"Too Many Requests"
- https://web.archive.org/web/20260531130034/https://jbkempf.c...
'AV2 decoding is roughly five times more complex than AV1 decoding. In practice, that means software running on today’s hardware will struggle to decode AV2 in real time without careful, architecture-specific optimization'
AV1 software decoding is already very intensive so AV2 decoding benchmarks are the next thing that would be really interesting (or mortifying) to see.
Intel's Arc dGPUs were really compelling for dedicated AV1 encode and decode, especially the small form factor of some cards. You could even fit it as a secondary card in a PC dedicated to recording and encode workflows for OBS.
Hope we get a similar option with future lineups that support AV2, especially given how popular video creation and streaming are now.
Is there a compelling reason encoding needs to be done locally?
The point of encoding is to reduce downstream bandwidth for the viewer, and upstream bandwidth for the distribution network.
The content creator only needs to upload it once.
Yes.
An uncompressed 1080p, 60fps video with 24-bit color depth would need around 3Gbps to be streamed. And even if you don't need to stream it, that would still consume a sizeable portion of the write throughput of the fastest SSDs currently available; if you go up to 4K, you'd actually exceed that by a lot (not to mention, 1tb of storage would last for about 10 minutes of video).
Who is regularly watching uncompressed videos outside of production environments? That’s got to be a very small population.
Using raw uncompressed bitrate is a bit disingenuous. How about comparing an older, widely supported codec like H.264 as a baseline?
If you don't encode locally as the video is created, you either need to store RAW frames which takes enormous amount of storage, or you use a different format and suffer quality loss by transcending.
> you use a different format and suffer quality loss by transcending.
Compressing to AV1/h264/265 etc is really only done for the final version, but that doesn't mean that videos are stored in RAW format during editing, where it is very common to store frames locally in Apple ProRes, Avid DNxHD, or some other compressed format that's targeted towards professional editing.
Contrary to AV1 or whatever similar format which offer compression ratios of 1000x and more, these formats have a compression ratio of around 10x. They are very simple, and the quality loss is low enough that it doesn't matter. They also tend to store images with 30 bits per pixel instead of the 24 bpp that's normally used for streaming.
Well yes? The platforms only accept certain resolution/bitrates and also most of America isnt running 1gig up. They're running 5-30 mbps up. So yeah they need to encode it.
Data caps make that hard. While everyone likes to claim unlimited data, I'm not aware of any providers that don't have a heavy data user clause where they'll deprioritize your data if you're a top ~5% data user (usually somewhere over ~1TB/month).
You also will need _some_ sort of encoding locally before uploading, even if it's minimal, which could lead to issues when encoded again (although there are codecs available to minimize this).
Video calls & streaming.
this
for other cases, I can just wait more for my cpu/gpu/cloud to do the job
I came to post this as well. Until widespread, inexpensive hardware catches up to a 2018 codec, AV# will remain a niche ideal.
Hardly niche. My laptop isn't new and it has hardware AV1 decoding and encoding. My 10 year old iPhone 7 can play 1080p AV1 video in software for over 200 minutes with VLC. The iPhone 7 was released in 2016, a year and a half before AV1. The dav1d decoder is mighty.
Netflix uses AV1: https://netflixtechblog.com/av1-now-powering-30-of-netflix-s...
YouTube uses AV1. It's tough to be more mainstream than that.
Right click on a YouTube video and select Stats for Nerds. If your system is capable of it, chances are it will be playing back in AV1.
Most of the YouTube videos I watch these days are AV1 encodes. Sometimes it's in VP9 and occasionally it's H.264.
Supported is different from doing it well though. You do notice the performance hit even on TVs that playback YouTube videos on AV1.
Even on 1080p videos running on AV1 on 1x, the TV system bogs down and any kind of interaction has a variable 1-3s lag. On some TVs if you do 1.25x the TV automatically "downgrades" the resolution to 480p to avoid dropping frames.
I wish there was an option to still use VP9 / H.264 on those systems (even limited to 1080p).
More reason to never use the builtin stuff in a tv. Cheap sticks can handle decoding fine.
