[niklas]

Has anyone tested this for an OpenFOAM application?

http://www.nvidia.com/object/tesla_c870.html

Niklas

[madad2005]

Is there not a problem with that considering it only has 1.5GB of memory? I generally work by the rule of 1-1.5GB of RAM per core for external aero problems. Maybe it'd be useful for VOF or engine calculations?

[billy]

Looks interesting!!

[gschaider]

Well Adriano: I assume that you're using double precision. As this thing only uses single precision it might be big enough for your stuff (if you can live with the precision)

[madad2005]

I normally work with meshes of around 4-6million cells, so I'm not so sure. I could be wrong though. Is the functionality of the NVidia Tesla not already available with both their 8/9-series cards and even on Ati radeons, which currently used for folding? I suppose it would provide a cheap option for a mini-cluster assuming your grid sizes weren't too large.

[hemph]

I am very interested in Nvidias CUDA framework for particle simulations. Am planning a small project to test the feasibility over the summer. As Bernhard mentions, single precision only for now, but double should be on the way.

Heree is a nice 300000-particle simulation. http://youtube.com/watch?v=Yb-C4CuvlXU&feature=related

Rasmus

[billy]

Hi,

It seems that double precision is already available.

http://www.nvidia.com/object/io_1213744368983.html

[pbohorquez]

Is it true that the only thing to be modified is the
List_FOR_ALL
definition in ListLoopM.H ?

[floooo]

I think we have to wait OpenCL.
http://en.wikipedia.org/wiki/OpenCL
I will be available for all and will work on linux Mac and windows.
It is a opensource project unlike to Cuda and directx11.

[sorcerer]

Originally Posted by floooo

I think we have to wait OpenCL.
http://en.wikipedia.org/wiki/OpenCL
I will be available for all and will work on linux Mac and windows.
It is a opensource project unlike to Cuda and directx11.

I think you are a bit mistaken. the necessary SDKs and drivers will still be closed source, only the standard is open, and there seem to be issues with the interface between the defined layers [see NVIDIA Forums > CUDA GPU Computing > OpenCL > True Heterogeneous computing (topic 98071)].