5 replies to this topic

[Rayphoenix]

Hello, I am new to DirectX (never touched directX) and currently I am going to start my Bachelor Thesis, which is to render a particle systems with dx11 compute shader (C++) and test the performance advantage of it.

My related prior experience:

- C++ language

- learning the very basic stuffs of OpenGL & Computer Graphics

- creating simple 3D strategy game with C++ & DarkGDK + Lua

- creating 3D FPS game with Unity3D

- creating simple 2D HTML5 Game with javascript

- playing around with game development/game programming (the high level approach)

 

I seldom go low level like directX, now I want to try it, and I really need your information,

do u guys think I can manage to finish it in around 3-4 months ? (for my bachelor thesis)

 

What's the approximated time to learn rendering a moderate particle systems with dx11 compute shader ?   (based on my prior experience stated above)

 

I heard that Compute Shader is on advance category, is it that difficult ?

 

Really appreciate your help

Edited by Rayphoenix, 18 February 2013 - 02:06 AM.

[Tom KQT]

That's really hard to tell without knowing how good in learning you are. I would bet some people could do it in a week (or course not the whole thesis, just to learn and make a working simple compute shader particle system). And many people wouldn't learn it in years.

Also, how flexible/universal should the system be? Is it supposed to have some kind of configuration (via a dialog or some "script" file) for different kinds of particle behaviour and appearance (flames, rain, show, smoke, sparks, missile trails etc)?

And against what are you planning to compare the performance advantage? Against a CPU particle system?

Edited by Tom KQT, 18 February 2013 - 01:52 AM.

[Rayphoenix]

That's really hard to tell without knowing how good in learning you are. I would bet some people could do it in a week (or course not the whole thesis, just to learn and make a working simple compute shader particle system). And many people wouldn't learn it in years.

Also, how flexible/universal should the system be? Is it supposed to have some kind of configuration (via a dialog or some "script" file) for different kinds of particle behaviour and appearance (flames, rain, show, smoke, sparks, missile trails etc)?

And against what are you planning to compare the performance advantage? Against a CPU particle system?

I am not going to use any external script in this case.

Yeah, comparing it with general particle systems which is computed by the CPU.

[Tom KQT]

Hm, really hard to say, but it is doable IMHO. Particle systems are not SO complicated, it's not a whole rendering engine.

I seriously think that you COULD make a first working prototype in like two weeks if you really wanted. And then have time to improve it and make those performance tests.

 

But maybe I'm too optimistic Also it depends on how much time do you have, if you have only evenings after school, it may be harder.

[Rayphoenix]

Hm, really hard to say, but it is doable IMHO. Particle systems are not SO complicated, it's not a whole rendering engine.

I seriously think that you COULD make a first working prototype in like two weeks if you really wanted. And then have time to improve it and make those performance tests.

 

But maybe I'm too optimistic Also it depends on how much time do you have, if you have only evenings after school, it may be harder.

I think I would have around 4-6 hours effective time per day. I am quite hesitant whether it's enough to finish everything in time, but I really want to do this for my thesis (instead of some other potential topics that came up in my mind).

 

Hopefully I can create particle systems that's not too complex utilizing compute shader, but sufficient enough for performance test.

Thx for your opinion

[Jason Z]

It sounds like an interesting topic.  Of course, everyone learns at a different rate, but if you already understand the basics of C++, then the main things you need to get up to speed on are the following:

 

1. Usage of Direct3D: This is a big topic if you need to learn every nook and cranny of the device context.  Since you are looking for performance advantages, I assume you do indeed want to know all about it...

2. Usage of Compute Shader: The mechanics of learning the compute shader are not so difficult.  You can get the concept in one day if you have a good reference.  The trick is trying to find the best possible memory layout and usage and the corresponding threading model to go with it.

3. Making the particle system on the GPU: That is what it is all about

4. Implementing the same thing on the CPU: That is also a whole other problem, with CPU architectures to worry about.

 

Overall, it seems like a pretty big task, but it should be possible for a thesis.  Of course, if you have questions you can always post them here and people are generally very helpful.