I recently switched from ubuntu to arch and i'm experiencing very slow render times. i used to be able to render in almost real time, meaning my video is 8 minutes long kdenlive would render it even with affine and composites and transitions in almost 8 or 9 minutes. I just tried a video that was 9 minutes and 16 seconds and it took 36 minutes to render. According to pacman the following versions are installed
kdenlive 0.9.10-1
mlt 0.9.2-3
ffmpeg 1:2.4.3-1
frei0r-plugins 1.4-2

i noticed htop was reporting only 1 core at 100% and the other 3 cores were bouncing from as low as 15% to only as high as 30%, shouldn't melt being using all 4 cores at 100%? can anyone help me figure out why rendering has gotten so slow please? Im on arch using 3.17.2-1-ARCH kernel and kde 4.14.2

Hi ubu, I've seen your post to the other board.

Although I am using the same versions, rendering to h.264 utilizes my i7-3632qm by constantly >90% for 90% of the rendering time. Although I am not using i7z for accurate CPU reporting, my fan and core temp of 80°C makes me believe the values seen may be quite right. I have set the rendering threads to 3 (hoping to keep one core for surfing while rendering). Rendering h.264 input from my camera shot with 24MBit takes 1-1.5 times as long as the video is. I am using an external USB 3 drive as project homebase and clipstore.

A first step for better anlysing could be to install i7z which is said to report CPU usage better than other tools. But the rendering times show the truth... But nevertheless it may give hints which screw has to be turned for faster rendering.

EDIT: Well, the things stated in second paragraph were right when rendering the last time. Currently rendering with h.264 high profile with 34-80% CPU usage, rendering takes about factor 1.5 to realtime.

so does htop or any other tool that shows cpu utilization show that 3 cores are being used? i could live with 1.5 x realtime but when i get 2x and 3x realtime something seems wrong. another thing my friend and i realized was that doing a composite (say track 1 onto track 2) on top of an affine (now maybe resizing track 3 onto track4) really affects rendering time.

On my last notebook I had i7z installed - which shows the utlization, clock multipliers and idles per core. kdenlive used to utilize all of the cores once in a while - but all in all about 3/4 of full load. To be honest - as the render times are alright, I have not checked the utilization on this machine with i7z.

The results of i7z and the system tools cpu usage differed quite a lot. i7z has been designed for core i7 and newer - give it a shot to check your core utilization exactly when rendering. It could be that lots of compositing makes your rendering less threadable, resulting in less core utilization.

are you seeing the same behavior if you put a composite over another composite like i described above? to your point i don't care about core utilization as much as i do about it taking 3 times as long to render a project than it used to. i guess im of the thinking that something was changed in melt where it's only using 1 thread or 1 core since i'm running kdenlive-git from the AUR

Ah, well... That may be the difference between our systems - I am running kdenlive & friends from Community/Extra repos.

i'll remove kdenlive-git and use kdenlive and friends from the regular repo's to see if that solves the issue. thanks

ok, so I installed i7z and i7z-gui and during video rendering 1 core ever goes to 100%, the other 3 stay a lot lower%. I have 4 threads selected, here's the command from ps aux | grep mlt

Code: Select all

/usr/bin/kdenlive_render -erase -pid:8808 /usr/bin/melt atsc_720p_30 avformat - /tmp/kde-ubu/kdenlivewS8808.mlt file:///home/ubu/kdenlive/make%20games%20launch%20on%20correct%20screen/force%20games%20to%20launch_new1.mp4 properties=x264-medium vb=4000k pass=2 ab=192k progressive=1 threads=4 real_time=-1

here's a screenshot taken at just the right time

Does the behaviour change, when selecting 3 threads? What are your cpu specs?

Today I have rendered a 7 minute 1080p project to 720p in 10 minutes (h.264 high profile). While rendering top showed various core utilization: One core at 100%, then to about all (8) cores 60%-90% and all states between. But in difference to an earlier version of kdenlive or my supercaramba plugin measuring core utilization, the plugin only showed 30% load for the complete core i7. As Dan Dennedy postet a while ago, kdenlive will definitely not utilize all threads by 100% all the time, so some alternations are normal and sometimes better, sometimes worse.

To be honest, I have no idea (and based on my experience I doubt) if the rendering could be any faster on my i7 with 4 cores and 8 threads (2.8GHz max clock speed when using 4 cores).

I'm having the same problem but I think this is a regression. Check this thread viewtopic.php?f=265&t=123194