The internet is killing old PC hardware [video]
youtube.com
youtube.com
I’m running 24 threads at 5GHz, sipping data through a GB pipe, and somehow, sites still crawl. Apps hesitate like they need divine intervention just to start. Five billion instructions just to boot up? My 5GB/s NVMe too sluggish for a few megabytes of binary? What the hell happened?
The internet isn’t just bloated—it’s an executioner. It’s strangling new hardware, and the old hardware? That’s been dead for years. Sure, you can run a current through a corpse and watch it twitch, but that doesn’t mean it’s alive.
This is nothing else than enshittification: https://pluralistic.net/2025/02/26/ursula-franklin/
After all; having paid gazillions to engineers and project managers to build the sludgefest, all that cash needs to be harvested back into the pockets.
(do relate. progress on personal projects can be slow due to this kind of stuff sadly!)
It's AltGr+[-] (–) or AltGr+Shift+[-] (—) on my keyboard layout (linux, de-nodeadkeys) btw.
AltGr+[,] is ·, AltGr+Shift+[,] is ×, AltGr+[.] is …, AltGr+[1] is ¹
[Compose][~][~] is ≈, [Compose][_][2] is ₂ (great for writing H₂O), [Compose][<][_] is ≤, etc.
I use all of these, and more, guess I'm an AI now :(
To be fair I intentionally use — incorrectly by putting spaces around it, just because I hate how it looks without the spaces ("correct" English grammar says there should be no spaces.)
Are you actually measuring the load-time bottlenecks in devTools?
I don't know the exact details but it appears a lot of sites are sitting around waiting on ad-tech to get delivered before they finish loading content.
Though true I would have spaces around it.
edit: failed to compile. missing semicolon
On Mac, it's configured by default: <option>+<shift>+dash
It would really be disappointing if typographic pedants start being falsely accused of being AI.
On Linux (maybe only certain distros, not sure) the keys are different,
but you can enable a Compose key and enable special character keybinds as well.
For example on Mint:
en–dash is "<compose> <minus> <minus> <period>"
em—dash is "<compose> <minus> <minus> <minus>"ZeroGPT: Your Text is Human written (0% AI GPT)
QuillBot: 0% of text is likely AI
GPTZero: We are highly confident this text was ai generated. 100% Probability AI generated
Grammarly: 0% of this text appears to be AI-generated
None of them gave the actual answer (human written and AI edited), even though QuillBot has a separate category for this (Human-written & AI-refined: 0%).
I was trying to figure out what made it so obvious, the dashes were one thing, the other things I noticed were:
- Bits of the text were in bold.
- The tone was horrible, very cringe. Full of superlatives, adjectives and cliches.
- I live somewhere where English is the third language, most people don't write emails in English without a few spelling or grammar mistakes.
- Nor do they write in paragraphs.
- It's also pretty unusual to get a quick reply.
Lots of these things are positive, I guess. I'm glad folks are finding it easier to communicate quickly and relatively effectively across languages, but it did kinda make me never want to do business with them again.
I wonder how much of this is caused by the AI companies using Reddit as a major training source.
For example on Mint en–dash is "<compose> <minus> <minus> <period>" and em—dash is "<compose> <minus> <minus> <minus>"
You might think that their thinking would change when you point out that the code is run millions of times each day on computers all over the world, so those saved seconds will really add up.
But all those millions of saved seconds do not affect the company's bottom line. Other people reap the benefits of the saved time (and power usage, and heat generated, and ...) but not the company that wrote the code. So it is still a complete waste in their minds.
Multiply this thinking across millions of programs and you get today's situation of slow, bloated code everywhere.
Could an AI just crawl through an organization's entire code and make optimization recommendations, which are then reviewed by a human? It seems like there could be a lot of low hanging fruit.
Even Electron probably could have been fine if the browser was just a document layout engine and not a miniature operating system. There was an article going around a few years ago about Chrome - and by extension, Electron - including a driver for some silly device I don't remember, like an Xbox controller or something. Googling tells me it wasn't an Xbox controller though. Every electron app includes an entire operating system, including the parts not needed by that app, including the parts already included in the operating system you already have.
Language runtimes don't have to be this way, but we choose to use the ones that are!
