eltonjohn
Joined: 22 Feb 2002
Posts: 1751
Helped: 28
|
 27 Jun 2009 20:51 DOING DSP with NVIDA MASIVELLY PARALLEL graphic processor |
|
|
|
|
I recently got interested in a new line of product by NVIDIA .which are massively parallel floating point processor .This cpus with 240 cores running at 1.3ghz are capable of a TERAFLOP .This is really extraordinnary!.
Such amount of computation can be easly harness with a programming model called CUDA .This is basically a multithread API that can be interfaced to a C/C++ host compiler like Visual C++ or a gnu .I have ordered a board ,but the product is in such huge demmand that i will not get my card until Agost or probably later .
This product can also be interfaced to MATLAB an even to MATEMATICA .
My interest is doing DSP with it .And if this type of solution harness the amount of possible computation .This can perfectly be used to replace FPGAs .
The only programming abstraction model required is threads in a well understood C enviroment. This is much simpler than complex parallel and timing matching requirements of FPGA demands . Also the price is way,way way cheaper that similar FPGA plataform .And please DONT .even think in comparing it to tradictional floating point DSP.
This new type of technology can be a new direction in INTENSIVE NUMERIC COMPUTATION
I leave the link to those products here !
http://www.nvidia.com/object/product_tesla_c1060_us.html
|
|
