Encoding Guide (AV1 specifically)

howto
1

WIP

Since I am uploading a lot of AV1 encodings I have seen many questions about how I do it and why the file sizes are so small. This thread will be where I explain things about encoding and AV1 without trying to make it too complicated so it should not be hard to understand and neither am I that knowledgeable anyway.

What is AV1? AV1 is essentially the newest codec for videos, it means that while it is not as supported on older hardware as the older codecs such as H.264 or H.265/HEVC and is also harder to decode (play the video), the benefit is that it is MUCH more efficient. What does efficiency do for videos? Basically more efficient codec means that you can pack the same amount of data into the video but making it look better visually, so it will either look better at the same amount of space taken or it will look the same but take less space. I believe that AV1 is the way to go in the future considering it is open source while the closest competitor (HEVC/H.265) is not and that a lot of the major companies out there such as Netflix, Google, Microsoft and Apple and more all suppport/use AV1.

How do you encode videos into AV1? The easiest way that works on all hardware is to simply use CPU encoding (SVT-AV1), you can do this in a lot of different programs but the one I would recommend for the features and ease of use is to use Handbrake. Specifically I would recommend using this version of it as it has AV1-SVT-PSY specifically. That PSY version of AV1-SVT is essentially an improved version of it that is focused on psychovisually optimal AV1 encoding, that means it is meant to look better subjectively for the human eye.
After installing it, I would recommend that you create a preset that is always selected when you start it and after that you only need to adjust 2 different factors (usually).

image
image999Ă—759 144 KB

Select MPEG-4 (mp4) and passthru common metadata in the first summary page.

image
image999Ă—759 88.4 KB

In the Dimensions page if you want all your videos to be in 1080 at most, then in the “Resolution & Scaling” section select 1080p Full HD in “Resolution Limit”, if you want 1440p or 4K as max then select that or simply “none” if you do not want a limit.
Nothing else needs to be adjusted here. Cropping on Automatic is useful so I would have that on.

The Filters, Subtitles, Chapters and Tags can be left alone and skipped. You can of course adjust the tags, chapters and subs as needed but I would not do that via presets.

image
image999Ă—759 78 KB

Now for the main thing we need to pay attention to, the Video tab.
Set the “Video Encoder” to AV1 10-bit (SVT PSY), this is for the AV1 codec using SVT-PSY variant and we also want it to be in 10bit as that is (afaik) always better than doing it in 8bit with better colors and also smaller file size sometimes, even if the original video you are re-encoding from is 8 bit you will see benefits from 10bit encoding.

Set framerate as “same as source” and have it on constant framerate.

The 2 factors that you need to pay attention are the “Preset” and “Constant Quality (RF)” factors.
Preset determines how fast it will encode, setting the preset value to a higher number (max 10) will make it run faster and lower will be slower (slowest being -3). Slower presets are slower because they basically go through the file more thoroughly and uses more advanced techniques to do so and MAY (keyword MAY) make the file size smaller and/or higher quality = more efficient encoding. So the main thing to keep in mind here is simply how much time and/or electricity cost you are willing to sacrifice for a file that MAY be smaller and/or higher quality. General advice from the Codec wiki is that you should stick to the 2-8 range for AV1-SVT-PSY.
And that going from one preset to the next generally means a 2x difference in speed. The sweet spots however are P2, P4 and P6. As going from say P2 to P1 takes more than 2x the time but the benefits are extremely small so not worth so and a similar case with P4 to P3. Anything faster than P8 however is just not good at all and you be better of using Hardware Acceleration, that is using the GPU to do the encoding instead, although if you have an AMD GPU and no Intel iGPU either I would recommend to stick with CPU encoding as AMD hardware accel is not that good (afaik). It is a completely different story if you have an Intel dGPU or Nvidia, but more on that later.

The RF value on the other hand, determines the compression, that is how small the file will end up and how much quality will be sacrificed in doing so. With a higher number meaning more compression and smaller number less compression and closer to the original file. This one will be the value you likely change the most depending on the quality, resolution and content on the original file you are working with.

For example if the original file is terrible and also using say an old coded like H.264 to begin with, you should just compress it more that way you can save a lot of space AND in some cases still looks the same as the original as AV1 shines at low file sizes compared to the older codecs.
If the original file looks good though you should not compress it as much though.

Depending on the resolution you may also compress it more or less, generally you want less compression with higher resolution files and more compression with lower resolution.

The content on the file will be a huge factor on how much you can compress though, for example with animation you can usually compress quite a bit more than live-action content without making it look worse.

