CFD algorithm design with reconfigurable computing
I wanted to take up this opportunity to briefly point your attention to an academic and scientific driven course with direct implication to our industries: Computational fluid dynamics (CFD) algorithm design with reconfigurable computing course. You can see the official communication at http://www.accelogic.com/training.html.
As I am sure you have read recently, Reconfigurable computing algorithms implemented in FPGAs are 10x to 1,000x faster than traditional parallel algorithms. Many high-performance computing applications are turning to FPGAs and reconfigurable computing to squeeze performance for their biggest computational challenges.
Although the algorithm implementation cycle in reconfigurable computing is similar in principle to that of a traditional software solution (code program, debug, compile, test), mastering the art of reconfigurable computing requires a paradigm shift in the thought process of the algorithm developer. Under this new paradigm, you free yourself from the limitations imposed by an instruction-based machine architecture (known in the literature as the "von Neumann or software bottleneck"), and open the door to a liberated world, full of opportunities for the discovery of new and groundbreaking algorithms, in which you can manipulate the machine architecture to best solve your problem.
The aim of this course is to disseminate this cutting edge technology among the scientific, research and industry professionals that can benefit directly from it.
If you think you might be interested, please let me know how I can help you finalize your registration process. However, if you are not planning to attend, I would appreciate your feedback as to what can be done to make your participation possible for this session, or future ones of similar content.
I sincerely appreciate your time and responses. Should you have any questions whatsoever, please do not hesitate in contacting me.
Beatriz Navarro Accelogic Phone: (954)249-4761 Email: email@example.com
Re: CFD algorithm design with reconfigurable compu
FPGA's have been around for a while, but I'd say the main reason they haven't made the due dent in the scientific community is the requirement to learn VHDL, which, to put it in perspective, is similar if not even more difficult than assembly language. Imagine writing a huge code in such low level language.
Of course, there are serious projects for developing high level languages but till then it's going to be tough for the average scientist to shift gear.
I've talked with Cray people on problems of interest to us, and the claim of 10x to 1000x is exaggerated EXCEPT for a focused number of specific problems
|All times are GMT -4. The time now is 20:31.|