Memory is infinite, CPU is infinite, disk space - we don't give a shit because they are all on the sucker's computer. Just like "your privacy" (aka your data) is very important for us, also your computing power is very important for us. I wish i was sarcastic.
Aside from websites, we will talk about that in a minute, how is performance? I am running Windows 11 on new hardware and it is running great. I built personal box with 64gb of the fastest DDR5 and a AMD 9900xtx. The most expensive component (and least bang for the buck)... the video card. This is my first time to have an NVMe disk and its absolutely amazing.
I am running Debian 12 on mini computer with much less hardware and its doing amazing there too. I can run anything on the box except AAA games and 4k video movies.
Now, for the web talk. I was a JavaScript developer for 15 years, and yes its garbage. Most of the people doing that work have absolutely no idea what they are doing and there is no incentive to do so. The only practical goal there is to put text on screen and most of the developers doing that work struggle to do even that. Its why I won't do that work any more, because its a complete race to the bottom. If I see a job post that mentions React, Vue, or Angular I stop reading it and move on.
No. It is buried under the laziness of build-fast, optimize-next. Except optimizing never comes. Building fast requires lightweight development on heavyweight architectures. And that means using bloats like JS frameworks.
I might rephrase that for our field: anyone can design software that works. It takes an engineer to design software that barely works.
Instrumentation of browsers to show performance constraints remains poor and/or a Black Art, so I'm somewhat winging this.
The same system performs wonderfully with scripts, non-browser based requests (e.g., wget, curl), terminal-mode browsers (usually w3m or lynx), and local applications (mailers, audio ripping / playback, GIMP, word-processing, and of course shell tools, as well as a few occasional games perhaps.
Which makes the loss of all our DEI programs over the past month all the more painful.
Further, the history of persecution of magic pracitioners would qualify them for DEI outreach. So it works either way. ;)
It's rare for jokes to play well: <https://news.ycombinator.com/item?id=7522885>.
Particularly where it's not clear whether or not someone's making a joke, or is merely misinterpreting language. A common occurrence where many participants' first language is not English.
And especially where the joke itself runs into a tired political fracture.
Back in the 8 and 16 bit days, companies managed to make software available across all of them, or at least the most common platforms, in a time where performance called for hand written Assembly, and each hardware platform was its own snowflake.
And yet in the age of high level programming languages, the best most folks can think of is shipping a browser alongside the application, not only they show complete disregard for the platform, they don't really care about the Web, rather ChromeOS.
And yes this includes VSCode, which has tons of C++ and Rust code, and partial use of WebGL rendering, to work around the exact point of being based on Electron.
Said childhood gaming, with what feels like a lot of latency: https://www.futrega.org/digger/
Sounds like a sacrifice I'm willing to make!
I'm pretty heavily involved in SD video restoration and I drive things from "retro" systems mostly because I need 32-bit PCI slots and Windows XP in order to interface with older broadcast engineering hardware.
I could shoehorn everything into a more modern system, but what I have suits my needs.
For most local applications, or simple over-the-Web fetches via curl, wget, etc., mid-aughts hardware or earlier often suffices.
Amongst my hobbies are occasional large-scale scraping of websites. I'd done a significant amount of this circa 2018-19 on a mid-aughts iMac running Linux. One observation was that fetching content was considerably faster than processing it locally, not in a browser but using the html-xml-utils package on Debian. That is, DOM structures, even when simply parsed and extracted, provide a significant challenge to older hardware.
I had the option of expanding RAM, and of swapping in a solid-state drive, both of which I suspect would have helped markedly (swapping was a major factor in slowness), though how much I'm not sure.
I'll also note as I have for years that this behaviour serves advertisers and merchants as a market segmentation technique. That is, by making sites unusable on older kit, in a world where physical / real-estate based market segmentation isn't possible, is an effective way of selecting for those with discretionary income to buy modern devices. Whom we presume have greater discretionary income / higher conversion rates as well.
(Multiple network round-trip requirements is also a way to penalise those making access from distant locations, as those 100--300 ms delays add up with time, particularly for non-parallelisable fetches.)
I'm not arguing that all site devs have this in mind, but at the level which counts, specifically within major advertisers (Google, Facebook) and merchants (Amazon) this could well be an official policy which, by nature of their positions (ad brokers, browser developers, merchants, hardware suppliers) gets rippled throughout the industry. In the case of Apple, moving people to newer kit is their principle revenue model as well.
