Is it possible to restrict this as a user? Like to force webpages to use under a certain amount of render/paint time/resources or else just break so that one dumb tab doesn't use up all my battery? Then I can opt-in to greater resources usage if it's a webpage I actually care about.
I've seen the "This webpage is using alot of resources" popup before but I don't think it would happen in this case.
Because honestly I think this is horrifying. I would rather it switch from grey to red "recording" dot than use even the 6% the author decided was "fixed". In 99% of cases I do not care at all about the "artistic vision" of the UI designer and in the other 1% of cases (say an in-browser game or some useful data-viz) I could choose to allow the tab to go crazy with my resources.
Seriously. It seems pretty reasonable to allow a web page a total amount of processing that is something like 100% CPU or GPU for 5 seconds. (E.g. can be 25% for 20 seconds.) And beyond that it gets throttled to 3% CPU cumulative max for however long it's been open so far.
And make that the default for popular browsers, so sites are forced to be efficient or else be super super janky and stuttering. And allow a permission for unrestricted processing that things like WebGPU demos or games can ask for.
Firefox already has a monitor module that AFAIU tracks UI stalls and prompts the user to stop the website if it hangs for a long period [1]. It probably can be further developed to also track resource consumption and show a warning if some threshold is reached.
There were those apps on pentium 133mhz era computers too, even started in faster tha 5 seconds.
For 99% of the people, word 97 was good enough then and is still good enough now to do whatever they need. They can literally still use it today. (yes yes, docx compatibility, i'm just making an example).
Web stuff is one of the few things where the amount of useful data on an eg. weather report site (webmail, news site, bus timetable site,...) hasn't changed, the cpu requirements have gone way way higher for no useful reason (except to serve more ads), and the users can't use the "old versio" of the site anymore. It's literally gigabytes of ram and multiple seconds of 100% cpu usage to service 5kB of article text and a 50kB article photo.
This isn’t unsolvable. A CAD app or similar could ask for permission to use extra resources. There’s already one for storage, extending this to CPU and MEMory is not far fetched.
A bookmarklet that injects custom CSS and selects all/certain items and applies the `contain: content` and `content-visibility: auto;` rules could do to trick.
Additionally, the pseudo `:empty { display: none; }` selector may get you additional mileage.
Probably too late, but uBlock Origin can disable animations (technically fast-forwards to the end, less likely to break the page) for all websites. In the Settings > My Filters tab, add:
> In 99% of cases I do not care at all about the "artistic vision" of the UI designer and in the other 1% of cases (say an in-browser game or some useful data-viz) I could choose to allow the tab to go crazy with my resources.
I'm with you 100%. Although I'd go one step further and say CSS just isn't needed at all, and should be removed from all browsers. Same goes for WebGL (if you want to play a game, download Steam).
This would fix all of your issues and save an unbelievable amount of energy across the planet. Unfortunately people like us are a dying breed!
Answering because I know the answer, even though I disagree:
There are some situations where having some animation can "attract attention" to parts of the UI, for guidance ("there's some info text here"), feedback ("operation complete"), state changes, timing. All this of course doesn't apply to decorative animations.
With that said: a lot of the situations I mention above are manufactured. They are often because of changes happening away from the mouse, or because the interface is brittle or too slow, and the user doesn't have confidence that something really happened, or because the organization of elements is not functional and things are too far apart that the user might miss something in a totally different part of the screen.
IMO, with enough thinking you can come up with alternative interfaces that don't require animations at all.
Imagine a list of 10 items that gets shuffled so that the order is completely rearranged. With no animation, it would take a lot of additional cognitive load to find where each item went. With animation, your brain would track the general movement of each item much more easily.
Basically, animation provides additional information about object state. Removing that extra information increases cognitive load.
This isn't to say all animations are useful. Many animations are excessive or completely unnecessary, which is probably what has given you a negative view of them.
Absolutely, without question. Do you think that humans, who evolved in a physical word where basically everything happens in an incremental way don't extract valuable information from in-flight state?
Also, as an example just try to use a window manager that just switches instantly to a new desktop, vs a very short animation. I find the former easily disorienting, and it is even more so when you have a more direct gesture controlling the action (e.g. touch or mousepad). Of course there is too much of everything, and too much animation sucks, no question. So is too much fat, but that doesn't make normal amounts any bad.
People are different, I guess. I remove those animations whenever I can. A desktop manager that switches virtual displays in a single frame is exactly what I want.
You can disable CSS in your browser. The fact that you even complain about this tells me that you didn't didable, yet you damned it to be removed. Pretty hypocritical of you.
Just wrap it in a container with fixed height and "overflow: hidden".
Now the layout engine knows that it doesn't need to recalculate positions of elements outside that wrapper, and it's much faster.
By the way, the same trick was speeding up large <table> rendering back in the day. As long as you know the size of your rows or columns ahead of time, which kinda defeats the purpose of <table>.
As an old-school webguy, fixes like this article make me sad.
The average front-end person these days has so little respect for the DOM and then gets mystified about why the McMaster-Carr website is so good despite being build with ancient practices.
Well, and especially because the “fixed” version is still - if I’ve read correctly - consuming 12% CPU before any user interaction on a page that’s otherwise substantially a plain or possibly RTFish text editor.
How is that considered acceptable? It’s still going to drain the battery faster than is justified, even if at a somewhat reduced rate from the pre-fixed case.
The acceptable range would be something like 0 - 1% when no interactions are taking place.
I’d still deploy the fix in the article as an interim solution - better is after all better - but I wouldn’t stop there.
Front-end web devs need to have more respect for power consumption and battery life (proxies by CPU and GPU usage).
> Front-end web devs need to have more respect for power consumption and battery life
they need to understand and believe that their wishes are not everyone's wishes, and that they alone are the cause of their own problems. Everyone else works hard to give them the creative freedom they demand, but for multiple decades now they have demonstrated over and over how they love shooting themselves in the foot at every opportunity.
