BD Rebuilder Beta - Bug Reports Only

[jdobbs]

Good to hear... I wasn't expecting that one to get fixed so fast. I'll test it tomorrow. Sigh... now I need to go back through my code and remove all the changes I made for outputting to .m2ts. It resulted in a lot more changes than I'd expected, and I was still testing it and finding stuff. Two steps forward and one step back.

[cartman0208]

Originally Posted by Mike-uk

yep can confirm it seems fixed stayed arounf 700 Mb, not tested on a disc as not got any RWs

--avhw --vpp-pad 0,276,0,278 -i F:\Movies\uhd2019\BDMV\STREAM\00014.m2ts --codec hevc --preset quality --profile main10 --output-depth 10 --repeat-headers --chromaloc 2 --colorprim bt2020 --transfer smpte2084 --colormatrix bt2020nc --master-display G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(10000000,1) --qp-min 0 --vbrhq 0 --vbr-quality 25 --sar 1:1 --aud --pic-struct --vbv-bufsize 45000 --max-bitrate 48000 --gop-len 24 --slices 4 -o "D:\test\VID_00000.hevc

RTX 2070 Super

encoded 175243 frames, 62.57 fps, 12390.37 kbps, 10795.88 MB
encode time 0:46:40, CPU: 7.2%, GPU: 6.3%, VE: 99.2%, VD: 18.1%
frame type IDR 7302
frame type I 7302, total size 1246.97 MB
frame type P 43811, total size 4712.77 MB
frame type B 124130, total size 4836.14 MB

I can also confirm, that there is a stable RAM usage.

And the best part ... it doesn't seem to matter which turing chip ... i tried with Mike's settings and got around 61 fps on my 1660ti, while the card costs about half
Maybe a Titan would make a difference ... but people could buy a decent car in that price range

BTW: I think with your settings you actually add some height to the video ( pad: [3840x2160]->[3840x2714] (right=0, left=0, top=276, bottom=278) ) ... forgot some dashes?

@jdobbs: why not use the code from 0.61.0.5 and start over?

[Mike-uk]

Originally Posted by cartman0208

I can also confirm, that there is a stable RAM usage.

And the best part ... it doesn't seem to matter which turing chip ... i tried with Mike's settings and got around 61 fps on my 1660ti, while the card costs about half
Maybe a Titan would make a difference ... but people could buy a decent car in that price range

hmm yeh interesting encoding fps is the same

[gonca]

Dedicated encoding hardware is probably the same

[jdobbs]

Originally Posted by cartman0208

@jdobbs: why not use the code from 0.61.0.5 and start over?

You underestimate how much code it took to integrate it in. I also have implemented a host of other changes since that version. Anyway, I've already reversed most of the code I'd written for it.

[Mike-uk]

Originally Posted by gonca

Dedicated encoding hardware is probably the same

hmm yeh quite possibly found this on a forum

Turing NVENC is slower at higher quality settings than Pascal NVENC. The NVIDIA application note for NVENC states that at highest quality you should expect no more than 130fps at 1080p (or around 33fps at 4K).

[gonca]

Extra switches (cmd line options) available
NVenc is on a dedicated chip, same chip across the "line" (family) of cards
https://developer.nvidia.com/video-e...support-matrix
Chip number and number of encoding chips per card

[MrVideo]

Originally Posted by jdobbs

You underestimate how much code it took to integrate it in. I also have implemented a host of other changes since that version. Anyway, I've already reversed most of the code I'd written for it.

Don't you use Revision Control Software? Programs like Git, CVS and SVN (subversion, which I've managed at a work place) would have allowed you to back out those changes. But, as you've pointed out, a little late now.

[jdobbs]

Originally Posted by MrVideo

Don't you use Revision Control Software? Programs like Git, CVS and SVN (subversion, which I've managed at a work place) would have allowed you to back out those changes. But, as you've pointed out, a little late now.

I am my revision control software.

[cartman0208]

Originally Posted by Mike-uk

...
Turing NVENC is slower at higher quality settings than Pascal NVENC. ....

I assume, that's because of the B-Frame capability of Turing ... which I don't want to miss, because there is the bitrate saving ...

[jdobbs]

Originally Posted by cartman0208

I assume, that's because of the B-Frame capability of Turing ... which I don't want to miss, because there is the bitrate saving ...

