26

Re: Some results from QTGMC & HEVC encoding tests

^ Yeah, we were typing at the same time smile

27

Re: Some results from QTGMC & HEVC encoding tests

Latest dev version crashes right before x265 gets to work when using Vapoursynth. DebugOutput attached.

28

Re: Some results from QTGMC & HEVC encoding tests

Clear your tool cache, from the looks of it neither vspipe nor vsedit were found on your system.

29 (edited by JDiG 2017-02-15 11:16:34)

Re: Some results from QTGMC & HEVC encoding tests

Ahhh. I actually noticed that yesterday when testing Vapoursynth. I guess I need to move those out of Vapoursynth directory and into Hybrid's root directory?

Moving the files and clearing tools cache resulted in Hybrid setting vsViewer to ".\Vapoursynth\vsedit.exe". Should I point it to vsViewer instead?

30

Re: Some results from QTGMC & HEVC encoding tests

No, those should be located inside the Vapoursynth folder.
Like I wrote: Clear your tool cache. (Config->Tools->Clear all cached tool infos)

Cu Selur

Post's attachments

Attachment icon tools.png 29.62 kb, 13 downloads since 2017-02-15 

31

Re: Some results from QTGMC & HEVC encoding tests

Okaydokey, files are in correct directory, cache's been cleared and tool paths match ones in your png. But something else went kablooie, see attached DebugOutput.

32

Re: Some results from QTGMC & HEVC encoding tests

"C:\PROGRA~1\Hybrid\VAPOUR~1\vspipe.exe" "F:\HYBRID_TEMP\encodingTempSynthSkript_11_22_05_8510.vpy" - --y4m | "C:\PROGRA~1\Hybrid\ffmpeg.exe" -y -loglevel fatal -threads 16 -r 30000/1001 -f yuv4mpegpipe -i - -an -sn  -vsync 0  -pix_fmt yuv420p  -f yuv4mpegpipe - | "C:\PROGRA~1\Hybrid\x265.exe" --preset slow --pme --input - --y4m --profile main --no-high-tier --level-idc 5.1 --bframes 8 --ref 6 --crf 25.00 --dynamic-rd 0.00 --vbv-maxrate 40000 --vbv-bufsize 40000 --aud --deblock=-2:-2 --repeat-headers --range limited --colormatrix bt709 --output "F:\HYBRID~1\11_22_05_8510_06.265"

call looks fine,..

x265 [error]: unable to open input file <->

sounds like there is some problem with the decoding.
Does the Vapoursynth preview work properly?

33

Re: Some results from QTGMC & HEVC encoding tests

Selur wrote:

Does the Vapoursynth preview work properly?

Ran it for the first time now... no. I get a weird error message:

https://s1.postimg.org/rob4s5swv/VPY_preview_error.png

34

Re: Some results from QTGMC & HEVC encoding tests

pack mal in Hybrid\Vapoursynth\platforms die angehängte qwindows.dll und versuch dann noch mal

35

Re: Some results from QTGMC & HEVC encoding tests

Okay, did that and got this from VPY preview:

https://s22.postimg.org/u5u9ld1ht/VPY_preview_error_02.png

36

Re: Some results from QTGMC & HEVC encoding tests

Okay,... will look into it.

37

Re: Some results from QTGMC & HEVC encoding tests

try if it helps to: close Hybrid, delete the Hybrid\vsscripts\__pycache__ - folder , start Hybrid
(will fix the download to remove the __pycache__ -folder and add the proper qwindows.dll, so others will not have that problem)

Cu Selur

38

Re: Some results from QTGMC & HEVC encoding tests

Damn, that didn't fix the issue,.. -> looking into it

39

Re: Some results from QTGMC & HEVC encoding tests

Seems like there is a problem with the portability of Vapoursynth.
-> looking into it, but this will take a while smile

40

Re: Some results from QTGMC & HEVC encoding tests

^ No worries!