> consuming 12% CPU before any user interaction on a page that’s otherwise substantially a plain or possibly RTFish text editor. How is that considered acceptable?
It's not acceptable, unless you are a front-end web developer. If you are a front-end web developer, you don't consider this a problem in any way because the page looks as it is intended to look.
You might be surprised at the CPU usage of rendering a GIF. I'm not sure why, but I've previously noticed Slack takes noticeably more CPU when there's an animated emoji on screen.
A few years ago a client that I hadn't spoken to in half a decade called me to thank me. He wanted to update his website to work with tablets, checked before calling me, and realized that his website just simply worked. Being the caveman that I am, I wrote the damn thing using the simplest CSS possible. Simple equals compatibility in this field.
Im seeing a lighthouse mobile performance of 64 with Javascript execution time, unused js, legacy js on modern browsers, etc. as some of the top reasons.
Yes, looks like this property isn't just a hack, but was designed with exactly the performance considerations laid out in the article in mind. Animating height is only expensive when the element’s height affects layout, which it doesn't have to!
Thank you! This is what I am looking for in HN comments. Layout engine cannot be that stupid per the article. I guess it is not once you give it enough hint.
You need to think about what the browser is actually doing when it displays a web page. You need to think about what happens when you type the URL in and what the browser is doing conceptually.
If you have any experience programming 2D graphics with SDL this will help you to understand what is happening as well.
There is an old school talk by Nikolas Zakas on JavaScript Performance from 2009 which while some of it is out of date now due to improvements in Browser engines, the fundamental ideas are still the same.
If you skip to the 35 minute mark he speaks about reflow specifically. Generally I still use many of these techniques when writing vanilla JS (which is unfortunately rare these days). There are other talks where he talks about many other performance specific topics.
Obviously none of this is as good as real world experience of dealing with a thorny issue.
> If you have any experience programming 2D graphics with SDL this will help you to understand what is happening as well.
No. No it will not. Nothing in 2D graphics will help you understand things like "simple animation required full layout re-calculation and re-painting of the entire page"
Which draw calls result in "changing border width will cause a full page re-flow in some browsers, and bot others"?
Which draw calls are "literally any animation will cause the full re-flow of the document unless you make sure an element is not a part of the document, but then you'll have issues with positioning, weird gaps etc."
Which draw calls teach you that "well, this primitive animation consumes 15% of CPU, but will only consume 6% if you go out of your way to make it 'performant'"?
You are being disingenuous. I can always use some hyper specific example to disprove any statement about what is generally true. That doesn't mean that understanding what is typically happening is useless. Which is essentially what your claim is.
> You are being disingenuous. I can always use some hyper specific example
These are not hyperspecific examples. We're literally discussing under an article which has a primitive animation which was, quote, "using 60% CPU and 25% GPU on [a] M2 MacBook"
How is 2D graphics programming helping there?
The article goes on to describe, correctly, layout properties ("The W3C spec is full of these!"), paint properties ("this tiny "bikeshed" SVG which you can find on lots of W3C spec pages. It costs ~30% CPU!") and composite properties.
These are not specific examples. These are literally footguns upon footguns that literally have no corresponding counterpart anywhere.
> That doesn't mean that understanding what is typically happening is useless. Which is essentially what your claim is.
Yes, knowing 2D graphics programming is 100% useless to understand what's happening with CSS and DOM.
> These are not hyperspecific examples. We're literally discussing under an article which has a primitive animation which was, quote, "using 60% CPU and 25% GPU on [a] M2 MacBook".
No we are not. Someone asked about how one learns some techniques that the OP mentioned (that were simpler than the ones given in the article). I gave some general advice on how one should think about what is happening and some general advice about how to think about performance.
Then you are pretending that none of this is relevant because you are concentrating on the hyper specific stuff in the article (which we are no longer talking about).
I also disagree that this cannot be reasoned about by learning a bit about how graphics works, Cartesian coordinate system and literally turning on the paint flashing tool that is in dev tools. It is literally obvious what is happening when you see it.
> Yes, knowing 2D graphics programming is 100% useless to understand what's happening with CSS and DOM.
In fact it was very helpful to help me to understand what was happening. I literally said to myself "wait a minute, that is kinda like this thing I did in SDL 1.2 in university". You are literally telling me, that something I know to be useful isn't, because you say so. That is unreasonable.
Yes. Yes we are literally discussing this under an article whose introduction literally says what I quoted.
> Someone asked about how one learns some techniques that the OP mentioned
Yes, and the techniques OP mentioned literally cannot be learned from 2D graphics because nowhere in 2D graphics are you going "ah yes, to fix 60% CPU utilization for a primitive animation, you need to apply this awkward workaround that will tell the browser not to reflow the page".
Edit: and this particular technique is only applicable to this specific property for reasons that do not exist anywhere else, except in the browsers.
> wait a minute, that is kinda like this thing I did in SDL 1.2 in university
Well, it's not. Just because it kinda somewhat looks like something different doesn't make it so. Hence my question in this comment: https://qqrl.tk/item?id=44658482
Because those are not hyper specific, no matter how much you pretend they are
Usually you try to build something, realize it’s slow, then do a combination of searching for possible solutions, trying them, and profiling again and again to gain this knowledge.
Not sure, but if CSS first principle is parent sets width going down to children, but sets height going up from children, then you could see how children changing its height leads to layout recalc all the way up.
So the you need to fix that height /remove resize, then google on which technique to use?
not really. the web is 30+ years of fixes, shortcuts, compatibility hacks, tech industry politics, and other assorted mistakes.
the "first principles" to understand it all is the complete set of mailing list archives of the WhatWG, plus the archives of the bug trackers of all the major browsers.
I don't think it would be different for any other system. Trying to make a piece of computer software that is completely responsive to big displays, small displays, touch screens, mice, etc is just a ton of work and a ton of small details you can't easily learn until you need to learn them.
yes. i don't mean to unfairly criticize the web as a platform, i just know it's warts because it's the platform i work on.
my criticism is reserved for the concept of "first principles", as if there should be a cohesive and simple set of rules that you can build up all the knowledge you need from. that's just not really how any complicated system works: building a comprehensive understanding of a complicated system takes a lot of work and a lot of time.