Youtube artificially limits the resolution, on mine if you cast the exact same video it doesn’t impose that limit and works fine.
Yeah I could imagine the AV1 codec sticking around for a very long while, even as a fallback for AV2. There's still hundreds of millions of people out there using old/cheap devices (especially in developing countries) where that battery drain from software decoding is a big problem, so AV2 would be nonviable.
Some of the early use of VP9 and AV1 was Netflix serving video to people in developing countries. Their metered bandwidth was more of a bottleneck than the CPU playback.
Same. Mostly AV1, sometimes VP9, and rarely h264.
What's missing mostly: live streams which are h264.
Currently, and I say currently, dav1d is so fast, no worries on that side.
> AV1 software decoding is already very intensive so AV2 decoding benchmarks are the next thing that would be really interesting (or mortifying) to see.
Yes, this is going to be fun to watch.
A codec spec isn't done until there is at least one decoder developed in the field. So reference + 1. The field implementations often become the de facto spec.
Reading the MPEG1 specs back in the 90s as a child opened my eyes to how to define complex systems. For a media coding standard, they spent most of their time saying how to interpret encoded bytes, which I realized is genius. Be descriptive about decoding and you don't have to be prescriptive about encoding. Encoding is where you can apply all the creativity, but you need to provide a way to have a shared understanding of the encoded bytes.
If decode is becoming so complicated and expensive the hardware can't handle it, why not just go full neural, send latents, and run decode on tensor cores?
The answer is probably the same as for why not AV2 everything; a lot of hardware couldn't support it today. But in 10 years?
It seems we're running up against fundamental limits of human-engineered video codecs at this point. There might be a lesson in there.
What makes you think that would use less resources?
And it's not really hardware hitting limits, it's specifically software decoding on somewhat weaker machines.
This is a software decoder designed to run on general purpose hardware.
Adding custom hardware like tensor cores to the stack would serves a different use case.
I'm not quite convinced a 25% reduction in size is worth effectively obsoleting all devices that have hardware decoders for AV1 but will struggle to decode AV2
Modern video services perform multiple encodings with different codecs, bitrates and screen dimensions, and serve up the most appropriate format that the client device can decode. Youtube has hundreds of format variants:
https://gist.github.com/MartinEesmaa/2f4b261cb90a47e9c41ba11...
Devices with AV1 hardware decoding - rare as they are - won't be obsoleted for a long time.
A new codec doesn't obsolete old devices. At least, not right away.
Studios still release new dvds with mpeg2 video. Online videos tend to be available in many codecs. Video conferencing tends to negotiate to best available or has settled on ancient codecs and won't change quickly.
When you host videos with near 17 billion views you're going to want to stream those videos in as few bits as possible:
https://www.youtube.com/watch?v=XqZsoesa55w
That extra 25% becomes worth it.
Nothing will become obsolete. AV1 will stick around for a long time. And YouTube still does H.264 encodes to support old devices.
> The page you have tried to access is not available because the owner of the file you are trying to access has exceeded our short term bandwidth limits. Please try again shortly.
HN hug of death
I thought this was about Dave2D
Soon to be decodable with dav2d!
Yeah I suppose it’s named after dav1d but still seems like a pretty unfortunate name collision.
Exactly
same
Same
Seems like the blog succumbed to the HN hug of death (`Actioning this file would cause "jbkempf.com//blog/2026/dav2d/" to exceed the per-day file actions limit of 160000 actions, try again later`), is there a copy available somewhere?
it got archived by the wayback machine: https://web.archive.org/web/20260531115337/https://jbkempf.c...
Ouch, looks like the HN hug of death struck again. Gives me error 429.
... improvements around 25% compared to AV1
AV2 decoding is roughly five times more complex than AV1 decoding
I understood it as compression is 25% better : a quality of 10mbps in av1 can be achieved with 8mbps in Av2. But, it needs 5 times more compute power for this 25% gain.
> I'm not sure what these two lines mean or if we can compare them, any help?
AV2 saves 25% bandwidth at the cost of 5x more decoding complexity.
What does "complexity" mean here? Computation required?