41 (edited by JDiG 2017-02-15 13:49:35)

Re: Some results from QTGMC & HEVC encoding tests

In the meanwhile, here are some results of my tests doing QTGMC, with GPU taking care of the encoding task:

https://s10.postimg.org/g1o5dge3d/x265_enc_results_gpu.png

The 889 frame source clip is a 30-second portion of the original 6580 frame video. It looks like that even though GPU rendering uses very little CPU, raising Edi and MT thread count over a certain limit begins to slow the process down. 12-13 Edi/MT threads appears to be somewhat optimal for my system, based on this limited sample of results.

What I found most interesting was the last run, done using VBR-HQ instead of Constant Quality. I was suprised to see that the visual quality of the output file was not hugely worse than that of the comparison output I ran using x265 (slow preset) at CRF 25 (~8.6 mbps). GPU HEVC encoding still demands a higher bitrate than x265 to reach similar visual quality, but the margin is getting smaller. It will be interesting to see what the next generation GPUs are capable of.

42

Re: Some results from QTGMC & HEVC encoding tests

x265 even at faster presets should look noticeable better, due to the use of b-frames which gpu encoders lack or can only use sparsely.

43 (edited by JDiG 2017-02-15 15:31:32)

Re: Some results from QTGMC & HEVC encoding tests

Selur wrote:

x265 even at faster presets should look noticeable better, due to the use of b-frames which gpu encoders lack or can only use sparsely.

That is very true, based on what little testing I've done using fast presets. BTW, unless I read the info too fast, it looks like AMD's new GPUs like the RX480 can actually do b-frames. I'm not 100% on that, however.

44

Re: Some results from QTGMC & HEVC encoding tests

Yes, but afaik even then those b-frames are far worse than those of x265 and x264 since they are not adaptive enough.  smile
-> I don't expect the general statement that gpu encoding requires a decent amount more data rate to archive a similar quality than a software encoder to be wrong any time soon. wink

45 (edited by JDiG 2017-02-15 15:48:26)

Re: Some results from QTGMC & HEVC encoding tests

I've got a RX480 here so I could do some testing. Not sure if Hybrid supports all the goodies in Rigaya's VCEEnc yet..?

46

Re: Some results from QTGMC & HEVC encoding tests

Haven't tested it myself, but in theory Hybrid should support b-frames for HEVC as soon as 'check-features' reports:
"Bframe support: yes" <- I suspect that is what would be reported.

47

Re: Some results from QTGMC & HEVC encoding tests

Selur, regarding the new dev version and Vapoursynth: Preview worked OK, so I ran an encoding job using the same test clip I sent you. No problems during the job, finished with an average speed of 3.65 fps using x265 Slow (MSVC-compile + 6 refs, 8 b-frames) preset. Didn't touch Pools or Slices yet.

However, the output's kinda messed up. It plays at correct speed, but looks very much like the field order's been wrong during processing.

48

Re: Some results from QTGMC & HEVC encoding tests

It plays at correct speed, but looks very much like the field order's been wrong during processing.

so output is jumping for and back?

49

Re: Some results from QTGMC & HEVC encoding tests

Selur wrote:

It plays at correct speed, but looks very much like the field order's been wrong during processing.

so output is jumping for and back?

A bit, yes, it looks almost like a big shimmering effect where there's movement.

I ran a quick test using NVEnc, and set (Field) Order to 0 in QTGMC (Vapoursynth)'s settings so the source gets processed as if it's Bottom Field First even though it's really Top Field First. Encode job finished OK (and it was almost 10 fps faster than using Avisynth), and the output looks OK. Field order is correct.

50

Re: Some results from QTGMC & HEVC encoding tests

Ahh,.. check the Vapoursynth script used, loading the source it seems like there's a small bug.
For me the QTGMC call is:

clip = havsfunc.QTGMC(Input=clip, Preset="Fast", TFF=False)

where TFF should be True!
Is that for you the case too?