I think the only way we are going to grow our developer base fast enough might be train existing users to be developers. While there have been threads on learning paths for C and python, understanding the fundamentals for rendering, rasterizing, simulation, etc. are all probably desirable.
So I’d like to collect learning resources that can take someone from just a basic C/python understanding and bring them to a level where they can make a useful contribution to various subsystems.
Actually wasn’t aware of your initiative till a few minutes ago I started this because I was being asked questions on gsoc and whether individuals had a suitable background to apply to work on certain areas and decided to google up a few resources that I felt would be of value.
These are definitely of value, and thanks for your willingness to merge.
This will hopefully help to keep this in one place and keeps all heads aimed in the same direction.
There’s no need to merge these threads. While there’s overlap, they serve two separate purposes. Useful results from the Sourcerers contest can be posted here and the links that Tom has provided can be of use to competition entrants. Furthermore, beyond the competition, this thread serves as a good general purpose resource on its own.
There’s no reason that these threads cannot peacefully coexist.
There’s no need for re-organization. We have an entire forum devoted to beginning coding and development in Blender (this forum, in fact). That forum has a large variety of topics posted by different people with different purposes. This is a forum website and as such it’s devoted to discussion; often scattered and disjointed (though we do our best to keep discussions focused and on-topic). If you want something more organized than that, then perhaps a different format of website is what you’re looking for.
Now, let’s let this thread stick to its original intent and if you would like to further the discussion, please create a new thread or send me a PM.
@LetterRip
That Kirill Garanzha (Кирилл Гаранжа) papers very attractive, as i understand, he specialized on dynamic GPU memory usage, on demo he get 200M triangle model raytraced semi- interactively on 1.5 GB GPU. Of course whole scene must fit in main RAM, he have 16GB.
you can use multiresolution and predictive loading and a few other tricks to avoid needing the entire scene in main ram. Also most of our models are with mutliple levels of SDS which can be generated as needed. Also we make extensive usage of instancing.
@LetterRip
Unfortunately, cannot. The problem is Monte Carlo solver, that walk semi - uniform random paths in scene, so any caching scheme not efficient, you get cache miss with almost every ray. But that guy use some black magic tricks to solve that.