AV1 codec version 3.5 greatly improves CPU encoding performance

Alfonso Maruccia

Posts: 90   +41
Staff
In a nutshell: Google just released AV1 3.5, the newest version of the video-encoding technology, which threatens to be a major player in the content and media industry. The codec now has much more efficient CPU encoding, and hardware support is growing too.

Just four years after its initial release in March 2018, the AOMedia Video 1 (AV1) codec is becoming a very appealing competitor to H.264 and HEVC/H.265 as the most used video standard in the industry. Hardware support is accelerating, and now developers have released a new version of the royalty-free technology bringing pretty remarkable gains for encoding performance in CPU-bound scenarios.

Google engineers just released the latest version, AOM-AV1 3.5, which retains compatibility with the previous iteration. It introduces a new API and features while improving speed and memory optimizations. Version 3.5 of AV1 is vastly more efficient when used with multi-threaded processors. Encoding times are significantly shortened depending on the source resolution. Testing showed a 34% reduction in encoding time on a CPU with 16 or more threads.

According to the official list of changes, the new codec can use frame parallel encoding to gain a 30-34% performance improvement for 1080p videos at best, while 2160/4K resolutions boast a 18-20% reduction in encoding time on a 32-thread CPU.

Streaming software, including OBS Studio, video editing, and recording tools, can take advantage of the AV1 increased performance to offer users a more enjoyable and effective experience. The codec already supported frame parallel encoding on a more significant number of threads using the "--fp-mt" parameter, but in version 3.5, the option is available by default.

The software side of the AV1 venture has improved with every new release, and the hardware side of things also looks increasingly optimistic. Intel Arc GPUs support AV1 hardware encoding and decoding, as does NVIDIA's upcoming GeForce RTX 40 series gaming cards. Meanwhile, Google is working behind closed doors to win industry support for even more widespread use of the AV1 video technology in media content and streaming platforms.

Permalink to story.

 

defaultluser

Posts: 510   +389
So in the end,you get much slower encode time than well-optimized hevc, along with roughly identical quality

such a waste of money just to get rid of all those patents
 

bandit8623

Posts: 442   +242
So in the end,you get much slower encode time than well-optimized hevc, along with roughly identical quality

such a waste of money just to get rid of all those patents
for them its huge money. for us the consumer its not something to worry about... jk.... until we need to buy new hardware which is expensive....
 

defaultluser

Posts: 510   +389
for them its huge money. for us the consumer its not something to worry about... jk.... until we need to buy new hardware which is expensive....


the added cost of Adding complex av1 decode is so high, Roku initially refuses google:


the initial encode has lower overall energy costs than hevc (but not not by much)), and the decode takes an area the size of the state of Montana to finish the encode > decode circle of life! sometimes die size outweighs the cost of patents; we're still several years from cheap chips used in roku / fire TV to get av1
 
Last edited:

bandit8623

Posts: 442   +242
the added cost of Adding complex av1 decode is so high, Roku initially refuses google:


the initial encode has lower overall energy costs than hevc (but not not by much)), and the decode takes an area the size of the state of Montana to finish the encode > decode circle of life! sometimes die size outweighs the cost of patents; we're still several years from cheap chips used in roku / fire TV to get av1
I agree with you. they are also thinking long term, so I doubt they will be using h.266 either.. maybe this next revision will start with the free codecs so we dont have this split.
 

Puiu

Posts: 5,881   +4,891
TechSpot Elite
So in the end,you get much slower encode time than well-optimized hevc, along with roughly identical quality

such a waste of money just to get rid of all those patents
the file sizes are the biggest gain, not the video quality. a 30% reduction in required bandwidth compared to x265 is huge for streaming/video platforms.
 
Last edited:

poltevo

Posts: 45   +27
I've seen mixed numbers for AV1 vs HEVC. Some researchers say 30%, others say its about the same. Any idea why? The complexity of AV1 is certainly higher, however, and the encoding takes longer.

There is also the newer H.266 VVC in development, which is apparently provides 30-50% better compression to HEVC, again at the expense of complexity. I'm not sure how much traction this will get.
 

defaultluser

Posts: 510   +389
I've seen mixed numbers for AV1 vs HEVC. Some researchers say 30%, others say its about the same. Any idea why? The complexity of AV1 is certainly higher, however, and the encoding takes longer.

There is also the newer H.266 VVC in development, which is apparently provides 30-50% better compression to HEVC, again at the expense of complexity. I'm not sure how much traction this will get.


because like any optimizing encoder, it really depends on two things

1 sea of settings you've selected
2. the quality level of source, and type of video

And on that subject, the only folks who care about "up-to 30% in the lab bandwidth savings, but more like 10% real-world" are big streamers like YouTube - good luck convincing he rest of the universe to buy-in!
 

ZedRM

Posts: 1,342   +943
There seems to be a serious level of confusion about video encoding and how AV1 & other encoders work. I would correct that misunderstanding, but such would be very tedious and might be akin to arguing with a brick wall.

So I'll just offer this thought: Many of you need to research the subject matter before making the comments you've made, as in actually read up on it.
 
Last edited:

Puiu

Posts: 5,881   +4,891
TechSpot Elite
I've seen mixed numbers for AV1 vs HEVC. Some researchers say 30%, others say its about the same. Any idea why? The complexity of AV1 is certainly higher, however, and the encoding takes longer.

There is also the newer H.266 VVC in development, which is apparently provides 30-50% better compression to HEVC, again at the expense of complexity. I'm not sure how much traction this will get.
On average you get smaller files with AV1. Settings and file source will always be a big factor.

One of the reasons why AV1 requires more processing power is because it introduces non-planar motion compensation:

“Due to the constant increase in processing power of consumer devices, this technique is now ready to see use in mass-market applications,” says Feldmann. “These techniques work extremely well for predicting large area movements, like background motion or camera movements. Additionally, they can handle consistent backgrounds and color schemes very effectively, which is one of the reasons why animated videos tend to deliver great encoding results, even with very high levels of compression.”

There are a lot more "smart" things AV1 does like how it handles film grain, the block size, etc. This should give you good idea about how it works.

In general AV1 will be the "next gen codec" until we see the some form of mass adoption for h266 (which is still several years away from proper hardware support and licensing is a deal breaker for many).
 

defaultluser

Posts: 510   +389
30% is a big improvement.


most previous new alg releases were at-least 50%

this also adds one of the largest functional decode block ever in-order to work this magic

its a tough sell all around: not enough improvement to gt early embedded soc device wins, while the unimpressive improvements meant most uploads used vp9
 

Puiu

Posts: 5,881   +4,891
TechSpot Elite
most previous new alg releases were at-least 50%

this also adds one of the largest functional decode block ever in-order to work this magic

its a tough sell all around: not enough improvement to gt early embedded soc device wins, while the unimpressive improvements meant most uploads used vp9
the large decode block is intentional because google is preparing for big videos shot in 4K and 8K. let's not forget that AV1 is a google product and they optimised it for their own workflow first. (god knows what settings they use for youtube)