We've got static batching, we've got paging geometry, we've got smart gpu/system ram paging for static batching, what next?

Unfortunately, we found that we're hitting a bottle neck with paging geometry. It was not batching up the pages fast enough. This caused us bad jerks and halts that happens every few seconds due to our high speed traveling.

I decided that it's not acceptable and decided to come up with a compromise idea. Instead of using the geometry batching provided with the paging geom library, we're going to rewrite one that is thread friendly. What happens here is that instead of generating the batches on the fly in the main thread, we're going for a background threaded batching method. This means that if a batch can't be ready within the next frame, it can wait till the frame after before actually being uploaded into gpu.

However, I do hope that I'm right as in it's the batching that is slow and not the uploading. It's a wild guess, but I think we're going into the right direction. Eventually, we'll need this background thread system for other stuffs like AI anyways.
Posted by Lf3T-Hn4D


Post a Comment

Liquid Rock Games and Project Aftershock. All Rights Reserved.