View Single Post
Old 14th August 2019, 09:27   #17134  |  Link
duffbeer
Registered User
 
Join Date: Mar 2019
Posts: 40
Quote:
Originally Posted by byteshare View Post
I've had similar results with AVC/HEVC inputs to 10bit x265 using AVISynth filters such as QTGMC and Megrain2. Without AVIsynth filters speeds seem close enough to the same that I can't tell, so I don' think it could be related to the updated x265.
Using the most recent update (including LSmash update), it seems to slow down the decoding process because extracting a frame is taking way longer when adding a job and using a lot of CPU while doing it.
--Maybe skip that step until someone clicks Edit > AviSynth? Or had Detect boarders on for cropping?
Unless something changed with FFMPEG, but I don't see an update to that since April of this year.
Another thing is I've noticed with filters my second Encoding Server on the same machine is taking way longer to start encoding than the first one, but with only one Encoding Server I'm not getting 100% CPU.
I'm not sure what I could send to help figure this out.
I tried a test with 1.24 vs fully updated 1.25 RipBot.
Same x265 version (one from the newest version of RipBot), same x265 profile, same filters for the same AVIsynth custom script.
For time used just one Encoding Server to encode the same chunk.
Custom:
Code:
video=QTGMC(video,Preset="slower",FPSDivisor=2,NoiseProcess=0).ConvertBits(16).ConvertToStacked()
.SMDegrain(tr=3,thSAD=400,thSADC=150,refinemotion=false,contrasharp=true,plane=4,pel=2,
prefilter=4,Truemotion=False,chroma=true,hpad=32,vpad=32,str=2,amp=1,lsb_in=true,lsb=true,lsb_out=true)
.ConvertFromStacked().ConvertBits(8).FineDehalo().FastLineDarkenMOD4().LSFmod()
(Added line breaks to keep the text from spanning too much on smaller screens.
Results:
confirmed extracting a frame was faster in 1.24
but for FPS of encoding:
1.25 0.20FPS
1.24 0.43FPS
...but, 1.24 was actually much faster than just FPS because the FPS isn't reporting correctly in 1.25 since the encoder isn't actually processing frames the whole time. It looks like the decoder stage is hanging and not passing frames to the encoder as quickly as in 1.24
--Sidenote, in 1.24 it is correctly splitting up the chunks for this file into 58 parts rather than 29 parts (but twice the number of frames roughly) and giving me the option for deinterlace drop down in the AVISynth settings.

Could the decoding slowness have something to do with the changes in the write buffer?
I think you may have misunderstood my problem. The issue I was referring to are with the encoded output file.
Encoding speed seems a little slower but the biggest problem is when I play back a file that was encoded with 1.25 using MDegrain2. The first 10-20 mins play back OK but then it plays at roughly 2fps. The audio is not affected and continues to play correctly.
I have tried the same source files on 1.24.1 and there is no problem so I believe it has something to do with LSmash.

The other big problem since 1.25 is that it takes roughly 2 minutes to auto crop. Generating a new preview frame also takes at least 1 minute. If I do exactly the same thing with 1.24.1 it crops in about 3-4 seconds.

It would be useful to know if anyone else has these issues. I've never had any serious issues with auto updates in the past, but this is the first time I'm going to have to ignore the latest version as it's virtually unusable in it's current form.
1.24.1 still works perfectly for me.

I've been using RipBot since 2012 and I think it's fantastic but the latest update seems totally broken to me.

Last edited by duffbeer; 14th August 2019 at 10:34.
duffbeer is offline   Reply With Quote