RF value is pretty subjective and there is no “best” values in general as it depends on your eyes (if you can tell a bad from a good quality file apart without pausing and A/B comparing the same exact frame) and also how you consume it, be it from a smartphone where the screen is much smaller and thus has a higher PPI (pixel density) than say a big TV screen where you will have low PPI and bad quality will stick out much more.

Note that you should not re-encode the same video that often as the more times you do the more data is lost, if you want a smaller file simply start with a higher compression (quality value) instead of trying to make it smaller by re-encoding the re-encode.

^ Everything said above is essentially the basics and what seems to be the consensus. Now from here on I will talk about my own experiences and my own recommendations based on that.

For those that have either an NVIDIA or INTEL dGPU I would recommend using it for encoding, as I personally only have an Intel GPU and am on Linux I am using a script to do my encodings using Intel QSV. Here it is if you want to reference it and use it for your own (NVENC if Nvidia) and/or if on Windows, I recommend asking Gemini to do the work for you as that is what I did.
Scripts ~ pixeldrain
One is a manual script where you can input the quality yourself and the other is a more advanced one that checks the video first and determines the optimal quality value to use. I personally have them both set so they can only use a value from 20-31 as under that is just not noticeable and a waste of space and above that is where I notice the difference (with 1080p content, I have not added a way for it to be able to use more compression at lower resolution but that is a good way to do it). The one that checks the files first also makes sure that the end result is at least 5% smaller than the original file.

I downloaded some random short clips from Rule34videos and encoded them into AV1 using both CPU encoding (AV1-SVT-PSY 10bit) on my AMD Ryzen 5 7500F and using hardware acceleration using my Intel Arc B580 GPU (Intel QSV). Using VMAF as the objective way to analyze the results, VMAF is a way to estimate the quality of the video with a focus on perceived subjective quality and was made by Netflix. With Netflix afaik aiming for a value of 93 for their streams as that is the best value to them where you get best quality and file size with a value of 95 and above being essentially lossless and indistinguishable. A value of 80 being considered good and 90 and above very good.

Dark Video:

Encoding Method Hardware Preset Quality setting Bitrate (Mbps) File Size (MiB) VMAF Score
Reference Original Source N/A Original 2,16 0,76 N/A
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 47 0,34 0,12 93,2222
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 46 0,37 0,13 93,2765
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 45 0,39 0,14 93,3871
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 44 0,41 0,15 93,5246
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 43 0,44 0,16 93,7234
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 42 0,46 0,16 93,8107
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 41 0,51 0,18 93,9366
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 40 0,53 0,19 93,9986
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 39 0,55 0,19 94,0961
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 38 0,59 0,21 94,1777
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 37 0,63 0,22 94,2944
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 36 0,67 0,24 94,3956
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 35 0,74 0,26 94,5214
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 34 0,79 0,28 94,6339
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 47 0,48 0,17 93,1339
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 46 0,51 0,18 93,2778
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 45 0,52 0,18 93,3407
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 44 0,57 0,20 93,5240
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 43 0,61 0,22 93,6410
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 42 0,63 0,22 93,7887
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 41 0,66 0,23 93,8716
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 40 0,71 0,25 93,9572
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 39 0,72 0,25 93,9935
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 38 0,77 0,27 94,1069
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 37 0,81 0,29 94,1769
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 36 0,86 0,30 94,2573
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 35 0,91 0,32 94,3086
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 34 0,96 0,34 94,3536
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 6 47 0,42 0.15 92,8465
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 6 46 0,46 0,16 93,0865
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 6 34 0,84 0.30 94,2252
GPU (QSV, AV1) Encoding Intel Arc B580 1 31 0,29 0,10 90,3107
GPU (QSV, AV1) Encoding Intel Arc B580 1 30 0,31 0,11 90,8827
GPU (QSV, AV1) Encoding Intel Arc B580 1 29 0,35 0,12 91,0972
GPU (QSV, AV1) Encoding Intel Arc B580 1 28 0,39 0,14 91,9427
GPU (QSV, AV1) Encoding Intel Arc B580 1 27 0,44 0,15 92,2590
GPU (QSV, AV1) Encoding Intel Arc B580 1 26 0,48 0,17 92,5025

Contrast video:

Encoding Method Hardware Preset Quality setting Bitrate (Mbps) File Size (MiB) VMAF Score
Reference Original Source N/A Original 6,62 2,8 N/A
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 47 2,07 0,95 98,4619
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 46 2,14 0,984 98,5882
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 45 2,34 1,1 98,7994
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 44 2,42 1,1 98,8809
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 43 2,62 1,2 99,0978
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 42 2,7 1,2 99,1309
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 41 2,92 1,3 99,3033
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 40 3,0 1,3 99,3062
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 39 3,23 1,4 99,4313
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 38 3,43 1,5 99,5326
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 37 3,62 1,6 99,5898
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 36 3,78 1,6 99,6394
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 35 4,0 1,7 99,6908
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 34 4,25 1,8 99,7477
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 47 2,22 1,0 98,1217
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 46 2,38 1,0 98,1955
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 45 2,51 1,1 98,5222
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 44 2,58 1,2 98,6154
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 43 2,79 1,2 98,8648
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 42 2,9 1,3 98,9122
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 41 3,13 1,4 99,1219
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 40 3,23 1,4 99,1591
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 39 3,45 1,5 99,2836
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 38 3,68 1,6 99,4040
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 37 3,86 1,7 99,4608
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 36 3,99 1,7 99,5149
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 35 4,22 1,8 99,5872
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 34 4,49 1,9 99,6618
GPU (QSV, AV1) Encoding Intel Arc B580 1 31 2,25 1,0 97,7989
GPU (QSV, AV1) Encoding Intel Arc B580 1 30 2,52 1,1 98,3457
GPU (QSV, AV1) Encoding Intel Arc B580 1 29 2,82 1,3 98,7163
GPU (QSV, AV1) Encoding Intel Arc B580 1 28 3,23 1,4 99,1337
GPU (QSV, AV1) Encoding Intel Arc B580 1 27 3,7 1,6 99,3723
GPU (QSV, AV1) Encoding Intel Arc B580 1 26 4,07 1,8 99,4721

Bright video:

Encoding Method Hardware Preset Quality setting Bitrate (Mbps) File Size (MiB) VMAF Score
Reference Original Source N/A Original 3,74 2,1 N/A
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 47 1,23 0,81 92,4064
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 46 1,28 0,84 92,6033
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 45 1,39 0,90 92,9008
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 44 1,46 0,93 93,0006
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 43 1,54 0,97 93,3897
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 42 1,65 1,0 93,5509
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 41 1,77 1,1 93,9570
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 40 1,83 1,1 94,0648
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 39 1,96 1,2 94,3419
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 38 2,12 1,3 94,6496
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 37 2,25 1,3 94,8760
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 36 2,38 1,4 95,1027
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 35 2,52 1,5 95,2739
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 2 34 2,73 1,6 95,5762
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 47 1,34 0,87 92,0926
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 46 1,39 0,89 92,3913
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 45 1,46 0,93 92,5546
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 44 1,57 0,99 92,8556
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 43 1,66 1,0 93,0970
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 42 1,73 1,1 93,3317
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 41 1,87 1,1 93,5472
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 40 1,94 1,2 93,7740
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 39 2,05 1,2 94,0271
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 38 2,23 1,3 94,3230
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 37 2,34 1,4 94,5244
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 36 2,45 1,4 94,6881
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 35 2,60 1,5 94,9279
CPU (AV1-SVT-PSY-10bit) Encoding AMD Ryzen 5 7500F 4 34 2,79 1,6 95,1663
GPU (QSV, AV1) Encoding Intel Arc B580 1 31 1,32 0,86 90,4133
GPU (QSV, AV1) Encoding Intel Arc B580 1 30 1,45 0,93 91,2654
GPU (QSV, AV1) Encoding Intel Arc B580 1 29 1,65 1,0 92,2063
GPU (QSV, AV1) Encoding Intel Arc B580 1 28 1,91 1,2 93,2431
GPU (QSV, AV1) Encoding Intel Arc B580 1 27 2,18 1,3 93,9532
GPU (QSV, AV1) Encoding Intel Arc B580 1 26 2,41 1,4 94,4178

Summarized table:

Encoding Method Preset Average Bitrate (Mbps) Average File Size (MiB) Average VMAF Score
CPU (AV1-SVT-PSY-10bit) Encoding 2.0 1.81 0.89 95.71
CPU (AV1-SVT-PSY-10bit) Encoding 4.0 1.97 0.95 95.51
CPU (AV1-SVT-PSY-10bit) Encoding 6.0 0.66 0.23 93.86
GPU (QSV, AV1) Encoding 1.0 1.77 0.87 94.30

As you can see from the tests, using CPU encoding does indeed give a better vmaf score and tends to do so at a smaller file size making it more efficient to use CPU encoding over GPU encoding even with Intel QSV which is top class for hardware acceleration.

