Transcode in H265

[Dibbes 431]

2 hours ago, softworkz said:

Storage - yes, but electricity - no.

When leaving files as H.264, there are chances that playback can be done by only remuxing the video (almost zero "electricity"). After re-encoding them to HEVC, they will always need to be transcoded - with two exceptions:

• DirectPlay
• An experimental and unreliable special playback case on some Apple devices.

Which brings us back to what I had written above:

With electricity I really meant the power required to keep that many extra harddrives spinning, though you're absolutely right on the difference between remuxing and transcoding. However, since I rarely have more than 4 external streams going at the same time, including Live TV and the rest is LAN traffic on recent devices that don't need transcoding, the old 1660 GTX is more than able to keep up 

Either way, I did convert about 75% of my downloaded stuff to HVEC and saved over 55TB. Since at the time I was running everything on 4TB disks (pre-covid), that is a substantial amount of disks in a RAID 6 config...

[softworkz 3341]

23 minutes ago, Dibbes said:

With electricity I really meant the power required to keep that many extra harddrives spinning, though you're absolutely right on the difference between remuxing and transcoding. However, since I rarely have more than 4 external streams going at the same time, including Live TV and the rest is LAN traffic on recent devices that don't need transcoding, the old 1660 GTX is more than able to keep up 

Either way, I did convert about 75% of my downloaded stuff to HVEC and saved over 55TB. Since at the time I was running everything on 4TB disks (pre-covid), that is a substantial amount of disks in a RAID 6 config...

If you saved 55, your total capacity must be > 250. Even when you'd watch 4 videos in parallel for the rest of your life, it will probably require your children and grandchildren (or more) to continue this in the same way in order to get everything watched  

[softworkz 3341]

1 hour ago, softworkz said:

If you saved 55, your total capacity must be > 250. Even when you'd watch 4 videos in parallel for the rest of your life, it will probably require your children and grandchildren (or more) to continue this in the same way in order to get everything watched  

What I actually meant to say is: If your primary objective is collecting and archiving rather than watching, that's a different situation of course and I understand that storage space plays a different role with that vast amounts of content. For the average Emby user (including myself), the gain in disk space would only decide about whether I buy the next HD a few weeks/months later. In your case it sums up quickly, though. When doing the Math, it's clear that a vast amount of your content will never be watched and for those items which will be watched, the drawbacks are probably acceptable, assuming the 75% which are re-encoded are on the unlikely side of being watched (IIUC).

The bottom line is: were talking about extreme and advanced cases and nobody reading this should think that he would do anything good to their library by starting to re-encode everything to HEVC. It saves space (especially for high resolutions like 4k) but there are significant drawbacks that need to be weighed against it. 

The message remains: Don't fiddle around with your videos and let Emby handle all that for you
(as long as you don't have some very specific reasons).

[sebasmiles 8]

8 hours ago, softworkz said:

If you saved 55, your total capacity must be > 250. Even when you'd watch 4 videos in parallel for the rest of your life, it will probably require your children and grandchildren (or more) to continue this in the same way in order to get everything watched  

Sounds like a challenge....

[Carlo 4330]

14 hours ago, softworkz said:

They will come in 2-3 years when that hardware will have become mainstream rather than being at the top edge of the price range.

Emby has always been very flexible and usable on a wide range of hardware from lowest to highest end. I don't see any reason to change that.

The latest Intel graphics do not even work with the 4.7 server, so you can put a zero behind this and only beta users are remaining. Nvidia's Ada Lovelace generation is still extremely high-priced, which doesn't stand in a healthy relation for using those "just" for an Emby Server - like you said. 
The vast majority of Emby Servers is running on mid-range to lower range hardware, which is clearly evident from looking at logs submitted alongside supported requests.

There's always a small number of enthusiasts who would buy such hardware in case - but well...small!

Everybody is free to do such things. My position is that the benefits are questionable and rarely justified, which you might not agree with, but you also need to consider that this is not a question of what YOU are doing - as somebody for whom these things are part of the professional work life.

It's about what normal Emby users should be doing. And they shouldn't do it:

• for once, because it's easy to do it wrong
• and secondly, because this is what they have Emby Server for them to do

Having Emby Server means that you don't need to care about those things - you give it whatever media you have and Emby plays it for you wherever you want.
You don't need to become a transcoding expert or spend endless hours on re-encoding your library. 