What are these webpages exposed in this video supposed to do¹? Display some text, pictures and maybe some videos. How does it feels it term of complexity and hardware requirement compared to a 1998 FPS which achieved impressive gamer-experience breakthrough incorporated into a customer grade product? Does it seems more fair as a comparison?
Now, obviously you can’t expect all webdeveloper interns out there to reach the level of Valve engineers in 1998, sure. But the frameworks they are asked to use should give them the sober way as the easy path, and let more complex achievements still accessible in the remaining computational resources.
¹ As opposed to something using WebGL or other fancy things incorporated in contemporary browsers.
A solution could be to put another layer on top of the internet. This could be done by means of a "presentation proxy", similar to cloud gaming, e.g. based on VNC, where only a VNC client is run on the old computer, and the browser is running on the presentation proxy.
That still leaves the attack-surface of browser-based activities (do you really want your recreational activities sites interacting with your financial services?), but both gives the option for fresh respawns (OS and/or browser) and physical and network isolation of your local storage and data from your browser.
(This presumes hygiene on any browser-based downloads as well, of course.)
In general, the idea that we'd want all our data on a globally-accessible network is seeming increasingly unwise to me over time, given both technical and political developments.
Late edit: Oh, and Browsh: text-only browser-via-proxy supporting CSS/JS.
My non-tech friends call me paranoid at the basic level of protection I do - separate browsers for everything, and a couple of VMs for e.g. finances.
Vlan has always been an important part of LAN security but we're just out here with always on, full Internet access? My home firewall logs show an insane amount of bot / scraping etc.
I usually I just go to archive.is and read the page there.
Making the difficult stuff possible really should not have made the ordinary easy stuff that we already had so much more difficult.
We used to browse the web with 486s and megabytes of RAM. If all I'm doing is reading a newspaper I shouldn't need giga bytes of RAM.
Using 2015 PC in 2025 is far less painful than using 1995 PC in 2005.
When even top-of-the-line older hardware is nearly free, it makes little sense to optimize for bottom-of-the-line older hardware.
It does very well on any modern internet task, as well as playing modern video games with a few-year-old used graphics card.
That is also one of the potential problems of pre-Haswell hardware, distros might stop supporting it in near-future
"MarkText is a lightweight, user-friendly Markdown editor that serves as a free and open-source alternative to Typora. It’s designed for everyday users who want a clean, intuitive experience."
https://myownsys.com/2024/11/24/everyone-should-have-a-free-...
Could you provide an example?
Anecdotally on my previous laptop from 2016 it was often laggy and took longer to process a single key input than Vim took to start up and load plugins, and natively compiled editors like Emacs and Sublime tend to be noticeably snappier.
Whether that matters to you or not is subjective, but I don’t like editors pretending to be IDEs anyway.
When I had an old computer that couldn't deal with the indexing done by JetBrain's CLion or Microsoft's Visual Studio Code - I used nvim. It was a pleasant experience, however it lacked support for visual debugging (and please don't talk about GDB TUI as if it is an option). Now that I have a computer that can deal with the indexing, Visual Studio Code is just fine. In fact, it is considerably more lightweight than JetBrain's CLion IDE and is very easy to setup.
For example, https://code.visualstudio.com/docs/devcontainers/create-dev-..., allows you to setup development inside a container. In practice this allows anyone to quickly pull the repository and start working on the code, including building and debugging without having to worry about setting up toolchains or environment as it'll be done automatically for you.
True, though performance is often completely unrelated to the amount of features. The build/debug actions you mentioned should not have any impact on text editing speed.