It could _probably_ figure out a lot of cases via some kind of static or runtime analysis.
Like the example from the linked article, there's one animation involved essentially and if the browser does the hard work for the entire length of one animation cycle and no layout ever changes, just assume it never will? It would have to have a bunch of edge cases to handle correctly I'm sure though.
Browsers layout engines are pretty smart, and also optimized for speed at the same time, which makes them very complex. I can do one or the other in my code, but both is too hard. Now add there security.
But Google having more people working on Chrome than on Android I would say all the low-hanging optimizations have already been optimized, and all that's left is pretty complex even for teams of thousands.
Translation: just make a completely different functionality and effect than what was originally intended because the DOM can only handle static content, barely.
The suggested change wouldn't affect functionality or the effect. It just tells the browser that there's a thing around the changing part with a fixed size that can't possibly change layout of anything around it, to the user it's invisible.
CSS contain has made so many performance "truths" obsolete, but so few developers realize it. I've seen massive efforts to do things like port text editors to WebGL where rasterizing text is a huge pain where putting `contain: content` on the individual elements in the DOM version would have delivered most of the perf improvements.
Browser rendering engines are now sophisticated GPU-accelerated compositors. Absolute positioning with contain: strict removes basically all CSS layout from the perf equation, and you don't have to write your own compositor or line-layout! `* {contain: content}` and flex/grid gets you the good parts of HTML and very good performance.
What amazes me is that so much a M2 Mac's resources would be used to render a website even if it rendered everything from scratch. There is almost no graphic content compared to a video game produced decades ago, and they would easily render a frame in 1/2 the time on decades old hardware
A live layout tree that has to be repainted and composited yet intersect with arbitrary layers like an accessibility tree and what would naively be N:N style calculations is completely different architecture.
Consider examples where people toy with the concept of replacing the DOM with a canvas but basic things like text selection don't even work anymore.
Keep in mind that the percentage in Activity Monitor is 100 % per core, so a 10 core CPU can go up to 1000 % usage. So 60 % means 60 % of a single core not of the whole CPU.
And it's sensitive to dynamic scaling as well, isn't it? So it's possibly not 60% of a core at max clock, but maybe of a core clocked to 800mhz or something.
I fell for that shallow analysis myself once when I noticed just how apparently bad the gpu usage was for a big of animation in a web page, until I noticed my 30% load gpu still drew as much wattage as it did on idle.
> On my M2 MacBook, the renderer process is now using 6% CPU (down from 15%), and the GPU process is now using 6% CPU and less than 1% GPU (down from 25% and 20%).
Wait, this is supposed to be an improvement? 6% CPU for a lame animation?
If the bars are not related to actual data and are purely just pre-canned animation, the gif would just come with additional file size for a one time cached type of download. If the bars do need to be animated to real-time data coming in, then the gif would not be the right fit for that need.
Can't imagine anyone is deriving much information value from a visualizer with 3 bars. Apple Music has a similar animation and it will start bouncing when you click play, while the music is clearly downloading and there's no sound.
Just because it isn't useful to the user doesn't mean some dev hasn't spent the cycles doing it. I know I've personally had to write code that was an absolute waste, but some PM wanted and it was coded. Usually so it can be a bullet point in a demo that impresses low skilled PMs or sales drone, but techy user types just roll their eyes when it's pointed out to them.
You've moved the goal posts. Nobody said an input levels indicator isn't useful. We're talking about this specific graphic not being a useful graphic for visualizing live data. You've wandered away from the pack here
I've almost never seen these tiny icons to have any actual useful information being conveyed (meaning they are tracking amplitude or pitch). They're usually purely decorative, or boolean in function, meaning you can see they're flat if you are completely silent or they're bouncing around if not silent.
You can have 2 GIFs for "silent" and "sound happening" if one cares to truly indicate if it's picking up sound / playing sound, and switch between them. Will be 1000x simpler and 99% fewer lines of code to maintain forever.
I don't normally have that issue, but one site had (has?) a bright green thing that jumped periodically that was so vastly distracting that I could not read the text on the page without covering the 'thing' somehow, eg with my hand! Absolutely pointless. I had to give up using the site entirely, even though it was otherwise useful.
Personally, I'd be annoyed by both the resource-consuming animations and the blurry GIFs/canvas. Infisical does use the latter (canvas) for icons in their UI, and I somewhat hate it. I'd rather look at crisp, but static icons.
Canvas should never be blurry. If it is, something is doing a bilinear upscale. I'd guess someone forgot to take the scale factor of your display into account.
Or there are images being used in the canvas, which would defeat the purpose for the use case you described.
GIFs historically were pretty unoptimized from a display perspective (as anyone with a moderately-active Tumblr dashboard could tell you), so I wouldn't be surprised if it was worse.
I disagree. The height of a DOM element is much more expensive to recompute as this article shows than a single element which gets frames swapped out with static images. The latter can be guaranteed to never impact any other part of the document. The end result of either is that the pixels in the animated area need to be recomputed, and end result which will be the same for both, but a GIF will have no further impact on anything else whereas resizing DOM elements has side effects that have to be constantly updated and checked for which scale with the number of sprites you're animating (unlike the GIF bitmap which is gonna be 'flat rate').
Likewise, for a tiny animation, having the frames of the GIF in memory and blitting them onto the screen in sequence sounds like orders of magnitude less CPU/GPU than performing all the ridiculous math-based transforms over and over again.
The Tumblr example is not instructive here because those were frequently many many frames, very colorful (often photos), and/or large in dimensions, taking up a ton of memory which was probably the main problem causing our browsers to choke. This would be none of those things.
Does a note taking app need such an animation if apparently it's that easy to accidentally drain the users battery?