However, the speed is not something that was tested in these short (3 ish seconds) clips and that is where the crucial point lies. From my experiences using the slowest preset (1) on my Intel B580 has an encoding speed of around 250-350fps usually, compare that with using P4 on my 7500F which has an encoding speed of around 10fps and around 1-3fps on P2, means that even on P4 it is around 30x slower than the slowest Preset on the GPU. This essentially means that a 2 hour 1080p30fps video would take around 12 minutes on the GPU but 360 minutes on the CPU at P4 and a whopping 1080-3600 minutes on P2.

Which is why personally I would simply stick to using hardware acceleration and only switch to CPU encoding for the files you REALLY like and only use P4 or P2 for that. My advice is that you run multiple “batches” at the same time, which is what I usually do to save even more time as you can run multiple encodings at the same time without it affecting speed much at all for each individual one. I also recommend running a CPU encoding alongside the GPU ones to really maximise your encoding efficiency.

Tested files:
Emma = contrast video
Jesse = dark video
Mercy = bright video

7 Likes
2

I have a pretty beefy machine, I didn’t use the psy version but it looks damn near the same on the 10 videos I tested. But I use svt non 10 bit and a P5 at a 40 RF and I’ve compresses more then 1.5TB into about 430GB. Though I think I messed up my audio passthrough… Oh well. I use a program called hbbatchbeast and make a preset in handbreak, then export it to be used with that software. It uses handbreak-cli to convert multiple files at once and I’ve set it to replace the original if it’s smaller. I find that there are a lot of videos that have a way to high bitrate for what it’s displaying so it’s a good compression ratio. The 2450+ files I converted took like 12 days on my amd Ryzen 9 7900x. And I have has to pause it a few times to reboot or windows updated and stopped it.

I will say that in my testing, if you want to compress and don’t care about the time/electricity cost… Then don’t use the GPU for Encoding. It’s a significantly more inefficient conversion process. Most of those methods are for streaming and hitting a specific bitrate. Not compression while keeping a good visual quality comparison.

3

I would try P4 or P6 instead, P5 is not really that beneficial over the faster P6.
And use 10bit always.

Handbrake should be gaining the PSY features near this years end afaik so when the time comes you can just use regular handbrake with AV1-SVT.

Sounds similar to my script then, there are some other batch/automation programs out there that does something similar though I cant remember the names now.

That sounds about right im still working with my own library and am just about half done with my last big subfolder, being JAVs at around 10TB and will probably be around 2-3TB when I am done.

Yeah, It is true though that with the older codecs, they really do look terrible at low bitrates so I kinda get it but if the content is low quality and/or can be easily compressed then that is just a waste of space..

Yup, speed will vary greatly depending on the CPU.

This is where I will disagree, as shown in my tables above and the rest of my experiences, the difference in quality when you get to at least 90 vmaf score is not noticeable in actual viewing and not a paused A/B comparison at the exact same spot, at least for me if you had me doing a blind test without knowing which is which, it would come down to pure guessing.

This is where I find it extremely annoying when I see people online just shitting on hardware accel since whats the point of taking so much longer for no visual difference for the human eye during actual use?

Of course, if you are able to actually tell the difference and not just pure guessing and time/electricity is not a concern then CPU encoding will of course be the best way. Otherwise it is essentially like a much worse problem than having people store WAV files when they cant tell the difference between it and say 320kbps MP3 files in the music world, video being much worse since it is MUCH bigger than music files.

4

For me it’s not the Visual Difference its the compression efficiency. The same video and a similar RF and as slow as I could go with the nvenc encoder still produced a file with a significantly larger bitrate but visually (not based on a scoring engine at all) it seemed the same. I played them at the same time on my dual monitor setup and i could not tell the difference. So as my primary goal was to compress as much as possible, I say it’s not worth it.

The only time it’s really worth it to me is if you’re dealing with h264 and it’s super large files or a direct copy/encode from a blueray that’s at a 160MB/S bitrate and just want to compress it as fast as possible

I am waiting till I get a proper server to not care as much, but as most of my files are just on my 12TB desktop drive, I am not trying to waste too much space. My plan is to make a VM server with 42TB+ of usable storage and make that what I store all my stuff on and keep the important stuff on my desktop / cloud services.

5

Understandable then, I misread you earlier and assumed you shat on the quality on hardware accel like most others I see…

For maximum compression CPU is indeed the way to go. Although from my own tests at least compared with QSV the resulting file sizes are really damn similar at the point where I cannot tell them apart.

For a server rec, I would recommend building your own DIY setup and use Unraid for it, certainly not buying from a prebuild one like Synology which is trying to kill their own company with their recent stunts…

If you also want the server to be able to do encodings, I would look at those cheaper first gen intel arc gpus like the Arc 310 or 380.

I remembered the automatic encoding software, its these.