1. Using uBlock Origin and NoScript would help
2. Sorry, but the AMD A4-1200 (2 Cores running at 1.0 (!) GHz, 4W TDP, Single Channel DDR3-1066[1]) was already slow when it was new back in 2013, it was introduced by AMD as a low budget *Tablet* option
3. Regarding Video playback: As mention in the video, forcing Youtube to serve h264 should help, since the iGPU supports h264 decoding [2][3]
This also makes me wonder: What does Youtube serve on this old machines? On my old Vega 56 (UVD 7.0 h264/h265/JPEG decoding support), yt runs without issues, but on the displayed HD 8180 (UVD 4.2, h264 decoding support) it doesn't? My current system has a Nvidia 1050 and gets served AV1, which the 1050 also does not support in hardware.My first instinct for yt's "auto" implementation would be to the serve a video in 480p and a codec the users hardware decoders supports, and if the user switches to a higher resolution, to serve it in av1 in order to preserve bandwidth. Maybe Youtube does not have access to the users hardware decoding capabilities, or they want to preserve bandwidth even at low resolutions.
[1] https://www.techpowerup.com/cpu-specs/a4-1200.c1706
[2] https://www.x.org/wiki/RadeonFeature/#index8h2
[3] https://en.wikipedia.org/wiki/Unified_Video_Decoder#Format_supportSo I visit thing-i-want.com and it doesn’t load because NoScript is currently disabling JS for that domain. No problem, I temporarily enable JS for thing-i-want.com
The page refreshes and suddenly NoScript is disabling JS for 10 more domains!
That seems excessive, maybe the page doesn’t need ALL of those scripts to function. I will enable that cloudfront domain and that one that has “static content” in the name.
Page refreshes.
Okay it mostly works now but also NoScript is showing disabled JavaScript from 5 more domains!
..Anyways Sometimes sites are running scripts from 15 or more domains and sometimes they are nested 4 domains deep. It’s absurd and OF COURSE it overwhelms older devices.
If you want to use a modern browser on an older device, use a browser with a script blocking plugin
Sites that just fetch JavaScript from their own domain, not as much.
So I agree it is an anti-pattern for typical use cases.
But if you’re trying to get the most out of old hardware, it will make some websites more usable.
My pattern is basically browsing the wild net with FF/Noscript and use Chrome for "apps" (gmail,sheets,etc) that i feel have a reason to use JS.
Very rarely it'll happen that I'll care enough to go through the list of possible domains to temporarily whitelist before finally giving GTM a shot, then immediately remove the whitelist. Usually I don't get that far, especially because if it hasn't worked by then, enabling GTM doesn't tend to work either, it's just a bad site that isn't actually providing what it claimed to provide. NoScript has never disincentivized me from visiting a new site, but it has made me give up on some or look for alternatives. My daily experience is pretty minimally impacted by it. (Still, I don't usually bother installing it on work machines or my travel laptop (which is remoting to my home PC most of the time anyway), and sometimes I'll just load the page up in a chromium tab (incognito or not) rather than play the game of five refreshes from whitelisting JS.)
The performance impact is quite minimal I think, especially if you compare the difference between Firefox with NoScript and Chromium without, the latter is just faster because it's not Firefox. The oldest machines I still use sometimes are from 2009 (with an i7 920, pretty good for the time) which as my old daily-driver I used NoScript, and 2017 (my travel laptop with an i7 7820HQ) where I don't bother. Neither is all that slower for web stuff than my current daily driver with a Ryzen 9 5900X. The web is just slow even with newer hardware. (In contrast to others here though, I immediately notice the difference of better hardware with local applications, especially content authoring ones like gimp or krita.)
Ain't nobody got time to live like this.
Also, companies don't want to invest in supporting multiple versions at any point in time, and can't afford reputation risk by not forcing upgrades.
My company let's the employees to request for and get a software installed, but can hardly allow them to use the features! The Risk & compliance department wouldn't like anyone to work or use any software properly. Any moving thing is a risk.
Hah, posted as a comment to a YouTube video.
I know you have to go where your audience is. But like, block ads, block most JavaScript, don’t go on YouTube, and the internet is much better.
This shoves you off into a smaller sub-sector of the internet. But if you are somebody who is nostalgic for the era when the internet didn’t suck so bad, it was quite a bit smaller back then too.
https://archive.org/details/software
National libraries across the world curate books, documents, and other formats but we as Humanity unfortunately are not very good at preserving the software side of History.
There was an interesting FB engineering post here about cost benefits to using a QLC storage layer... No comments. I feel like a lot of technical people nowadays don't even appreciate that we get FAANG engineering blogs and talking about hardware. Maybe I'm just jaded from having worked in a data center.
also, rhetorical: did you comment in the FB engineering post?