can do that in a weekend, because of the technologies we have at our disposal. Something that would have taken a team of developers several months to do.

On the other hand, the fact that we’re abstracting everything except the business logic away is a big advantage. As soon as Chrome pushes a performance update we can see apps across the board performing 10% faster.

And when computers will be 30% faster in a couple years, vs. multiplying the developer time/cost it's a trade off.

What the hardware gives, the developer takes away. Developers will get computers 30% faster, then release 30% slower software, in a never ending cycle.

Right now, an M1 is pretty fast, but wait until all developers use it and it starts to become barely adequate.

Rephrasing Wirth's law: It takes slow hardware to develop fast software.

> Except, that they are not

My experiences with NixOS show me that they are. What do I mean by that? I am forced (MSFT Intune) to use Ubuntu for work, and was using MacOS prior to that. Both took a good heft of time to boot up, especially compared to the WinNT example. They are general purpose and come with everything under the sun installed in-case the user needs it. In the latter case (MacOS) your hands are also pretty tied when it comes to slimming it down (to be clear, apps are easy to remove, but not system cruft).

The slowest parts of bootup on my personal PC (NixOS) are POST and the GRUB timeout. NixOS takes less time than either (< 2 seconds). I chalk that up to NixOS installing very little more than I tell it to.

I agree that WebApps make situation significantly worse, but the OS itself is full of garbage that does eat CPU cycles and IOPs.

> Except, that they are not,

What a weird claim. If the new apps aren’t doing anything more, then just use the old apps.

Except you’ll quickly find that the old apps are quite simple and limited relative to what we have today.

I think the point is that they are not doing much more relative to the hardware improvements. How much less functional was windows 3.1 notepad compared to win10 notepad? 50% (utf8, maybe multiple windows)? RAM and CPU have increased 25000%. That should be way more than enough to handle the extra features in notepad.

> And even if they were, we have a hundred-fold more compute power [...]

This doesn't sound right. Sure, GPUs are probably that much faster, but we certainly don't have the equivalent of a 60,000 MHz CPU.

The clocks have not become a hundred-fold faster, I grant you that. But, combined with specialized instructions, the improvement of the instruction pipeline, the growing cache, the multiplication of shadow registers, the addition of hyperthreading, and the increasing number of cores, we probably do have a hundred time more computing power in a modern laptop.