Yes, much higher computation required to encode it, and decode it, both.
Encoding speed even on Mac Studio is atrocious, it’s in range of single frames per second as opposed to realtime+ for even h265
dav1d is the av1 decoder and it’s an insane feat of engineering. Written in assembly, it even eschews the normal c calling convention to get even better performance.
Unfortunately, in absence of inlining, compilers mostly respect calling convention even when they don't have to.
Yes
Smaller files but harder to decode
Related:
The AV2 Video Standard Has Released (Final v1.0 Specification)
> Make it fast on older desktop, by writing asm for SSSE3+ chips
I guess 5 years ago (around the time when Intel stopped making SSE-only chips) is technically "older", but I wouldn't prioritize avx2 when devices intended for consuming media definitely experience much less pressure to upgrade than workstations…
Almost every Intel CPU released since 2013 has AVX2 support. Some Atom SKUs were longer holdouts, but the fraction of x86 CPUs shipped in the last decade that have AVX2 support is very high.
What's next, Dav4d?
Dav3d?
I would love to see comparisons with AV1 on very low bitrates.
Return of the 8MB Shrek encodes?
https://web.archive.org/web/20210416200451/https://cdn.disco...
Shrek 1 at 8.34MB including audio.. insane
Video resolution: 128x72, hahah. Late 90s RealPlayer postage stamp video is back! To its credit, that whole movie is probably smaller than RealPlayer itself was.
There's a 64MB game boy advance cartridge with shrek on it [1]. Looks pretty horrible [2]. But the GBA only has 16KB fast / 256KB slow RAM, and a 16MHz CPU.
[1] https://archive.org/details/Shrek-Video-GBA [2] https://www.youtube.com/watch?v=CyOfPZQl4MI
I love this, hope to see a AV2 version at 8MB
6MB should be enough for everyone!
Not to be confused with Da4vid (world-class hacker and owner of the Black sun) or D4vd (rap artist and alleged murderer)
Or Dave2D, popular tech youtuber
This thread made me think of this song/sketch from Kids in the Hall.
Or dave, the command to start Dangerous Dave.
I don’t think the chances of confusion between a niche celebrity and a video decoder are very big.
> Not to be confused with Da4vid (world-class hacker and owner of the Black sun)
*Da5id
Is codex working on novel decoders 24/7? I hope
One would imagine given the name that it would specialize in codecs
Codex is actually short for `Code Execution`
When AV1 was first announced, I got the impression the name was chosen partly as a pun/reference/homage to AVI, the classic but outdated format with used to be popular. Then when I saw Dav1d, OK, good way to continue the pun.
But now with AV2 and Dav2d, that completely breaks. Are we eventually going to get AV3/Dav3d and AV4/Dav4d, which will read like Ave/Daved and Ava/Davad? Seems a bit awkward. Was the idea from the start to have the 1 be the version number, and have it specifically be part of the name?
I'm pretty sure it is a homage. As for dav1d it's not a reference decoder (although partially funded by AOM iirc) so they might not know that the next iteration will simply be AV2, we have h264, h265, h266 naming though
Tangent but I cannot wait for h269 (or h267 for the younger gen)
I think it's a reasonable decision. The only people who will interact with dav2d by name are codec nerds, and a simple increment makes the lineage more obvious to that audience.
As with all naming schemes in the tech world, I am sure no future scenarios, including successor names, were ever considered
1dav2codecs?
2av2furious?
And then AV3: Tokyo Drift, and after that AV Episode 1.
Or go the Apple Watch naming scheme route.
Just “AV”
Next, AV Series 1 and 2 (released simultaneously)
Later, AV Edition but it costs $10,000
AV360. AV365. AV2030. AVXP. AV8. AV10. Perhaps some here will be around for AV95.
Young AV?
Already predicting which versions to avoid, huh.
Da5id could potentially work as a Snow Crash reference.
> experience Dav... Now in 3D!
I’m fascinated by the flurry of downvotes to a simple commentary and question, especially when the replies are normal. If you’re one of the downvotes, please do share what you found offensive about my comment, I am genuinely interested in what you perceived as problematic.
