Hacker News new | past | comments | ask | show | jobs | submit login

NVENC on the Turing architecture (RTX2080/2070/2060) is at the highest quality setting comparable to x264 fast. During complex scenes closer to faster. Older architectures (Pascal etc) are worse.

Gold standard for H.264 quality/encode time/bitrate is still x264 medium / slow.

For H.265, the situation is much different. x265 is far from as mature as x264 and frankly the licensing is such a mess that it looks like H.265 might actually lose the war of next generation codec. Instead VP9 -> AV1 seems to be the industry’s codec of choice with adaption from Google, Twitch and Facebook. Also note Apple is back on the AV1 page.




I like messing around with ffmpeg from time to time, and while I love free formats I have say VP9 is practically unusable for me, it just takes insane amounts of encoding time. I'm unsure what hardware people are using for it, especially when it hardly multi-threads...


That's all too true. However it's important to remember that when H.264 started to get public traction (this is back in 2005, when Apple started pushing H.264 movie trailers), realtime CPU encoding was deemed impossible at that point in time. It just wasn't realistic back in 2005 to produce H.264 on a CPU, instead hardware encoders were used to accomplish the task.

Fast forward 10 years and basically every single laptop has hardware accelerated H.264 encoding through Intel's QuickSync FFHW and desktops have it through Nvidia. On top of that the software encoder (x264) is so fast that it's possible to do it in realtime on a CPU.

There is a fair chance that VP9 will see the exact same pattern here. Today CPU encoding is unfeasible, as you've noticed. However Intel is releasing VP9 encoding on their QuickSync encoder with newer CPU's and Nvidia will from my guess have it either 1 or 2 GPU generations from now. Same goes for AV1 here.

It's also interesting too think that the age of these aggressive improvements in CPU speed we've had over the past 14 years (from 2005) might come to an end, where efficiency instead comes from specialised hardware, such as fixed function hardware doing encoding on the chips, as NVENC/QuickSync does today. I don't know if realtime AV1 encoding will be feasible on a consumer grade CPU within the next 10 years, but i sure know it's hardware counterpart will be.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: