Another performance query
Until last month I used Cinelerra-GG on an 11 year old early Intel Q6600 Quad core computer with a modest old nVidia GTX Video card running the Nouveau video drivers. In Cinelerra-GG I would use the X11-XV driver and play up to 2 full 1080P tracks of Audio and Video (without effects) and with proxy editing I was getting along pretty good...
Then I bought two 4K Cameras and since my main Desktop rig was 11 years old and I was moving up to higher resolution editing I decided it was time to upgrade my main Desktop PC so I built a new machine with a 24 Core/48 Thread AMD threadripper CPU and an AMD RX 570 Video card and expected that world domination was next on my agenda... not so fast!
I opened up an existing 1080P Cinelerra project that didn't make my old machine sweat too much and I found with some horror that Cinelerra (or at least the new computer) was performing MUCH worse than it ever did on the old machine to the point that I was having difficulty running the project with proxies... The UI was extremely slow with several seconds of catch up delay when hovering the mouse over things and it was taking forever to draw the thumbnails and waveforms, scrolling the timeline with the bottom slider would induce freezeups for several seconds at a time so I though I'd better look at some of the settings..
I went into the preferences and Playback A & B were at their default X11 settings, I tried X11-XV since that had worked on my old PC and it seemingly made no difference so I moved to OpenGL and that improved things about 50% but although the UI was less sticky things were not nearly as smooth as they had been on the old computer. I double-checked that I hadn't accidentally enabled "single processor use" and saw that indeed 48 threads were showing in the CPU count so that wasn't the issue.. To get the project I was working on finished I disabled thumbnails (which I've never done before) on the Video tracks and that gave me another boost of performance but something is very amiss here..
I also noticed that rendering of proxies and rendering the final video is far slower than in should be, with 24 cores there should be a puff of smoke after the nanoseconds of rendering for a 5 or 10 minute video and I'm not even getting 1080P renders in realtime.. ( a 7 minute video just took 12 minutes to render) If I didn't know better I would swear that the version of ffmpeg in this build (March 2020) didn't have multithreading enabled. Decoding and encoding are far too slow..
So what do I need to send or attach to debug this?
"Speed improvements for line drawing using Bresenham’s line algorithm implemented now in
Cinelerra (motivated by LFOM forum notes). " Since this was for Ubuntu 19 which is based on Debian, perhaps this will help here too. I will have GG create a static tar build tomorrow for Debian 10 to see if this helps. It was not a problem on Arch.
When and If you have time, please see if the speed improvement mods using Bresenham line drawings improves your speed. A Debian 10 static tar is at:
On a whim I also tried the Ubuntu static build you made for LFOM but my Debian system doesn't satisfy it's libc6 requirements so it doesn't launch at all. I'm very puzzled about the playhead issue and it's not allowing me to playback the project at all to even see if the line drawing optimization is helping.. As you could see in my screencap Cin launches and opens the project like lightning..
Do you have an AV Linux download there to see if you are getting the same problem with that static build?
We can put AVLinux here to use? I missed your previous note but will test the Debian 10 we created to see if we have that same playhead problem.
Booted the Debian 10 system here and loaded the same static tar we created the other day. We could not generate the playhead problem at all but there is so much going on in your recording that we do not know about that it is impossible to create the same situation. Is it at all possible to send the session data? or put it somewhere for us to grab? ([email protected]).
OK some good findings..
In my previous efforts to improve performance my Audio preferences had been set to PulseAudio for Playback A and B and this seemed to be the problem with the static build.. I went into the preferences and set it back to ALSA and then the playhead worked again so either the static build doesn't like the PulseAudio driver (or perhaps its not enabled at all?) So good news #1!
I opened the project from my screencast again, re-enabled media view on the tracks and it seemed to play 4 HD Video tracks and one WAV Stereo Audio track (not in proxy mode) quite smoothly even when jerking around the scroll at the bottom of the screen. I would say it's behaving much more "freely" than it was before so even though I don't have an installed CPU meter to correlate the 2 there is definitely an increase in performance in daily use case scenario of how I would normally use the program.
To summarize I got the first big performance jump by selecting OpenGL Video driver and I would say the new line drawing algorithm was what closed the gap on getting the program to perform "normally". If I were to get you some CPU usage data between the 2 Cin versions what CPU Utility would you recommend so we're all on the same page? htop?
Oh, we never thought to use PulseAudio. Have tested again with all kind of strange configurations and had no problems. Maybe sending your $HOME/.bcast5/Cinelerra_rc file and just the session file would help us to set it up like you did (do not need the video files at all). There must be a bug somewhere related to PulseAudio.
"top" is the statistics gg likes to see.