Any thoughts on GPU Programming? Graphics Programming Units!!!

[OceanJeff32]

I just bought GPU GEMS 2 the new book NVidia just released at the Game Developer's Conference.

It seems totally awesome to me, and I've never even heard anything about programming with the GPUs of the video cards, other than from Andre LaMothe...

Anybody have any source code, or helpful advice for learning about this unique approach to graphics programming.

If not, hopefully, I can post some examples here later on,

Jeff C
::)

[Mark_Larson]

I just bought GPU GEMS 2 the new book NVidia just released at the Game Developer's Conference.

It seems totally awesome to me, and I've never even heard anything about programming with the GPUs of the video cards, other than from Andre LaMothe...

Anybody have any source code, or helpful advice for learning about this unique approach to graphics programming.

If not, hopefully, I can post some examples here later on,

Jeff C
::)

There has been a lot done over the past year with GPU programming both for graphics and for non-graphical purposes.  Their is a webpage dedicated to non-graphical purposes.  Such as using it to compute a FFT.  http://www.gpgpu.org/

[James Ladd]

If you google you may find some articles on C compilers that translate code for running on GPU's.
Since most GPU's are faster than your average P4 CPU it makes sense.
Especially for lots of floating point math.

[OceanJeff32]

Cool website, thanks Mark!

Later all,

Jeff C
:8)

[Mark_Larson]

 
There are a few compilers for what striker is talking about.  The big one is called Brook GPU.  All the compilers can be found on www.gpgpu.com under one of the links.

http://graphics.stanford.edu/projects/brookgpu/index.html - brookgpu.  Allows you to run general purpose code on a GPU.  This should allow you to exploit "parallelism" by letting you run some code on the CPU and other code on the GPU.  Or maybe it's just faster running it on the GPU, and so you do a complete GPU implementation.

  What I am looking at right now is doing raytracing partially on a GPU and partially on my CPU.  Break up the task between the two to be able to handle more data per frame.

  The www.gpgpu.com webpage has several projects on it where people did raycasters completely on a GPU.  So I know it's feasible.  So I want to try and split it up to make it even faster.

  Keep in mind that some tasks will never be able to made to run faster on a GPU.  Some things it does well and other things it doesn't.  Some people have been overhyping how cool this is, so people think you can do anything faster on the GPU.  That's not true.  They have some posted algorithms in the Brook GPU forums that are slower on a GPU than on a P4.  Also keep in mind that if you have a Geforce 6 series of the latest ATI series it is significantly faster than previous generations for a lot of this stuff.