PDA

View Full Version : Hypervoxels, poor performance on Multi-Cores



lserlohn
05-28-2008, 07:52 AM
I don't know why Newtek didn't do something on Hypervoxels, which performance is too poor on Multi-Core CPUs. Particals is one of the most processes which require massive calculations, but 3/4 of my Quad Core should be waitting while the only one core is busy calculating...

If it is difficult to divide the Hypervoxels to multi-threads, I think maybe it can calcuates several frames at one time.

clagman
05-28-2008, 02:57 PM
On my machine rendering hypervoxels will utilize all 8 cores (a good thing too). The only time I can see that I'm not using all 8 is when using certain pixels filters (which I do in post 99% of the time anyway) or when using objects with lots of layers and/or geometry that requires a lot of setup time. Make sure you have the proper number of threads selected in your rendering options.

Stooch
05-30-2008, 04:14 PM
the best way to increase the speed of hypervoxels is learning how to use them correctly. Not using too many particles and not making the voxels too big. the trick is finding the best ballance between the number of particles and the size of voxels. I use sprites for most of my voxel work and usually they render VERY fast, faster than anything comparable infact.

too bad voxels dont compare to much since they arent true CFD, greatly limiting the realism of the output and requiring ALOT of tweaking and experimentation to get things to look half decent.

clagman
06-02-2008, 07:00 AM
Speaking of realism, I wonder why NT decided to be rid of the fantastic feature "distance between particles". Something I really miss from earlier versions...