The UI of this app literally only has a piece of text, a button and this animated indicator. To render this it needs an entire chromium browser loaded into memory, and still it has performance issues.
I will always prefer software written in native graphics toolkits. I understand the benefits of using electron but the only electron based software I've ever liked was vscode.
> Instead, we can create the illusion of a changing height by using two rectangles, applying translate to each.
Its a very clever solution and props to the engineer, but this being the fix makes me truly despair at where we are as an industry around web UI. That html and css won despite these sorts of counter-intuitive horrors.
UI layers that make me feel good reflect intent. I can take an image and write some code to darken that image (any image) and show that to the end user. It makes sense. However, in html+css I have to introduce a third element, another rectangle, slap it infront, paint it entirely black and set its opacity to something low. Sure, it works the same but it just feels so conceptually ugly.
I had a similar problem trying to animate a book flipping pages. At the half way point I had to update the text to show the backside and update the back pages.
However no matter what I did, I could not get the update to sync correctly with the page being exactly half way. There's probably a solution but I just gave up
My head is sort of reeling from this. If height animations are that expensive today, imagine how expensive (and still commonplace) they were twenty years ago.
I fortunately quit animating height quite a long time ago in favor of similar transform techniques, but… wow, still crazy to learn the magnitude of this common operation.
You can animate an absolute positioned div without it obviously altering what a normal person would call "the layout." That's not simple unless one is willing to dig very deeply into the how web browser rendering works. And really still not simple. Otherwise this article wouldn't have been written.
The heart of the issue isn't that animating height is expensive, it's the downstream effect of that animation repeatedly changing the page's layout and forcing redraws.
I actually did quite a bit of browser animation 20+ years ago. The trick was (as mentioned elsewhere on this page) to only animate absolutely positioned elements. Even back in the version 4 browsers you could animate hundreds of divs at 20+ fps on a decent computer (with JS of course — CSS animation didn't exist yet).
So, the CRAY-2 built in 1985 was rated at 1.9 GFLOPS. The M2 the author uses seems to be benchmarked at around 6GFLOPs per core [1]. So this 6% of the author's CPU (assuming charitably this is a single core) corresponds to about 20% of the mid 80's peak supercomputer capacity.
People were already using those computers for applications that go slightly beyond animating 5 x 3 green little bars up and down at the time...
I understand dev time cost > CPU time cost (for the company that hires the developpers anyways) but aren't things getting a little out of hand?
Even without comparing to an 80's supercomputer, what if someone with a 15 years old laptop tries to use their app? What will those 6% CPU on the dev's shiny M2 MBP become?
I did not find other comments like this in this thread.
6% M2 cpu (even single core) is a huge computing budget for such a small feature.
At this point I don't really understand why OP seems happy with this result (sorry).
I think even a naive canvas implementation can really and quite easily cut most of this computing budget.
Also, a pure css animated thing should use mostly gpu in a right DOM implementation. I think some issues remains in this result.
Maybe OP displays it's app on a 240hz external screen which makes browser compute DOM animation at 240 fps requiring slighly more compute in a passive way (4o-mini suggests this, I am not sure browser works like this)
I remember too much of my pentium 1 166Mhz let run entire age of empires 2 game in 1997
I read this article because just last week I set up CSS height transitions to show/hide divs in a form which displays conditional content (based on radio button selection). "Why shouldn't I do this?", I thought.
It looks like the point of this article is that you should avoid continuous animation for CPU performance reasons. These performance reasons are probably inconsequential for occasional transitions.
For many of us in the target demographic of "people who animate height", the scary title of this article is misleading.
I created a little CSS-pulsing 'live' button and give it a fixed number of cycles of pulsing to cap worst-case resource consumption. But it took me a while to work out that that was the right thing to do!
How about we create tools that don't consume 5+ % of user's CPU (and keep it busy, not allowing it to sleep, decreasing battery life significantly) when idling or doing some background work?
>On my M2 MacBook, the renderer process is now using 6% CPU (down from 15%), and the GPU process is now using 6% CPU and less than 1% GPU (down from 25% and 20%).
I honestly can't imagine typing this out and patting myself on the back rather than thinking "wtf am I doing? It's 2025, I have a high end laptop, and this is a note taking app. The total resource utilization should be close to zero."
So good to see a developer who actually cares about resources. Thank you!
I usually develop my apps on a 1300 MHz single-core Intel Atom mini PC from 10 years ago. When something like this goes wrong, the app becomes unusable and I immediately notice it.
Yeah author congratulating themselves at the end because their note app is only using hundreds of millions of cycles every seconds, to do essentially nothing...
Yes! And the electron app 90% of the time can't work offline anyway (or often would be useless even if it tried). So it's just an extra 2 gigs of space wasted on my SSD for another copy of Chromium, instead of having the website be an "App" installed in Chrome or Edge and granted notification (etc) permission.
In my opinion, most things shouldn't bother to make an "app," and certainly shouldn't try to push their apps on me, unless they want to make an actual native app for their target platforms.
> In my opinion, most things shouldn't bother to make an "app," and certainly shouldn't try to push their apps on me, unless they want to make an actual native app for their target platforms.
It's as people have just forgotten that software used to integrate with the rest of the computer. I know people who install things like Netflix as apps just to have an icon on the taskbar.
Would the taskbar people be older people? There's something to be said about having a quick common place to get to the things you access the most. To that point, there are many many variations of launchers. I really wish webapps had taken off more, or having website create custom shortcuts for the various OSes so they can something like their red letter N logo on the desktop or wherever that just takes them to the website with a click rather launching the blue letter e then have to type something into a location bar or find it in some list of favorites. However, I think the sweet data analytics that apps allow companies to collect is too tempting and the web link just something to avoid.
I no longer have the app, but Slack uses <500MB of memory and .2% CPU in a tab in my browser. Back when I had the Slack electron app it was way more than that in use of system resources. I don't know what else to tell you. I don't know what other documentation would be interesting other than my observations made on my own equipment
This helped explain why my PC uses so many resources "with just a few webpages open". I didn't realise that graphical updates that seem so simple could be so resource-intensive.