Doing these things for you - and in the right and best ways - is the job of Emby Server and an area which it handles with excellence.

It's not just about transcoding a file and having it play on the end-user device but about doing it efficiently and with the best quality possible, IMHO.

Allowing Emby to be soley responsible for all streaming/conversions/transcodes would cost me 2 to 4 times the disc space as well as quality the user receives. I would have a crazy number of different formats including flash, xvid, DVD, BlueRay, interlaced SD & HD mgep2/ts, just to name a few older formats. These are all semi-problematic files when streaming. How about all the typical files people download that use 10 bit AVC files or have wrong headers or timing issues? Not to mention the quality and bitrate differences. Emby can transcode a file using more bitrate than the original, yet degrade the picture noticeable. Compare that to a pipeline conversion where you can do multipass encoding,  and can test a couple variables on 10 to 30 second samples to see what gets the best scores on noise and perceptible quality, then use that profile for the video.  Sure that's probably a lot more than others do but there are tools like TDarr that can do similar things.

Why Now?
Because you're then 3 years behind the curve of people being able to use features already supported and making an impact in the industry. Quality at any given bitrate is superiour for AV1 compared to AVC or HEVC. NVENC AV1 offers substantial compression efficiency with respect to H.264 and HEVC at better performance. You can plot out different profiles then compare signal-to-noise ratios (PSNR) as well as use the video multimethod assessment fusion scores (VMAF) to compare different encodes to gauge encoding quality. This is the ideal way to find the combination of setting (profiles & adjustments) that provide the best outgoing stream for any codec and bitrate combination.

Right now there is no compelling reason to have anything more than a 1650/60 as Emby won't really take advantage of "more" card. If however we supported full functionality of the ADA cards, we would see much more of them in use as it could open up the world to much better transcoding. ADA can support up to 3 encoders and 4 decoders on a single card automatically using all encoders/decoders for even higher throughput.

There are even media accelerator cards such as the Alveo MA35D which lists for $1695 and has 32 separate stream engines that can do up to 8K, AV1 encoding in real time. They have a 16 stream engine version at $995 as well which is cheaper than some GPUs with only 3 encode engines. Is hardware like this over the top for a home media server? Maybe, maybe not. If you only have 20Mb or 30Mb upload and can't do anything about it (ie Comcast) then being able to make much better use of your bandwidth is easily justifiable for many.  Hell, many people spend 2K for a GPU to play games! 

He's a stunner, An H.264 video is encoded using medium presets at 30 Mbps. The same video is encoded using AV1 video at 18 Mbps using the high-performance presets. The VMAF score is equal so there is no perceptible difference between the two videos during playback. The H.264 encoded at roughly 55.5 fps while the VC1 encoded at roughly 500 fps or 9X times faster! Stepping out above an apples to apples comparison both HEVC and AV1 can do 10 bit encoding in addition to 8 bit encoding. 10bit encoding can prove even faster conversion times with better quality yet!

I believe the above is the thought pattern of a lot people in this thread trying to save bandwidth and get better quality encodes along with reduced latency which is a triple win.

[Carlo 4330]

On 9/12/2023 at 2:37 PM, softworkz said:

The bottom line is: were talking about extreme and advanced cases and nobody reading this should think that he would do anything good to their library by starting to re-encode everything to HEVC. It saves space (especially for high resolutions like 4k) but there are significant drawbacks that need to be weighed against it. 

If a high end GPU such as a 4090 is purchased for $1500 or so and used to save storage space, it doesn't take long to break even and start saving costs on HDDs.

[softworkz 3341]

31 minutes ago, Carlo said:

Allowing Emby to be soley responsible for all streaming/conversions/transcodes would cost me 2 to 4 times the disc space

Saving 30% by encoding to HEVC gives a ratio of 100 / 70 = 1.4 times the disk space, not 2 and not 4 times.

36 minutes ago, Carlo said:

These are all semi-problematic files when streaming. How about all the typical files people download that use 10 bit AVC files or have wrong headers or timing issues? Not to mention the quality and bitrate differences. Emby can transcode a file using more bitrate than the original, yet degrade the picture noticeable. Compare that to a pipeline conversion where you can do multipass encoding,  and can test a couple variables on 10 to 30 second samples to see what gets the best scores on noise and perceptible quality, then use that profile for the video.