You can run them via docker on your server so it would work very well with the Arc 310/380 for a low cost.

6

Great write up! I have used ffmpeg to convert h264 to h265, because they were choppy in Heresphere. Now I will look into changing them AV1. Have you tried ffmpeg?

7

I had 2 questions:
First is not using web optimization intentional? It generally is a negligible increase in memory used for improving the streaming performance at no cost to size or quality. I haven’t done mass encoding so not sure if there is some other downside (other than if used with some really slow reading hdd network setup).

Secondly, what has Synology done recently? I have had a Synology NAS for a while but I don’t generally interact with much of their integrated software (just use docker for stuff) so I don’t keep up. I do plan on building a DIY unraid server with more slots eventually, just have to dedicate some time and money to the project.

Anyways, thanks for the guide as I have been considering doing some mass encoding as I usually just do it in a case by case basis so far.

8

Yes, my script is using ffmpeg. And Handbrake is also using ffmpeg under the hood.

Give AV1 a try, but if HEVC and AV1 are both supported but HEVC is choppy then AV1 may be worse since it is more demanding to decode.

I am honestly not sure what it is which is why i have not enabled it, I have had zero issues with my files like this and I do technically stream the files from my NAS to my PC either when I am just watching anime then it is done via MPV connected to the NAS via SMB, or when I use the handy I use stashapp via the browser.

oh boi they have gone down the hill step by step over the years with feature removals here and there, but the biggest dealbreaker is that the 2025 models ALL lack support for 3rd party drives if you use a 3rd party drive from like Seagate and not their “own” models (they dont even make them themselves…) you wont be able to even start/install your nas if you have 3rd party drives, if you have an older syno and want to move to the 2025 models that works however you cannot expand your storage pool if you have 3rd party drives, cannot use M.2 NVME drives for neither cache nor storage pools. And more issues, essentially cannot use 3rd party drives at all.

The problem is also that while the older models still work, we do not know WHEN they will remove 3rd party support on them as well as they have removed features on old/current models in the past such as the USB wifi adapter compatibility.

I personally do own a model from 2023 and this just makes me even more pissed off since now i am worried about if it would suddenly die on me, I would HAVE to buy an older model or I wont even be able to restore the data even if I just put the drives into a DIY build using say Unraid and buying the new models would also not work due to my 3rd party drives…

For your own DIY unraid build I would look at a case from Jonsbo such as the N3 or N5 are popular.
Or you can also get the Sagittarius case from aliexpress which is also very good.

1 Like
9

It just moves the frame index to the front of the file at the end of the encode allowing it to be played faster. The performance boost does depend on how the file is served I believe though.

Ok yeah, this is 1000% a dealbreaker. I am surprised this wasnt a top result when I googled them. This feels like committing company self deletion though maybe they feel they have a large enough base of customers who dont mind throwing extra money around so they dont have to think.

Thanks for the rec, I may start piecing together a pcpartpicker list, though I just expanded my NAS with 2 more drives recently so may wait for next Black Friday to start buying. The Jonsbo cases do look like a nice option as they are actually built for NAS and not just a repurposed tower case.

10

Exactly, from what I read it seems they are trying to cater to pros the problem is that you will never see anyone go for Synology when they are just gonna get something from a much older and reputable brand.

It is truly company seppuku 101.

Yeah the N lineup from Jonsbo is aimed at NAS builds and probably the most popular and used brands for that. With the N3 and N5 being the most popular, the N4 was a dissapointment apparently and iirc the N1 and N2 are just way too small so I never really looked at them.

And look up powerrtop if you want to save a lot of electricity by lowering idle power drain, here it is for unraid.
https://forums.unraid.net/topic/98070-reduce-power-consumption-with-powertop/
But works on other OS as well like TrueNAS.

11

I will be putting proxmox on it and setting up media storage along side my VMs. It will be a raid 5 setup. Proxmox is basically a custom kernel Debian install so i can do all kinds of stuff under the hood and still get the full proxmox VMs and containers benefits. I don’t ever plan on buying a preset built NAS of any kind.

12

I tried proxmox myself on my old PC but had a lot of disconnects from it and was not a fan overall but I do not need to run more than 1 or 2 VMs if at all so Unraid is much more suitable for me and also due to the unraid pools being a lot better for having a low power consumption. Powertop does still work on proxmox though so look it up.

And bookmark this.

1 Like
13

I will be using it as a virtual lab too so that’s why I’m doing that. If I have to I can use a VM to do the shares and pass through a folder on the storage volume. I book marked that a while ago. Thanks for the reminder.

1 Like
14

Can always run unraid in a VM in proxmox so yeah.