I am currently working with models containing millions of faces, and the time it takes to import these into Blender is killing me. The models I am creating cannot be decimated without losing needed detail, and the entire model changes with each iteration (i.e. I create a completely new models on a regular basis).
Once imported the viewing speed is fine, but importing from .ply files takes an obscene amount of time. Here are a few speed tests, run on a Win7 machine with 32GB ram & Intel 2600k processor:
Blender importing binary .ply file (86MB): 89 seconds
Blender importing text .ply file (142MB): 152 seconds
MeshLab importing binary .ply file (86MB): 3 seconds
MeshLab importing text .ply file (142MB): 6 seconds
I’m aware that having the import code written in python would probably be somewhat slow, but 30x slower is a lot. I’ll be working with files easily 5x larger than that tested as well.
Does anyone know an appropriate place that I could put in a request for import speed improvements? This isn’t a bug, and I haven’t found a good place to post something like this.
since the importers are written in python, the performance is quite capped. Due to Python’s GIL, it can’t utilize the full power of your quadcore cpu. My dualcore i5 has 4 virtual cores and python runs at 25% cpu usage max, yours will certainly use 1.7 ghz of your 4 * 3.8 (*2 ?) ghz at most.
Thus i would expect it to run 8x slower (maybe only 4x) just because of the missing multicore support. Then it’s python and no optimized C/C++ code, probably 2-4x slower. And finally, blender python has to interact with blender’s c-coded api… In total, 30x slower seems reasonable to me.
Some optimizations to the import script might be possible, but i wouldn’t expect too much of a performance gain.
You should rather look into the Bratwurst branch, there are builds around with assimp integration (= importers for OBJ, DAE, PLY etc. written in C/C++). The performance should be a lot better.
May try and see if I can figure out how to use IRC, but so far I’ve just seen random people on there.
I’m guessing that Python just uses a single core for importing, which would work out to using 12% of the potential computing power. Of course I kind of doubt that other software packages are multi-threaded for importing, and considering my saving the file from c# takes far less time (while also parsing gigabytes of input data single-threaded no less), I’m guessing the slowdown is just due to the fact it is Python.