Can, can, can....nobody does that. And the misunderstanding here is that we're talking about two different things:

• Emby dynamically decides the best possible way for streaming - each time and specific to the actual situation
• Thinking you could do this any better is a misconception (except we're talking about DirectPlay only)

 

44 minutes ago, Carlo said:

Because you're then 3 years behind the curve of people being able to use features already supported and making an impact in the industry.

At the moment, such people are not in sight and neither do I see us behind in any way ("experimental features" do not count).

 

For the remaining points, I think I have addressed them often and detailed enough already in this conversation.

[softworkz 3341]

1 hour ago, Carlo said:

If a high end GPU such as a 4090 is purchased for $1500 or so and used to save storage space, it doesn't take long to break even and start saving costs on HDDs.

For this price, you can get 8 HDs of 14 TB capacity, makes 112 TB, so you would need to have a library of 370 TB to get to a break even (with 30% savings from converting to HEVC).

So, doesn't this lead us back to my statement which you had quoted?

1 hour ago, Carlo said:

On 9/12/2023 at 8:37 PM, softworkz said:

The bottom line is: were talking about extreme and advanced cases and nobody reading this should think that he would do anything good to their library by starting to re-encode everything to HEVC. It saves space (especially for high resolutions like 4k) but there are significant drawbacks that need to be weighed against it. 

Edited September 13, 2023 by softworkz

[softworkz 3341]

1 hour ago, Carlo said:

I would have a crazy number of different formats including flash, xvid, DVD, BlueRay, interlaced SD & HD mgep2/ts, just to name a few older formats. These are all semi-problematic files when streaming.

I'm not aware of any current streaming issues with those files (when HLS streaming with transcoding).

1 hour ago, Carlo said:

How about all the typical files people download that use 10 bit AVC files

There's no hwa which can decode this, but again, I'm not aware of any current issues.

1 hour ago, Carlo said:

 or have wrong headers or timing issues? 

Such things do not get automatically fixed by re-encoding to HEVC. It requires manual intervention.

1 hour ago, Carlo said:

 Not to mention the quality and bitrate differences. Emby can transcode a file using more bitrate than the original

This is not necessarily wrong. Emby looks at the bandwidth that is available for streaming to a client. When there's strong compression or rare key frames, Emby cannot replicate this for streaming when transcoding on-the-fly.

But re-encoding to HEVC doesn't change this.

1 hour ago, Carlo said:

 Emby can transcode a file using more bitrate than the original, yet degrade the picture noticeable.

Again, re-encoding to HEVC doesn't change this behavior.

Encoding quality can be adjusted by changing encoder parameters.

Edited September 13, 2023 by softworkz

[Carlo 4330]

18 hours ago, softworkz said:

For this price, you can get 8 HDs of 14 TB capacity, makes 112 TB, so you would need to have a library of 370 TB to get to a break even (with 30% savings from converting to HEVC).

So, doesn't this lead us back to my statement which you had quoted?

I ended up with 30% the size, not 30% savings. That's going from a nearly 2 PB of storage down to 600 TB ish. That for the online as well as backup as well.

Let's suppose this person with 8 HDDs is using ZFS with one set of parity. That would give them 65.9 TiB of useful storage.  Assuming this person also has drives to backup the online data then a savings of roughly 35-40 TiB would be break even for a 4090. Even the most expensive GPUs are far, far cheaper then storage at scale.  For me I was able to save 280 10TB HDDs, 12 SAS Enclosures as well as the cables and SAS cards used to drive things.  The cables and cards alone cost more than a 4090. That's roughly 1.4 PB less data needing to be backed up which of course figures into costs/savings as well. 

On 9/12/2023 at 2:37 PM, softworkz said:

The bottom line is: were talking about extreme and advanced cases and nobody reading this should think that he would do anything good to their library by starting to re-encode everything to HEVC. It saves space (especially for high resolutions like 4k) but there are significant drawbacks that need to be weighed against it. 

I'm not seeing the drawbacks as such GPUs are quickly becoming the defacto standard used, from small WOWZA setups to Live streaming via online social platforms like YouTube, to gamers using OBS studio pushing out feeds between 1080 and 4K.  Bottom line is everybody else seems to be embracing AV1 and any GPU (especially NVidia) that can be used to generate the streams. All the popular video editing software packages support AV1 and is especially popular for low latency environments requiring high quality streams.