The worst modern invention. HTML/CSS is so bloody awful at rendering user interfaces. We should if at least contained it, if not killed it outright. But instead it’s a contagion that has spread far and wide.
How the hell does the optimized version use 6% of CPU? It should render at roughly 5000 frames per second.
I'm not 100% convinced. Likely because the article doesn't show the difference between animations that are announced with 'will-change' and/or elements that are animated with position absolute. Also, it looks related to the browser and there was no evidence that this is also a issue in FF or Chrome.
> On my M2 MacBook, the renderer process is now using 6% CPU (down from 15%), and the GPU process is now using 6% CPU and less than 1% GPU (down from 25% and 20%).
This still feels way too much compute for a tiny animation updating a couple times a second.
In my wimpy work Virtual Desktop, this page eats 25-40% of my CPU just by scrolling up or down after fully loading. It's a severely limited VDI, I agree, but this performance is crazy for what is functionally a static page.
It certainly doesn't give me confidence in using their app, given this landing page can become as heavy as GMail while doing nothing. How optimized is their electron app? Is the iOS version as heavy? Are they OK with wasting all this energy for nothing?
> On my M2 MacBook, the renderer process is now using 6% CPU (down from 15%), and the GPU process is now using 6% CPU and less than 1% GPU (down from 25% and 20%)
Ouch. All of this for a tiny visualizer animation.
This article doesn't apply to every single use case and should not be taken purely at face value.
A few week ago I needed to animate height for a banner like message that appeared in the layout (e.g. it was not absolutely positioned) once a user clicked a button. This banner message would cause a layout shift no matter what because its height was being added to the layout and shifting the elements underneath it down. Thus, to mitigate a jumpy layout, making the height animate made the layout shift easier on the eyes.
And before you say "well design it differently" - I didn't design this.
An alternative solution to this is to place the height animated objects inside of containers with a static vertical size that is equal to the maximum height of the animation
here's an idea - don't animate anything on a webpage.
Just. Don't.
NO ONE needs animation to convey information, unless that information is an animation, such as in an educational context, where interactive things demonstrate what is being taught.
I don't need to see your widget spin in 3d while I scroll down the page. I really don't. I don't even want to see that.
Web pages are slow because people make them slow. Stop making them slow.
I'm not sure how it's implemented on the JS side, but I can imagine the CSS getting a lot of updates per second as the voice level changes, constantly interrupting the current animation.
Might be worth it to buffer the input until animationEnd() event fires. That might reduce the amount of calculations and redrawing that needs to be done.
Or the CSS engine is robust enough to handle continual updates and this won't solve a thing. I'd have to test it.
CSS animations can sometimes cause weird performance issues or even crashes. 10 years back, animating box-shadow caused browsers to crash for our some of users by consuming a lot of memory: https://www.sheshbabu.com/posts/my-binary-search-debugging-s...
Does anyone have the courage to perform a similar inspection on the Cloudflare homepage like the author did? On my M4 Mac mini, Safari lagged with single-digit frames per second (FPS), which scared me so much that I had to close the page as quickly as possible.
Not just height, anything that causes continuous repaints can become expensive. Search for "jank free" to find more resources on the subject, or check this oldish website [1].
I’d be interested to see the performance of the same visualizer in a canvas or animated svg instead. Forcing these animations through the dom seems a little absurd to me, but I’ve been surprised by web performance before.
Did throw together a quick SVG example - on my computer, a dummy page with just this animation spends 1ms rendering, so basically a solved issue - and cleanly.
My personal call would be to use SVG for this - you can do CSS animations there just fine without triggering layout. Not a frontend expert but don't see why that wouldn't work.
Is it just me, or have a lot of modern startups lost the plot? Imagine doing all this work for an element of your app which has nothing to do with its feature set. No one cares (certainly no customer) about an audio visualizer bar animation.
The post is also riddled with all kinds of misunderstandings, as well. For example, the authoritative language used to describe the differences between layout/painting/compositing is just simply untrue generally speaking (though it might be w.r.t. Electron/Chromium). The W3C does not care how you implement your rendering engine, and even bringing up the spec betrays a misunderstanding of how these things work.
In typical fashion, we see yet another grandiose blog post about fixing a "perf issue" for your hipster new startup that is essentially notepad.exe. Who funds this shit?
Why distract users with giving an auxiliary feature so much prominence? Why wasn’t a GIF enough? Or just an SVG that consists of 3 frames?
Spending this much time on fixing this bug was worthwhile (at least it created a HN front page article) but the solution was not. 6% CPU usage is horrible.
Why do browsers even allow this shit. I was trying out a scheduling app the other day, and it was using 3% of CPU just sitting there doing nothing. That’s a Pentium 1 level of computing power to do jack shit.
The fact that this little animation can have such heavy performance impact tells us one thing: CSS design is flawed. That‘s it. That‘s the whole story.
How can the browser know if an arbitrary animation can change an element's height? I'll bet 10 bucks this is equivalent to halting problem.
Unless you 1) don't support full SVG spec or 2) separate SVG elements from the layout completely, it's not a problem that is solvable by browser. The developers need to wrap their animation in a fixed size element and it's very reasonable to expect them to do so.
Plus, the CSS spec doesn't prevent browser vendors from optimizing this specific use case if it's really that needed.
(Of course in the case of this simple web app in OP, the correct solution is to not use animation at all.)
This looks like a clever implementation of a useful feature. I enjoyed that the developer was able to link the streaming audio API to a useful visualizer, recognize problems with his approach, debug issues in that visualizer, and find a clever solution using graphics fundamentals like masking.
I'm less than impressed with the general consensus that he's somehow negligent for launching a feature that needed a fix, or that users don't want or need feedback about audio connectivity, or that the poster did something much better sometime back in `02.
I think the idea is to have the animation respond to the actual volume in realtime. The animation property is only so that the 10 updates per second get interpolated according to the framerate, I think. The real solution would be to remove the gimmicky animation and show a red circle.