[dead]
D4vd
Sorry if this sounds naive, but does it make sense to write a codec library in C/ASM considering how well Rust is progressing, especially when, as the author puts it, AV2 decoding is roughly five times more complex than AV1 decoding?
The algorithms deployed in these kind of codecs take into account not only human vision and mathematical laws of information, but also nitty-gritty details of how computers work, which are optimally exploited by directly having humans write detailed assembly rather than a compiler make a best guess and effort.
Encoder and decoder writers frequently need extremely fine grain control over SIMD instructions in order to get good performance.
The way they weave these instructions can be very hard to express with a high level language.
Further, there's a ton of work with arrays and importantly parts of arrays. They can, for example, need to extract every other element up to 1/2 the array. Unfortunately, rust has runtime array bounds checks which make writing that sort of code slower. The compiler can elade those checks, but usually only in simple cases.
The authors would be writing a bunch of unsafe rust to get the performance they want and rust makes that more painful on purpose.
I like rust, but C/ASM really is the right choice here. This is one of the few cases where rust's safety is a major detriment.
Performance should not be priority #1. Security should be. Why do we slow down all CPUs to prevent SPECTRE attacks yet continue to write in C? As rav1d shows, the perf loss is far less to migrate from C to Rust than it is to apply SPECTRE mitigations, and adding a sandbox around a memory-unsafe codec is going to be way more expensive again than using Rust code to start.
> Performance should not be priority #1. Security should be.
For a web browser, or a server in a bank, sure. For anything else, questionable.
> adding a sandbox around a memory-unsafe codec is going to be way more expensive
In modern world, overhead of strong sandboxes is surprisingly small. A nuclear but most reliable option is hardware assisted VM. On modern computers with SLAT and virtualized IO the overhead for most use cases is negligible. If you want something lighter weight, can use a multi-user nature of all modern OS kernels and isolate into a separate process with restricted permissions. Sandboxing overhead is approximately zero.
> As rav1d shows
rav1d is not a full rewrite of dav1d to rust. So it really doesn't show that. It's currently C + rust + asm.
I don't think we can say anything about what this does or does not prove about the performance of safe code.
> Performance should not be priority #1. Security should be.
Entirely depends on the application. The reason rust has `unsafe` is because there's some situations where performance needs to preempt potential security problems.
> Why shouldn't safety be the default?
Because safe code isn't fast enough to decode live video.
> If you really want to, it wouldn't be too hard to maintain a patch on top of rustc to drop the bounds checks if you want to compile object files without them.
Yeah, but then you are undermining safety in a critical way that does lead to security vulnerabilities (buffer overflow). And you are also now maintaining and requiring other devs for a project to use a custom version of rustc. That's certainly part of the reason that's simply not happened.
But another major part of it is that encoders end up with a lot of custom ASM regardless. That custom ASM is going to be where vulnerabilities end up. You don't really escape that by using rust.
If you are already abandoning where you critically need safety the most for performance, then why pick a language that additionally penalizes you for using unsafe constructs?
> Software decoding has a safety culture problem, and we need to talk about it.
Compilers and languages have an optimization problem that we need to talk about. SIMD optimizations remain a very hard thing for compilers to get right. We should talk about what it'd take to make compilers better and the reasons for why codec devs need to drop down to asm instead of using a high level compiler.
There might not be a solution to this problem, there are reasons for it.
Because it's 5 times more complex, you need to get the maximum performance available. Therefore more ASM than ever.
Rust does not bring more performance. Just more safety.
> Rust does not bring more performance. Just more safety.
Though more safety can in some cases bring a bit more performance. For instance, with Rust you can often avoid "defensive copies" of objects.
When writing a high performance video codec avoiding defensive copies of objects is something you want always, not just often.
C makes it easy to be fast but hard to be safe. Rust makes it easy to be safe but hard to be fast.
Also note that video codecs tend to wrap C or Rust around handcrafted ASM. Performance is king.
The safety can be worth it in certain cases. Like when handling untrusted input. And it's not just Rust: look at WUFFS for example. WUFFS can actually rival handwritten implementations in certain cases.