Please don't take anything said as me being negative toward our transcoder as it has a purpose and does that pretty well. I'd just like to see it expanded to catch up to current protocols and codecs that present devices & browsers are able to take advantage off using resources more efficiently when-ever possible.

16 hours ago, softworkz said:

Can, can, can....nobody does that. And the misunderstanding here is that we're talking about two different things:

• Emby dynamically decides the best possible way for streaming - each time and specific to the actual situation
• Thinking you could do this any better is a misconception (except we're talking about DirectPlay only)

At the moment, such people are not in sight and neither do I see us behind in any way ("experimental features" do not count).

 

For the remaining points, I think I have addressed them often and detailed enough already in this conversation.

Lets look at what our transcoder does. It encodes to a format known to play on x device. It can cap the bitrate used, transform audio, manipulate subs, etc but from the standpoint of video it basically targets AVC.  It doesn't try to support the best video codec available to both server and client device or support low latency streaming  nor adaptive streaming. It really doesn't try to supply the best stream(s) to the client using a low latency protocol trying to minimize bandwidth. It doesn't remove audio tracks or subs not currently being used. It doesn't really do anything to to save resources or improve quality of received streams.

15 hours ago, softworkz said:

There's no hwa which can decode this, but again, I'm not aware of any current issues.

Such things do not get automatically fixed by re-encoding to HEVC. It requires manual intervention.

Exactly, hence why being able to fix things like this offline can take the burden off the transcoder and give you the exact type of stream you want to have as the default.

[softworkz 3341]

23 minutes ago, Carlo said:

I ended up with 30% the size, not 30% savings. That's going from a nearly 2 PB of storage down to 600 TB ish. That for the online as well as backup as well.

Let's suppose this person with 8 HDDs is using ZFS with one set of parity. That would give them 65.9 TiB of useful storage.  Assuming this person also has drives to backup the online data then a savings of roughly 35-40 TiB would be break even for a 4090. Even the most expensive GPUs are far, far cheaper then storage at scale.  For me I was able to save 280 10TB HDDs, 12 SAS Enclosures as well as the cables and SAS cards used to drive things.  The cables and cards alone cost more than a 4090. That's roughly 1.4 PB less data needing to be backed up which of course figures into costs/savings as well. 

LOL - that's really insane figures...

24 minutes ago, Carlo said:

I'm not seeing the drawbacks as such GPUs are quickly becoming the defacto standard used, from small WOWZA setups to Live streaming via online social platforms like YouTube, to gamers using OBS studio pushing out feeds between 1080 and 4K.  Bottom line is everybody else seems to be embracing AV1 and any GPU (especially NVidia) that can be used to generate the streams. All the popular video editing software packages support AV1 and is especially popular for low latency environments requiring high quality streams.

Yes, this will come. But as mentioned above, the cases where it would be applicable right now are just a fraction (in Emby context). That fraction is so low, that it wouldn't even provide sufficient coverage for testing it.

24 minutes ago, Carlo said:

 everybody else seems to be embracing AV1 and any GPU (especially NVidia) that can be used to generate the streams

To my knowledge, only the Ada Lovelace generation can encode AV1.

29 minutes ago, Carlo said:

Please don't take anything said as me being negative toward our transcoder as it has a purpose and does that pretty well. I'd just like to see it expanded to catch up to current protocols and codecs that present devices & browsers are able to take advantage off using resources more efficiently when-ever possible.

Give me a team of 5 developers and 20 testers, as well as a fully stuffed hardware lab, then it would be a different story.  

We have limited resources and that requires us to make good decisions at the right points in time in order to maximize achievements while minimizing the required effort.

[speechles 1920]

But what happens when you play these AV1 streams? Android 14 is where the line is drawn for AV1 support. My Samsung S20FE cannot render AV1 without dropping half the frames and dropping behind on audio sync. The issue isn't about encoding AV1 since that is practical. But what is impractical is your friends keeping up with the Jones's. It is impractical for your friends and family to upgrade things just because you have the next best thing since sliced bread. Once the prices become something more mainstream and the R&D costs of support AV1 have been paid for by prior adopters things will get cheaper. Right now the cost of AV1 is very high. The cost of HEVC is becoming very low. There are cost/benefit ratios for everything. You have to evaluate your friends and family into this equation. The future is AV1 but is that future right now?

