Parallel Rendering Thesis Advice

[Anonymous]

With the semester over here drawing to a close and the time for our thesis proposals approaching, I was hoping I could get a little bit of advice regarding one of the projects I had been considering.

You see, during the course of the year I developed a bit of a fondness for graphics programming. Thus when it came to thinking about my thesis I decided to go down the graphics route, with a possible emphasis on the performance implications of parallising real time rendering techniques across multi-core platforms.

The problem arises, however, when I consider my potential platforms – PC, PS3 and CUDA. While the former is fine it’s the latter two that are proving more difficult. Basically, in terms of implementation, the only way that I can see that it would work across all three platforms is either with a ray tracing or voxel based rendering demo. While both approaches would of course work, I’m just unsure as to whether doing such a body of work on these two areas would be of any real benefit to me in the future, given they are so far removed from the current state of rendering in the industry.

As such, I wondering what you folks think of the possibility of focusing on the above area or is there something else in the field you think would be worthy of similar investigation and/or treatment?

[Anonymous]

Without a doubt.

There’s being alot of buzz recently by various games industry gurus ( such John Carmack) about Voxel based rendering becoming the next big thing (again) but this time combined with ray tracing techniques. He spoke about this some time last year. You should be able to find the blog/post etc off gamaustra or google. I think ID 5 tech is supposably going to possibly use techniques like these but little is known about how far he got with it. Remains to be demonstrated how useful his theories are in practice.

As to whether its worthwhile I would say definately. Ray tracers are always good things to show you can write, and parallelized ones, even more so nowadays.

Question though ? I was under the illusion you can’t do RSX related stuff with a retail PS3 with linux installed, you can only access the cell’s PPU and SPE’s? I was under the impression that the graphics was accessible to PS3 development kit environments only? Perhaps I’m wrong?

[Anonymous]

Yep, I’ve being reading up on both Carmack’s and Tim Sweeney’s standpoint on some of these technologies, as well as their reactions to the recent announcements from Intel and nVidia regarding Larabee and their respective ray tracing engines.

With regards CUDA, what sparked my interest in adding it as a viable platform was the work done on a Voxel Rendering Engine based off Ken Silverman’s Voxlap. More info here. That, combined with Carmack’s comments about porting his previous software-driven voxel work onto CUDA, sounded really interesting.

As for the PS3, true using the retail kit and Yellowdog you don’t have access to RSX however, most of the projects I’ve seen that used the console for ray tracing have done without. For example, one project used SDL to provide a framebuffer API on top of Linux. Info on that one can be found here. The Blue Steel ray tracer done by the MIT students is another example of this and was linked to in another thread.

On the other hand, we should be getting PS3 devkits over here during the Summer so that restriction may even be gone by the time I get to the implementation stage. Either way though, I’d be able to implement in some form and go into more detail on how it could be expanded upon when access to the RSX was enabled.

Basically then, I think both options or even a combination of them are open to me, it’s just a matter of which one (if any of course) would be more beneficial to me when I finish up here.

[Author]

Posts