15min is a little bit long...jamnique wrote:Is there anything that can be done hardware-wise to significantly speed up exports from Softimage? Would a RAM upgrade from 4 to 8 gigabytes or buying an SSD have a noticeable positive impact? I'm dealing with massive amounts of strands and exports take very long, like 15 minutes often followed with a total system freeze, so I'm guessing I am badly underresourced. Also, out of curiosity, is there room for improvement in the OBJ exporter itself? Can it be multithreaded and would it be worth it, or would hard disks slow everything down anyway?
BTW, strands seem to work better and ... sorry for off-topic ;]
On my C2Q 6600 with 8GB ram i can export around 1mio tris in ~12sec.
This can be a single mesh with 1mio tris, a pointcloud with 83333 cubes or ~2700 strands with a cube which have 24 subdivisions on V.
When i calculate you 15min export, i would export 75mio tris.
To export strands, i create internal for every strand a bspline with the strand positions.
Then i grab the y positions from the instance shape, remove doubles, rescale them from 0 to 1 and use this values as U to read the positions, normals, binormals and tangents from the bspline.
I calculate the U only if the shape is changed. That means, if you use only one shape, i calculate it only one time.
If you use multiple shapes, it can be, that the plugin must calculate the u positions for every strand. That can cost some time.
An other thing wich can cost time is the subdivision of your strands, called number of segments in Softimage.
Because i use internal a bspline, the strand can have minimum 1 seqment. The rest of needed points will be calculated so that my bspline will have 3 points.
My hint is to use so few segments how possible. It makes no sense to use 1000 segments if you can have nearly the same result with 100 segments.
Here a example:
You should also avoid to rotate the shape because then the plugin must calculate more U values.
Example:
For simple particles, i have a routine which sorts the shapes, so that the plugin export first all particles with shape one, then with shape two and so on.
This will save time to read in the shape geometry to the number of used shapes. The plan is, to implement this also for strands.
The use of a SSD will not speed up the export speed, maybe more ram will have an effect.
Also the use of multithread export will haven´t much speed improvements, because the logic and managment behind it will negate the effect...
Hope this helps you a little...
face