Great writeup. With LLMs doing an increasing amount of the coding now, it would be great for the browser or development environment to have built-in validations that enforce good performance. The coding agent (or human) would get direct, immediate feedback at development time that there's a performance threshold violation, at development time.
This looks like a very typical audio recording UI element. The bars are animated to the volume of the recorded audio. Helps the user see that the application is picking up their mic.
It's a digital analog of a VU meter on a mixing console, though the UI has morphed over time. (Like that the lines vertically centered, vs. being on aligned bottom like you might see on a real-world meter.)
… because it doesn't need to be bigger…? The purpose is to convey that "hey, the audio equipment is working". I use these all day to debug things like "am I not recording, or is the other side's volume muted?" We need "videoconferencing" or "notetaking" app levels of accuracy; it's not a sound studio app?
Not to mention how much dB as a unit freaks people out.
... Why does a "note-taking app" have an "audio volume visualizer"???
Edit: kinda funny how I asked what I thought was a reasonable question and expressed entirely understandable confusion; got an upvote almost right away; then got multiple downvotes after people answered the question — as if it somehow suddenly became obvious why I should have expected such a feature a priori.
Man, I hate what the web has become in 2025. There is absolutely no respect from frontend developers for their users. It's obscene the amount of wasted CPU cycles, energy, battery life, bandwidth, and time we have in aggregate due to horrible frontend development practices.
Here's the part that makes me confused/angry: this is a flat style icon with maybe 4 or 8 colors displayed on a static flat green background. I could have built this 25 years ago with a single GIF with about 6 frames, that was a couple of kilobytes and would be performant on the (relative) potato computers we had then. With CSS you can easily make that GIF a background image and position it correctly in the <button> or whatever. Do it at 2x and scale it for retina sharpness.
But modern "web designers" feel the need to spend how long carefully crafting that CSS animation which adds dozens of lines of code to the codebase and burns a ton of CPU...why? Just because you can, I assume? The same reason the same people use embedded <svg> documents inside the HTML for simple icons like "edit pencil" or "close", wasting bandwidth instead of at least putting them into asset files that can be cached.
Frontend web "advancements" of the last 10-15 years, at least the way they are used, are mostly cancer in my opinion. I will allow for the usefulness of display: flex, and that's about it.
The app taking "only" 6% of a $3k laptop's CPU to show two lines of text and a little 2D animation also gave me pause, but maybe this is just how it is. Still I wonder how people were squeezing pacman out of prehistoric machines, and we're using an 80's supercomputer worth of CPU power (not that far given they were in the ~ 1 GFLOP range) to animate three green bars today.
They actually knew how to code back then. I fancy myself someone who cares about performance and take a great effort to not do wasteful stupidity like those CSS animations for a decorative element, but compared to 80s programmers, I'm actually a complete fraud, because I do everything in interpreted languages that do 10,000 things behind the scenes for every line of code I write. I could never make an Atari computer or NES work.
There are of course a few who do, they all either work on embedded systems or making homebrew for retro computers. So much respect for that level of skill!
I aspire to be one who cares about performance, but I would like to address:
"I do everything in interpreted languages that do 10,000 things behind the scenes for every line of code I write."
Is it even possible to create a modern website, or product, without this? How long would the development cycle take, if everything is to be written in C?
I'd wager that replicating a "minimalist styling" for a react-based website with a dozen or so components would take dozens of times longer to produce.
> How long would the development cycle take, if everything is to be written in C?
Once upon a time I worked on a website backed by C. Development times were not appreciably longer.
We had a in-house templating language for generating HTML (also written in C). That implementation got to the complexity point to where even the original devs did not want to touch it to add anything.
But in terms of add a new screen collecting fields x,y,z it was fine. This was a job board, allowing search, applications, saved resumes, bulk opening uploads, i.e. there was some real functionality there.
If this is the case, I do wonder why we do not see more of this.
Were you around for the original design phase? Why was C chosen? Did you and yours collect performance metrics to see if anything meaningful was gained?
C on the modern web would terrify me. Devs are bad enough with interpreted languages, I can't imagine most of them using a language with a howitser sided footgun.
In particular I remember one bug where a pointer to a local variable was returned by a function. It worked on our development environment which was apparently more forgiving than our production environment.
Why were they different? Because they were actual servers and the production one was the expensive DEC Alpha while the development server was something less expensive (my memory fails me as to the exact machine).
I didn't mean to suggest it would be practical to do the practical stuff people do today with Java, Ruby, Python etc -- web stuff or "app" stuff -- in say, 6502 assembly, or C. Just that doing a good job of the stuff contemporary to those in those languages (especially the 8-bit era stuff) is a tour de force in programming talent.
Another way to look at it, is only those who could write very optimized code could get anything meaningful running on those machines. Hence all (?) software from far enough back was optimized down to death.
I'm personally grateful I've never had to wrangle hardware bugs into features and other such hacks one stumbles upon in anecdotes from those eras... even though there is a certain charm to it all.
That may be 6% of the lowest clock speed of an efficiency core, as I suspected when I saw a similar figure for a very tame CSS animation on my MBA M1. In the end I capped the duration of the animation to cap resource use unconditionally.
GIFs don't scale well (including 2x version is not enough); with CSS (and SVG) you get crisp graphics at any scale. And of course you can use asset SVG files that are cached [1].
That GIF weighs one third of a megabyte. You could literally fit a fully featured bespoke 2D/3D rendering engine in the same space. Now, it could probably be optimized, but an aPNG would likely be vastly smaller. And these days aPNG is even reasonably widely supported! That said, an animated SVG would likely be the best option.
What is your point? The random first GIF I found is enormous in pixel size and many frames, thus large in filesize, but a few dancing green blobs would be tiny. You don't need that one's 640px size to scale, and 2x will work perfectly fine.
I’m looking at that on an iPhone, and if I scale (in either direction, pinch or zoom) it I can see a semitransparent halo around it, I would assume from the antialiasing.