[Carlo 4330]

4 hours ago, softworkz said:

LOL - that's really insane figures...

I agree with you on that but when you collect everything digitally for 40 years and upgrade them when available your storage needs kind of grow and grow. I've got things like 20 years of NFL games (not a team, but every game for the week. Same for some other sports as well. 2+million episodes, 100K plus movies, etc... I only have a subset of this actually in Emby because the sports are of no interest to anyone but me on my system. Movies and shows before 1970 for the most part except some classics aren't in Emby either for the same reason.

4 hours ago, softworkz said:

Yes, this will come. But as mentioned above, the cases where it would be applicable right now are just a fraction (in Emby context). That fraction is so low, that it wouldn't even provide sufficient coverage for testing it.

Not sure I follow you on the fraction of users who would benefit right now.  If you mean AV1 specifically, I'd agree but in the context of using the best codec available to both the server and client, then HEVC would get a good workout and would offer many of the same benefits as AV1 would (just not as extreme).  The only negatives I'd see about HEVC is the included codec in ffmpeg is subpar compared to other HEVC codecs. But considering I'd only advocate HW used for anything past AVC it's probably not a big deal. The real negative t me is possible license issue but considering we are a "home media server" using ffmpeg I'd highly doubt there would ever be an issue since the devices themselves have a license to use HEVC.

4 hours ago, softworkz said:

To my knowledge, only the Ada Lovelace generation can encode AV1.

3000 series added dedicated decode for AV1 while 4000 series do both decode and encode. Any modern quad core CPU should be able to decode AV1 as well without much issue.

4 hours ago, softworkz said:

Give me a team of 5 developers and 20 testers, as well as a fully stuffed hardware lab, then it would be a different story.  

We have limited resources and that requires us to make good decisions at the right points in time in order to maximize achievements while minimizing the required effort.

LOL, nice try.  Information is available to from Netflix, Prime as well as real time WOWZA encoding ladders to see what protocols and codecs are being used for what client devices to provide the most efficient use of bandwidth.

4 minutes ago, speechles said:

But what happens when you play these AV1 streams? Android 14 is where the line is drawn for AV1 support. My Samsung S20FE cannot render AV1 without dropping half the frames and dropping behind on audio sync. The issue isn't about encoding AV1 since that is practical. But what is impractical is your friends keeping up with the Jones's. It is impractical for your friends and family to upgrade things just because you have the next best thing since sliced bread. Once the prices become something more mainstream and the R&D costs of support AV1 have been paid for by prior adopters things will get cheaper. Right now the cost of AV1 is very high. The cost of HEVC is becoming very low. There are cost/benefit ratios for everything. You have to evaluate your friends and family into this equation. The future is AV1 but is that future right now?

Ideally of course, AV1 is only ued for devices that can handle it. If a device couldn't handle AV1 then HEVC would be a viable alternative and of course the trusty old AVC. Ideally, the client would get a manifest with several different stream resolutions and codecs so if it couldn't handle the bitrate/codec fast enough it would switch streams until it got a stream it's happy with.

I totally agree on not forcing people to upgrade client devices, nor server GPUs if they are all ready happy. A good many devices and mobile phones/tablets already support AV1. HEVC is everywhere so that's almost a no brainer. So VC1 is used if both the server and client support it, if not fall back to HEVC which the large majority of clients do support. If the server itself doesn't support this it could be added fairly cheaply with a used Nvidia model supporting Turing or better architecture. Anything Geforce GTX 1650 (some were Pascal) or better would work. The 1660 Super editions can be had for $50/60 off eBay pretty easily.

NAS boxes with Intel chips should be able to transcode both HEVC and AVC but NAS boxes without Intel chips are always the wildcard, just as the are already.

