How much speed improvement by cooperative rendering?
Posted: Fri Nov 04, 2005 2:16 pm
Hi, I read all topics about cooperative rendering (one frame on multiple computers, for short: coop) and I am still confused. Some questions:
In older topics some people claim it doesn't work at all. Could be this was on the alpha or is this still a problem on the beta?|
As I understood the sampling level on the individual machines doesn't get higher as they don't exchange info while rendering. At the end however the two solutions get merged and the total of used rays (or samples) is higher thanks to the extra cpu time. Correct?
What is the effective speed up by using extra pc's? I would expect it to be nearly linear as there is hardly any overhead involved, but somewhere I got the impression that people claim this to be incorrect.
I even read Tom's nice explanation of the algorithm as he understood it, but it was still not clear to me what the speedup would be.
The questions are relevant as a large network would make maxwell useable for actual production work and as long as the beta's are cheap...
In older topics some people claim it doesn't work at all. Could be this was on the alpha or is this still a problem on the beta?|
As I understood the sampling level on the individual machines doesn't get higher as they don't exchange info while rendering. At the end however the two solutions get merged and the total of used rays (or samples) is higher thanks to the extra cpu time. Correct?
What is the effective speed up by using extra pc's? I would expect it to be nearly linear as there is hardly any overhead involved, but somewhere I got the impression that people claim this to be incorrect.
I even read Tom's nice explanation of the algorithm as he understood it, but it was still not clear to me what the speedup would be.
The questions are relevant as a large network would make maxwell useable for actual production work and as long as the beta's are cheap...