You really don't need an alpha channel for some tiny lineart thing such as the article's example. A single color would even be perfectly sufficient to make an attractive 'dancing bars' animation to indicate "sound is happening." That's my point. Some overpaid "UX designer" decided something wildly complex was needed here, and they're wrong.
In the broader sense, especially in the 'flat design' era where we don't use gradients or complex backgrounds anywhere, a 256-color GIF with its 1-bit alpha (plain transparency) will work perfectly fine to do an animated flat non-3d, non-photorealistic icon that wiggles or whatever, on a background that is almost certainly a plain flat color, and not complex enough to need 16 million shades of transparency.
Why the embedded SVG hate? A simple icon should be just a few hundred bytes, is reusable, and styleable with CSS. It's not the right fit 100% of the time, but I think it's a very nice feature to have.
They're generally static documents that could be served independently, and thus cached, like an image.
Many times, though, they're served up inline in the JavaScript code that templates the html, and thus is cached less aggressively, especially if the FE is deployed frequently.
Ok, I want to add a pencil icon here, the pencil should be #ccc, here is #ddd, there is rebeccapurple, and over here it should be the same color as the container color.
Please write a nodejs server that produces those tiny images, don't forgot to correctly set the caching headers...
Yeah, this is what I was referring to and is my main use for embedding SVGs.
For really simple ones, the size of the icon isn't a whole lot bigger than a CDN/object storage URL. And if you reuse it a couple times, it might even be a net bandwidth reduction.
Honestly, when people are making use of CSS to "reach into" the svg dynamically, or like your example, to color it dynamically, I'm not opposed to the strategy.
99% of the inline <svg> usages I see are people using js includes to inline an SVG which is itself stored in a .jsx file. This is just as lazy and wasteful as it would be if Styled Components were built not to generate classnames and a single stylesheet, but instead with all the rules simply inlined in style attributes on every element.
Wasting hundreds, well maybe a thousand of bytes in the process ! (I understand it's better to cache and things can go out of hand fast. But we're talking about small animations here.
I don't disagree at all. There's way bigger fish to fry when it comes to maximizing network efficiency in websites.
I'm just pointing out the rationale of OP since someone asked, and I've worked with people who've made this and similar arguments before. It's been awhile, though- the ones I'm thinking of would treat code golf as a best practice for CSS (only slightly exaggerating).
Believe me, a few kb on every icon on every pageview adds up to a lot of wasted bandwidth and not everyone is on a gigabit connection at all times, even fancy iPhones are on shitty connections a lot more than you think. Inside parking garages, on deprioritized LTE, etc.
> I could have built this 25 years ago with a single GIF with about 6 frames, that was a couple of kilobytes and would be performant on the (relative) potato computers we had then. […] But modern "web designers" feel the need to spend how long carefully crafting that CSS animation which adds dozens of lines of code to the codebase and burns a ton of CPU...why? Just because you can, I assume?
The animation (which is not suppose to be constant/static) conveys whether audio is being picked up by the application. I.e., it's VU meter.
I can't use the app in question, but I haven't seen any proof it doesn't just "dance when receiving audio". If it tracks amplitude and pitch, I'd still say it's overengineered, since in practice no one can discern anything useful except 2 or maybe 3 states (silent, quiet, not quiet).
True, the article didn't say that it for sure isn't being used as a glorified GIF, but … why not just give the benefit of the doubt that it UI element is doing the thing the UI element ought to do? Then, "Here's how that turned into a performance problem" followed by the rest of the article.
> I'd still say it's overengineered, since in practice no one can discern anything useful except 2 or maybe 3 states (silent, quiet, not quiet).
For most use cases … that's all that's necessary? It helps confirm that the audio equipment and various software layers are working.
… like I use these UI elements all the time, because Bluetooth has decided that "nah, I'm not going to connect, even though every device reports as connected".
We agree that 2-3 states is all we need to discern -- I was just saying that if it's actually wasting even more CPU by truly analysing both the amplitude and frequencies in the audio, I'd chastise whoever decided to do that for wasting time when "sound" "quiet sound" and "loud sound" serves 100% of the use cases for this widget.
And the above is a separate question unrelated to implementation. A true VU meter in CSS animation is the most overengineered, a glorified GIF built as a CSS animation is somewhat overengineered, and a couple of simple GIFs is the simplest.
I think it's funny that you think sites can't be attractive unless they include many megabytes of complex JS and CSS on every page. Craigslist lasted and was dominant a whole extra generation of the Web until Facebook Marketplace unseated it without even bothering with colors and certainly no animations. And FBM didn't win on clever animations and React UI complexity, it just won based on the fact that it was easier to detect Nigerian scammers since you can see when the buyer's FB account was created and a few other cues. Users don't care how clever your programmatic animations are. They hate how slow your site is. Look how much Jira sucks now. It's more sluggish on an M4 Pro than it was on my Core i5 in 2013 because of how much frontend bloat they've added in the meantime.
This contempt is misplaced. It is obviously better in one sense to generate an animation on the fly using a short program instead of encoded its literal pixels in a much larger amount of data even after compression.
The ecosystem has largely moved towards that abstraction because it scales better for virtually every purpose, particularly for rendering in different formats.
Now web developers are brought up inside the abstraction and so don't even think to go outside of it by default. Which would be fine, except that the abstraction isn't designed very well: the browser is massively underpowered at building applications in an foolproof way. Part of that is historical cruft and part of it is (imo) a lack of vision--there should have been a (not-backwards-compatible) replacement for html/js/css years ago which solves its fundamental abstraction problems.
But anyway the present is still much better than the past. You must be nostalgic for the era when all monitors were the same size and websites weren't expected to do anything without reloading the page.
The problem is that everyone (browsers + developers) need to agree on a replacement, which is notoriously hard to do. Chrome got laughed out of the room with Dart/Flutter.