[iiiJoe 55]

Question: would this potential migration to HVEC reduce the amount of data used through my ISP to stream m3u’s? Would really like to reduce my data usage

[sshanee613 3]

The intel arc a380 is a $120 usd gpu that can encode av1. There are higher end arc gpus but the 380 is very cheap, obtainable and low power.  This av1 stuff would be of most importance to your highest paying users with lots of streams/clients to best utilize resources and bandwidth. As Carlos has said, the percentage is low because there's no reason to add hardware with these capabilities... emby can't support it. What percentage of embys income comes from the people with 15 year old machines vs the people with 5000 dollar servers? That 1% statistic doesn't have much relevance when you think of it this way.

Edited September 15, 2023 by sshanee613

[sebasmiles 8]

12 hours ago, sshanee613 said:

The intel arc a380 is a $120 usd gpu that can encode av1. There are higher end arc gpus but the 380 is very cheap, obtainable and low power.  This av1 stuff would be of most importance to your highest paying users with lots of streams/clients to best utilize resources and bandwidth. As Carlos has said, the percentage is low because there's no reason to add hardware with these capabilities... emby can't support it. What percentage of embys income comes from the people with 15 year old machines vs the people with 5000 dollar servers? That 1% statistic doesn't have much relevance when you think of it this way.

I thought support for ARC gpu's are fairly bad still or am I wrong?

[ebr 14935]

15 hours ago, iiiJoe said:

Question: would this potential migration to HVEC reduce the amount of data used through my ISP to stream m3u’s?

Only on the upload side.  

[rbjtech 4289]

1 hour ago, sebasmiles said:

I thought support for ARC gpu's are fairly bad still or am I wrong?

Support has been a lot better in the last 6 months or so.   

tbh, I think until mainstream CPU's support AV1 encode in hardware, then it's not going anywhere soon.

ARC is a discreet GPU option only if you want to 'dabble' in AV1 encode (they are still not cheap enough to seriously compete with Nvidia/Amd) - and the 4000 Nvidia GPU's are still silly money if your only interest is AV1 encode...

I think 14th Gen Intel and the Amd/Zen5 will both have AV1 encode built it - THEN I think we may see some traction....

[softworkz 3341]

2 hours ago, sebasmiles said:

I thought support for ARC gpu's are fairly bad still or am I wrong?

Emby Server Release/stable versions: Not supported
Emby Server Beta (and upcoming 4.8.x): Well supported

Edited September 15, 2023 by softworkz

[sshanee613 3]

1 hour ago, rbjtech said:

Support has been a lot better in the last 6 months or so.   

tbh, I think until mainstream CPU's support AV1 encode in hardware, then it's not going anywhere soon.

ARC is a discreet GPU option only if you want to 'dabble' in AV1 encode (they are still not cheap enough to seriously compete with Nvidia/Amd) - and the 4000 Nvidia GPU's are still silly money if your only interest is AV1 encode...

I think 14th Gen Intel and the Amd/Zen5 will both have AV1 encode built it - THEN I think we may see some traction....

The current gen cpus have av1 support and an arc a380 cost 120 dollars. It's here, now.

[iiiJoe 55]

1 hour ago, ebr said:

Only on the upload side.  

Ahh. Don’t do any uploading atm. Thanks for answering

[jaycedk 389]

6 minutes ago, iiiJoe said:

Ahh. Don’t do any uploading atm. Thanks for answering

Your upload is what users use to watch movies / shows etc.

The smaller footprint that has the better, and the more user you can cater to. 

[rbjtech 4289]

9 minutes ago, sshanee613 said:

The current gen cpus have av1 support and an arc a380 cost 120 dollars. It's here, now.

You have a Meteor Lake Intel CPU ?  Interesting as they have yet to be released ...

Pleas read what I wrote - AV1 Encode is not available in any mainstream CPU ...

Yes many CPU's have AV1 Decode - but we are not talking about that ... 

[TMCsw 122]

19 hours ago, sshanee613 said:

What percentage of embys income comes from the people with 15 year old machines vs the people with 5000 dollar servers?

Probably very little from either as most people opt for the 'Lifetime Emby Premiere plan'. Although some do choose the monthly/annual plan to help support the dev. team. 

[sshanee613 3]

3 hours ago, rbjtech said:

You have a Meteor Lake Intel CPU ?  Interesting as they have yet to be released ...

Pleas read what I wrote - AV1 Encode is not available in any mainstream CPU ...

Yes many CPU's have AV1 Decode - but we are not talking about that ... 

Ah yes I was misinformed. But amd and intel cpus will be out soon that do encode.