I'm with you. I want the B-Frame efficiency.

[Mike-uk]

jdobbs do you think you'll add true CPU first pass, then GPU second pass ?? seems the only way to get true 2 pass or maybe add it as an option so people can choose what they want,

maybe get some comparisons of CPU first pass and the lookahead feature to see if theres any quality difference

[MrVideo]

Originally Posted by jdobbs

I am my revision control software.

[Mike-uk]

Nvidia 3xx series rumoured to be released in aug, wonder if they have improved NVenc, and x266 being developed ohhhh its like xmas

[Mark_Venture]

Originally Posted by cartman0208

...
And the best part ... it doesn't seem to matter which turing chip ... i tried with Mike's settings and got around 61 fps on my 1660ti, while the card costs about half
Maybe a Titan would make a difference ... but people could buy a decent car in that price range

So are you saying that as long as its card with a Turing chip, the Cuda cores, clock, etc. doesn't really impact the FPS much when using NVEncC? They're all about the same?

[cartman0208]

Originally Posted by Mark_Venture

So are you saying that as long as its card with a Turing chip, the Cuda cores, clock, etc. doesn't really impact the FPS much when using NVEncC? They're all about the same?

I was wondering, too ...
But i used the same settings as Mike-uk, only a different source (with the same dimensions)
I could try different sources, but my experience so far was an almost unnoticable FPS change, even when changing presets.
We need someone with RTX 2060 / 2080 to verify

[jdobbs]

Originally Posted by cartman0208

I was wondering, too ...
But i used the same settings as Mike-uk, only a different source (with the same dimensions)
I could try different sources, but my experience so far was an almost unnoticable FPS change, even when changing presets.
We need someone with RTX 2060 / 2080 to verify

I'm seeing quite a bit of difference between presets when running against the same source. The "performance" preset, for example, is running much faster than "default" or "quality". It can also change based on whether I am using --vbr or --vbrhq.

For example, on a specific UHD source I'm using for testing with preset "performance" and "--vbr" I am getting 127.49 fps -- the same source with "quality" and "--vbrhq" it is doing 77.67 fps (with an otherwise identical command line).

I am also seeing fluctuations depending upon the size (on disc) of the source. An imported 10GB source M2TS encodes faster than a full 60GB untouched source.

The largest factor to speed I've seen (as you might expect) is whether I am using --avs as the input type as opposed to --avhw or --avsw.

[cartman0208]

Ok, I didn't do extended tests ... only constant QP on a small source file ... I wanted to check out resulting file sizes

[Mark_Venture]

Originally Posted by jdobbs

I'm seeing quite a bit of difference between presets when running against the same source. The "performance" preset, for example, is running much faster than "default" or "quality". It can also change based on whether I am using --vbr or --vbrhq.

For example, on a specific UHD source I'm using for testing with preset "performance" and "--vbr" I am getting 127.49 fps -- the same source with "quality" and "--vbrhq" it is doing 77.67 fps (with an otherwise identical command line).

I am also seeing fluctuations depending upon the size (on disc) of the source. An imported 10GB source M2TS encodes faster than a full 60GB untouched source.

The largest factor to speed I've seen (as you might expect) is whether I am using --avs as the input type as opposed to --avhw or --avsw.

you mention different presets. I was wondering about using the same source, with the same preset, across different cards.

cartman0208 indicated using the same command line that Mike-uk posted, resulted in similar FPS using different video cards (GTX1660ti vs RTX 2070 Super) based on their posts. And using the same string Mike-uk posted on a UHD rip on my PC also came out with the same FPS (using GTX1660ti here).

When all is said and done, it will be interesting to compare results if we could each run tests on our respective systems using the SAME pre-sets and (if possible) the same source.

[gonca]

Originally Posted by Mark_Venture

you mention different presets. I was wondering about using the same source, with the same preset, across different cards.

cartman0208 indicated using the same command line that Mike-uk posted, resulted in similar FPS using different video cards (GTX1660ti vs RTX 2070 Super) based on their posts. And using the same string Mike-uk posted on a UHD rip on my PC also came out with the same FPS (using GTX1660ti here).

When all is said and done, it will be interesting to compare results if we could each run tests on our respective systems using the SAME pre-sets and (if possible) the same source.

https://forum.doom9.org/showthread.p...67#post1917767