Well they don't really. What ought to happen is someone just makes a new browser (or adds support in an existing browser) unilaterally on an experimental basis. If it's good enough I imagine it would catch on. The limiting factor is presumably that that's a lot of investment for an experiment, but getting anyone else on board without a successful experiment is going to be impossible. (well, also, you have to have a good enough plan to be worth switching. I don't think dart/flutter was that.)
A nice starting place would be if one of the major browsers added an easy to use extension point for swapping out languages. (at least I'm not aware of this existing, nor that I've looked)
> A nice starting place would be if one of the major browsers added an easy to use extension point for swapping out languages. (at least I'm not aware of this existing, nor that I've looked)
What would that actually mean in practice that would be different from how Dartium did it?
Well I don't know how dartium did it really, but it seems to me like "add a new website implementation language to Chromium" would be a tutorial on Chrome/Chromium's home page.
Such that one could throw together, I dunno, qqrl.tk/index.z and see if they can do a better job than HTML/CSS.
As far as the actual solution to try for, table stakes is a native implementation of something that looks like React (but presumably reimagined from scratch).
> Such that one could throw together, I dunno, qqrl.tk/index.z and see if they can do a better job than HTML/CSS.
What would be the use of that? It would help people who want to make prototypes of a different web with no hope of it ever getting adopted, but that's not a huge market.
> table stakes is a native implementation of something that looks like React (but presumably reimagined from scratch)
Like a component layer? Browsers have tried with e.g. XUL or WebComponents, how would you avoid the problems of those?
So people can experiment with it. Creating a good place to experiment is the best way to make lots of good ideas happen. If 10000s of people can play with ideas instead of the 10 groups with enough time and energy and expertise, you get a much richer exploration of the space. Heck if it was easy I'd be doing it right now.
> Like a component layer?
I'm not aware of anyone creating a component layer that is even passable for abstraction compared to React's programming paradigm. Usually because they stay imperative-first when all signs seem to point to declarative-first being the more scalable model.
> how would you avoid the problems of those?
Umm... they're terrible, so I would just not copy them? Look at XUL for seconds and try not to vomit. WebComponents adds abstraction of HTML in Javascript, rather than in HTML / declaratively, which is obviously never going to work as a foundational technology. It is far too imperative-first. I would start by copying React... but then doing everything differently because most of React's design is forced by being implemented in JS and having to work around CSS being quirky. The greenfield version would do a much better job.
Dart/Flutter was never a technical issue, but a political/monopolistic issue. The underlying monopoly has not changed, so I don't really see a path forward unless the chromium team magically changes leadership.
I guess my point is, if the big Chromium team couldn't pull something like this off, who else would? At this point major web updates tend to get born in one of the big tech companies funding web development, either directly or indirectly (Mozilla being funded by Google)
I personally don't think dart/flutter was good enough of a solution to be worth everyone's time. When this actually happens it will be more of a paradigm shift than that.
That's why (which I said in the other comment) what needs to happen is that innovating in this space needs to become a lot easier.
An engineer could build that much faster than it would take to build an animated GIF in an editor. (For a start, it's a much more complex animation than 6 frames.)
Does it need to be more than 6 frames to serve its purpose? It's 3 wiggling bars to be stuck into a 1/4 inch spot inside a button to serve a purpose of communicating a boolean state (sound happening or not?). The artist who actually designed that probably spent an hour plus getting it just right, wasting a ton of time before the engineer even wrote the "clever" GPU-destroying CSS, and then he had to rewrite the second version after discovering how resource-wasteful the first version was.
An icon served from an <img> is going to be served from browser cache (disk or even RAM) after the very first time it's encountered. Unless you're building something designed to be used for one visit ever (maybe a promotional landing page?) it's better to have a few more requests (HTTP2 especially makes this a non-issue anyway) and have a cache full of assets than to make your dynamic webpages bloated.
That tab is using 4-5% CPU on my M1 Pro, and I also have a "GPU process" in my task manager, which is showing about 8% in the "CPU" column when it's onscreen. So, no, using a 64x64 GIF (1/100th the area of my example) you'd need to do this animation would absolutely not use any significant CPU or GPU. Which is why those worked very effectively to animate small icons with zero lag on the potato computers we had in 2000.
ironic that the benefit of using web tech (platform abstractions) can be totally negated by such a small footgun, which then requires insane knowledge of the browser rendering pipeline to solve
One problem is that this is supposedly considered insane esoteric knowledge, when the concept of reflows, and why and when they're triggered, ought to be fairly basic knowledge to every reasonably experienced frontend coder. Up there with "O(n^2) may bite you in the ass". But we've decided that optimizing SEO or conversion rates or whatever is much more important than optimizing rendering.
I think there's a bit of an "everybody knows that" [1] phenomenon when it comes to knowledge like this. Devs come from different backgrounds and work on different types of projects. There are 10000 things you expected to know to be an expert, and all of us are continually learning.
Except that this is the developer of an apparently popular and well-invested product.
If you expect to make a note taking app, which only has a piece of text, a stop button and a floating animation, I hope that you figure out why your performance is being tanked by your single animation.
> insane knowledge of the browser rendering pipeline to solve
Saying it having not much of that knowledge myself: it's reasonable to sometimes need platform knowledge to make intricate things (here a realtime animation) behave efficiently.
That's the same for game engines, native rendering, flash animations etc. There will always be edge cases, building enough stuff you'll hit them, and if you care about doing it right it will require knowledge to overcome the issues.
None of this should be esoteric knowledge if you work with web technologies for a living. Animations, box shadows and blurs have a very long history of performance issues. Similarly, standard tools easily highlight the issue, as shown in the article.
I've seen the "This webpage is using alot of resources" popup before but I don't think it would happen in this case.
Because honestly I think this is horrifying. I would rather it switch from grey to red "recording" dot than use even the 6% the author decided was "fixed". In 99% of cases I do not care at all about the "artistic vision" of the UI designer and in the other 1% of cases (say an in-browser game or some useful data-viz) I could choose to allow the tab to go crazy with my resources.
reply