Wanting to move my library from H.264/MP4 to AV1 (SVT-AV1) to save rack space.Looking for maximum space savings without noticeable VMAF drops.Is the efficiency gain over HEVC/H.265 worth the encode time on a [RTX 5070 Ti ]?What CRF/preset are you guys using for "transparent" 1080p/4k?Is the hardware decoding support stable enough now or am I going to regret this in 2 years?Don't want to lose quality, just want smaller blobs.
>gpu encodinglol
>>108752275isn't gpu encoding better?. i really don't have experince in this stuff
>>108752241hardware encoding even with the best consumer av1 encoders is still totally pointless (nvenc av1 barely competes with svt-av1 preset 8 which is already stupidly fast to do on cpu anyways), you should only ever consider hardware encoding for real time applications where you need to use that specific codec for some reason.by encoding av1 in hardware you are losing compatibility over h264, while maintaining x264 medium levels of efficiency pretty much... makes 0 senseeven compared to h264 hardware encoders the gain is extremely smallsee: https://rigaya.github.io/vq_results/
>>108752275this, encoding AV1 video through svt-av1 is now literally faster than encoding H264 through x264. GPU video encoding is becoming pointless.>>108752318It's on par with like Preset 10 svt-av1. Preset 4 svt-av1 achieves like 50% better compression than AV1 GPU encoding.
>>108752318it's much faster, and much less efficient, the gap increases with modern codecs.you're trading off efficiency/quality for speed, which you don't want for archival.it also of course doesn't use cpu which means it's better for streaming/making game clips in the background, but they make no sense in your use case.
>>108752348i will use my i5 14400f then. but can you explain the presets thing. thanks in advance
>>108752342i see you are using ffmpeg is it better than HandBrake ?
>>108752402preset number lower = video size smaller for similar quality but encode takes longerYou'll also want to use -pix_fmt yuv420p10le to get another 10% compression boost and eliminate color banding problems that plague virtually all H264 tips right now.
>>108752430i was told by a friend to use HandBrake is it better/worse than ffmpeg ?
>>108752424>>108752441Not mine but generally yes, you get access to fancy things like frame lookahead, variance boost, vq tune, and so on and so on.If you don't know what you're doing download staxrip instead and select the svt-av1-psyex AV1 encoder. That has all the movie-friendly parameters enabled by default. All you have to do is select CRF (30-40 generally advisable) and preset (4 for modern CPUs and 8 for old CPUs).
>>108752441this bullshit site won't let us upload av1 video yet but here's a link to one where someone posted an av1 staxrip tutorialhttps://umigalaxy.com/explore/general/275-i-made-a-video-tutorial-of-how-to-encode-av1
>>108752464thanks man
>>108752441i'm pretty sure handbrake IS ffmpeg, it's just a GUI for it that's supposed to make things easier for retardsbut the reason i switched to ffmpeg a bijillion years ago is literally because handbrake's ui is so retarded and convoluted that i found typing a couple of commands in a commandline much easier...
>>108752241>Is the hardware decoding support stable enough now or am I going to regret this in 2 years?Pretty much all modern x86 CPUs have it. Even the old ones from 5-10 years ago that don't can software decode 10-bit 1080p AV1 video at hundreds of FPS, you probably won't even notice the fan spinning up.https://openbenchmarking.org/test/pts/dav1dARM CPUs, as usual, are lagging behind hard. If it's not a flagshit then expect hardware av1 decoding to not be present. However from my testing if you see Cortex A78 or better on the CPU spec, software decoding 10-bit 1080p AV1 at 1Mbps has an estimated battery life of 8-10 hours. Which imho is usable from this point on.
>>108752335>while maintaining x264 medium levels of efficiency pretty muchis NVENC really that bad?
>>108752660kind ofthe thing is, nvidia says "nvenc h264 is on par with x264 slow" but they literally just cheat at vmaf (https://arxiv.org/pdf/2107.04510)doing side-side comparisons and using traditional metrics like ssim, you can easily tell this is the case.but besides, while nvenc might be able to compete with x264 (which is no problem doing on cpu anyways), once again, you can clearly tell that the gap increases with h265 and even more with av1.newer codecs just have too many features compared to older ones, you'd need a huge amount of die space to implement every function, it's just not possible to make it work for consumer hardware outside of certain use cases (streaming/background recording), and as cpus and software encoders get faster this gap will keep increasing.Picrel is a Video Coding Unit designed by google to transcode youtube videos, notice how each encoder core is huge compared to the decoder core, while on an nvidia gpu, nvenc is barely twice the size of nvdec.