Are video codecs in the present day able to be sandboxed? In my fantasies at least I’d like the worst a malicious video file can do is cause garbage output or cause the codec to crash.
Forgive the ignorance, I have worked entirely in the abstracted layers of the stack, and mostly web.
not really. they're mostly pure assembly and sandboxing assembly isn't really a things
but not these cases
It really should be, though: https://en.wikipedia.org/wiki/FORCEDENTRY
WUFFS like approaches work better for algorithms like lz77 that are substantially bandwidth constrained. for something like a video codec, the computational intensity is much higher so you need better codegen to reach max speed
The ffmpeg devs have said many times in public that they routinely get speedups of 10x or more over C code. I'm not a reputable source on this myself but I highly recommend looking into their channels, mails, or posts.
yes it makes sense to use C/ASM here, but if you're curious, there is a rust port of dav1d named rav1d: https://github.com/memorysafety/rav1d
it's not much slower than the original C/ASM implementation (last i checked ~5%?) but that matters here
It's a Rust/ASM port, look there: https://github.com/memorysafety/rav1d/blob/main/src/ext/x86/...
I am not sure if it is that much safer than the C version when raw assembly is still required.
It is much slower than 5%, there were other independent tests that put it around 20%.
there's a rav2d now too fwiw — https://github.com/stukenov/rav2d same playbook: safe Rust + asm kernels via FFI. just shipped 0.1.0.
https://youtu.be/nepKKz-MzFM&t=7195
If you can stand Lex Friedman for a bit, the VLC authors talk about why you use ASM for a video decoder instead of pure C or rust.
fyi the Rust port already exists: https://github.com/stukenov/rav2d you keep the hand-written asm via FFI, rest is safe Rust. same trick rav1d uses.
Go ask FFmpeg what they're writing their encoders and decoders in.
That isn’t particularly helpful to someone asking a question in good faith. What others are using doesn’t clarify why they are using it. Plus, FFmpeg is itself a decade older than Rust. The OP is asking about starting a new project today.
> What others are using doesn’t clarify why they are using it.
It does if you ask them, or at least research the topic at hand.
Isn’t that just the same as answering “Google it”, then? We’re on a discussion forum, where matter experts visit, talking about a specific topic. If one can’t ask their questions in this highly relevant situation, where can they? The point of HN is supposed to be gratifying curiosity.
I don't know why you've been down-voted. It definitely isn't an optimal decision. A video codec isn't all assembly. There's plenty of plain unsafe C code. E.g. this is the first random file I clicked. It has a ton of raw C pointer stuff just begging to be exploited.
https://code.videolan.org/videolan/dav2d/-/blob/main/src/dat...
There is a project to write an AV1 decoder in Rust: Rav1d (really stretching the name here).
https://github.com/memorysafety/rav1d
They got within 5% of the performance of dav1d and held a contest to close the gap but I think I read somewhere that this wasn't achieved.
https://www.memorysafety.org/blog/rav1d-perf-bounty/
They claimed
> This is enough of a difference to be a problem for potential adopters, and, frankly, it just bothers us.
But in my opinion nobody actually cares about 5% in absolute terms. It's likely just Rust naysayers using that as an excuse.
I think the likely reason for dav2d using C is that they can reuse lots of code and infrastructure from dav1d. But I agree it would be much better if they worked on Rav2d instead (these names!). You can hardly complain about a 5% overhead if you're opting in to 5x more decoding complexity.
funny you mention it — rav2d exists now: https://github.com/stukenov/rav2d full C-to-Rust port, asm kernels still via FFI like rav1d does. early (0.1.0) but passes conformance against dav2d.
Of course any random C file is going to have pointers. Where can anything in the linked code be exploited? It seems like they're testing for bad input data with asserts to catch bugs in some functions, and properly validating bad inputs in others. Just because they're writing C doesn't mean it's vulnerable.
How can you claim nobody cares about 5%? A 5% performance increase is significant. And video decoding is not always for playback, where 5% may not matter as much.
Yes? There is 5x more code to optimize the ASM for.
Ok whose idea was ‘Wiener filtering’
Norbert Wiener in the 1940s. He invented the technique.
