check this out
http://users.skynet.be/fquake/
This guy should make some Blender code! This is awesome.
200x grid fluid sim using Navier- Stokes differential equations.
http://users.skynet.be/fquake/
that is absolutely amazing, i dont understand how that is real time.
impressive! thanks for the links…
It could be proted relatively easy to opencl i think, we need seriously GET INVELVED with openCL, in the SAME TIME a new Blender is is birthing it is already getting OLD!
I know it is done by DX11 magic. but I guess that these are the things that are coming our way once OpenCL is implemented.
Direct compute != OpenCL, I reckon there’s some infrastructure that needs putting in place before we all jump on the OpenCL bandwagon too. I very much doubt it’s a simple “lets port things over” thing, it rarely ever is. Let’s have a nice stable and user friendly Blender before we begin bolting bits on it 
As I understand it, OpenCL is much more versatile than DirectCompute. OpenCL compiles to run on both GPUs and CPUs, while DirectCompute is GPU only. OpenCL also has the advantages of being created by The Khronos Group, the same group behind OpenGL, an it isn’t tied to a specific platform.
I have all of the tools that I need to start developing OpenCL code, I just haven’t tried anything yet. It might be interesting to try to port something like this. I may give it a shot…
As far as the usefulness of OpenCL in Blender goes, I think that it will only be useful in certain situations. From what I have seen, OpenCL doesn’t run as quickly as plain C code on the CPU, so porting everything will most likely hurt most of the blender users.
If you are interested in some cool OpenCL examples, check out this website:
He has some pretty cool demos, including a GPU accelerated raytracer.
Thats why i said PORT. I know that blender being opensource dosen’t have the corporation development power, so bet for new technologies is a bit risk, because that technology could not reach the standard status, but i clearly see that openCL will at least be supported for a long time.
Man, programing in openCL involves new paradigm, is not a trivial task porting something that wasn’t designed to be multithread, some time you must recode it.
If we should ever start with opencl it should be now.
Replacing a car around the engine?
Let’s start a collect to hire him for the blender foundation.
I thought the Navier- Stokes equation was the basis of all fluid simulations?
The shader and performance/high-resolution grid is amazing.
The problem right now with OpenCL is that is not finished. I mean, only Nvidia provides OpenCL for windows and linux right now (Don’t know the situation with Apple and their machines). AMD only provides an experimental driver to mess with OpenCL in both Windows and Linux.
DirectCompute drivers are out and ready to use with Windows. So that’s why some students are using it right now. AFAIK DirectCompute is very easy to program, in direct contrast with OpenCL, but this is what my programmers friends tell me, i myself am no programmer at all, also DirectCompute is tied to windows systems only.
Also right now i think the priority should be to finish the 2.6 series of Blender and then, when the drivers and the hardware needed are in hands of the developers any OpenCL work should start. AFAIK right now OpenCL is in “Results may vary” status.
If you visit his website and download the application, he does provide the source code within the download. So if a programmer here were to take a look…