Hi,

This is a summary of a request i sent to the list the 18th of February,

The problem in short is that i generate very larege meshes (50 Mb) and have
a (too) limited amount (288 Mb) of memory. When rendering this resulted in a
large amount of swaping - cpu utilization went down to 10% - for full
description see the original thread, use advance search for author idoit).

After rearanging my code as corroborated by Erkki (thanks again, by the way)
the CPU utilization went up to 95% percent. I was wery happy at first but
noticed that  the image rendered _very_ slowly. After activating my brain i
realized that it was not enough to change the order of the vertex vectors -
a change of the vertex indices was also nessesary (this is partly why my
summary has been delayed).

In any case after doing all the corrections the CPU utilization went up to
20-30% (from the original 10%). The conclusion is that indeed the rendering
speed does matter for
how triangles are laid out in memory - again see Erkki's description in the
previous thread - with respect to the tracing ray but depending on your
schene the improvement may not be as big as hoped for. In my case the
swapping problem only went away partially.

My final conclusion is that layout of the triangles in memory definitely
matters  but perhapps bying more memory might be a better solution in those
cases when a schene is rendered from different directions (in my case I
allways render from the same place).

Perhapps it would be possible to have the PovRay engine optimize the data
layot in memory automatically, but the payoff in rendering speed is
probably not worth the effort, since i can imagine that this is qute a
challeange - especially if there exists several light sources.

Best regards
Jan

Post a reply to this message