It's a semi-common last name.
standard in DSP. Super cool and can be used on stock trades and other predictive markets.
This seems like an interesting case to test AI agents on.
Like we had weird examples like C compilers and Bun. This is a much more interesting example because its highly nontrivial.
AV1 exists, Dav1d exists. Lets see AI take the AV2 spec and Dav1d code and try to make a working high performance AV2 decoder.
How is AV2 expected to avoid the patent-pool issues AV1 ran into?
AV1 was designed as royalty-free, but Sisvel’s pool and the recent Dolby/Snap proved the contrary.
https://accessadvance.com/2026/03/24/access-advance-licensor...
They filed a suit, henceforth making a claim of an issue...... They haven't "proved" anything other then they have lawyers on staff that can file some paperwork until the suit is settled in court...
How does that prove anything?
They're claiming that there are patents, but that doesn't mean there are.
Dolby is only the most recent case, Sisvel consorsium actually bills licences per device:
Consumer Display Device: EUR 0.32
Consumer Non-Display Device: EUR 0.11
(source here: https://www.sisvel.com/licensing-programmes/audio-and-video-...)
Sisvel allows you to pay them if you believe their claims, they haven't actually taken anyone not paying to court yet to prove that. The only court cases for VP9/AV1 from Sisvel so far have been their patents being found invalid/irrelevant.
Dolby is somewhat more interesting in that rather than scare tactics, media hype, and attempting to form a pool about it they are actually taking a patent assertion claim to court.
That crowd are just deeply concerned one of their lucrative revenue streams is disappearing.
So they seem to be attempting to pull a fast one and use unproven claims to try and convert their competitor into a replacement revenue source.
It'll probably be a case of whoever has the best lawyers + contacts + persistence wins.
But it'll be interesting if discovery shows evidence they know they don't have a case and are trying it anyway. "Piercing the corporate veil" can theoretically be a consequence of that AFAIK.
How does how they bill for their product, matter in terms of if their lawsuit holds merit?
That doesn’t prove their claims are valid.
I can claim the same and offer licenses per device.
Can you point to any other patent lawsuit over AV1? AFAIK the Dolby case is the first.
Sorry, I have a patent on questioning whether open source codecs are parent encumbered. Venmo me $1000 or you will be speaking to my lawyers
Every single AV2 news here in the last week has seen exactly the same question.
Either go back read the answers there first, or I will assume you are part of a FUD campaign (yes, I know HN guidelines, but again every single AV2 news in the last week has seen the same rhetorical "questions" as top "comments").
No codec can ever avoid patent-pool claims.
Should have used AV2
Too much traffic from HN?
``` Too Many Requests The page you have tried to access is not available because the owner of the file you are trying to access has exceeded our short term bandwidth limits. Please try again shortly.
Details: Actioning this file would cause "jbkempf.com//blog/2026/dav2d/" to exceed the per-day file actions limit of 160000 actions, try again later ```
I don't know if I'm underestimating HN's reach but I doubt we did that, probably traffic from a much bigger aggregator/forum
[delayed]
You are underestimating HN's reach, this happens all the time. As someone who has been on the front page of HN it's a pretty big rush in traffic!
I'd wager that the load is amplified by other sites that treat HN as a goldmine of tasty links.
Hacker news doesn't generate much traffic, despite what people are saying.
The host here has a limit of 160000 files served each day. That is extremely low. If the site has an icon, css, a js file and a few images it's 10 files each visit. That's will limit it to 16k visits/day. If there are more files loaded it might just handle a few thousand visits, and they have received more than that from HN now.
Well, if every asset request hits the origin (no CDN, no caching...) that would be a misconfiguration of the website. It should never happen for a blog.
A decade ago I was on the front page and saw ~16k uniques/hour I think?
Wait really? I'm not really sure what to think and I posted before I saw this... I wonder why the limit is so low?
Yeah, it's hard to tell thb, just a guess. But potentially the site also misconfigured their server, causing too much cache misses and hitting the server direclty.
i had that too once i used dyndns address my linux apache crashed when some one posted it here