rollin ocean
This is just a debug note, not a complain.
The session crashed my computer twice in a row. As soon as the rendering started, my RAM filled up to about 3gb, all 4 cores were at a 100%, then complete lock up, no mouse move or keyboard input possible - reset button ;)
On the second run I let it that way to see if it's commin out of that state after a while, but it didnt (let it run bout 10 minutes)
PC is a AMD X4, 4gb RAM, Kubuntu 64bit, recent boinc client.
Cheers, Manu.
How should we select memory usage when submitting a job?
I manually ran a workunit for "Rolling Ocean" (with 4.77) and saw it peak out at 476mb. While a successful task result shows 1.1gb.
I also saw that my sumbission (100 Points of Light) was changed from 256mb to 1gb. I had tested this and it peaked out at 116mb.
I'm confused how I should caclulate my memory usage on new jobs. Should the blender value be doubled or tripled?
I'm asking myself the same thing, but I think this partly could be a blender issue. Just before submitting my caprica scene I did a test render with the final quality settings and it peaked to 1,5gb, came out fine. Later I did another test render, nothing changed, my computer loaded up to 3gb ram and 5gb swap - almost froze. Had to kill the blender task -> blender memory leak?
EDIT: or maybe an ubuntu issue in my case, dont have a windows or mac machine here to confirm that.
You can specify the memory setting at one point of the uploading process in the Renderfarm distro of Blender, or at some point of the uploading process via the web form. That setting hardly matters for you as I will be doing a test render and increase the memory requirement if needed. The issue is that BURP is for some reason not using this value properly. There probably needs to be a feature that checks if a computer can receive 8 tasks with it's memory (Sure, I have 6GB of memory, but 6GB of memory != able to run 8 * 1GB tasks)
Also, the job dispatcher should take into account that ie. windows32 systems don't really work well with >1.3GB RAM requirement (1.3-ish GB is maximum per process, larger than that = crash). I'm not sure what exactly the limits are for the other OSes, but I'd imagine something similar.
That is probably due to the very high memory consumption. All your cores are starting a render and 4 cores are eating more than is available. This problem has been noted and discussed with jbk, and hopefully we can do something for this soon. For now, decrease the amount of cores you donate to 1 or 2 for this session.
- Jesse Kaukonen, www.jessekaukonen.net