Instance Optimisation

[Anonymous]

Hi,

I was wondering if there was any way to speed up time to first pixel when rendering a large instance object set.

In this particular example I have 1 million points and about 16 attribs per point. The point cloud is about 114MB on disk and I'm instancing 1 object.

It takes about 2 mins to get to first pixel. If I save out the instance setup to ass it gives me a file of about 1.3GB and saves about 10 seconds for the time to first pixel.

I guess the ass file just has all the instance info baked in seeing as it's huge and investigating the text itself.

Anyway, I wonder if there is something I'm missing to get the time down further?

Cheers

J

[Stephen.Blair]

Most of that time to first pixel is reading the instance nodes from the ass file and initializing those nodes. If you look in an Arnold log (verbosity level Detailed) you should see something like this:

00:26:16 13474MB         | ---------------------------------------------------------------
00:26:16 13474MB         | frame time                  26:16.20    machine utilization (50.09%)
00:26:16 13474MB         |  node init                  26:07.10
00:26:16 13474MB         |   ass parsing (procs)        0:37.34

That was for instances of one object.

I don't think HtoA uses the new Arnold instancer node yet. Eventually it will. That will give some speedup, but millions of nodes will still take some time to initialize.

I'm not sure there's much you can, unless you can reduce the number of nodes somehow eg fewer instances of a bigger point cloud???