FPGA or GPU for monte carlo simulation problem??

[Himanshu_Varshney]

I am working on a project "acceleration for probabilistic load flow analysis using monte carlo simulation". At distribution level bus size will be less (<500). But thousands of load flow analysis needed to be done within critical time. Which will be good solution from performance point of view. FPGA or GPU?

Thanks in advance.

 

[guy.lemieux]

I'm not familiar with the "load flow analysis" problem, or the calculations required to make it work. This will be crucial in answering your question. Generally, some excellent approaches exist on FPGAs for generating random numbers at high data rates. If the load flow calculation is simple enough, eg can be written with just a few fixed-point operators, then you will probably do very well with an FPGA.

More importantly, you are very vague on your precise requirements -- does this need to be embedded into a stand-alone system, or just accelerate a PC as a PCIe card? Do you know how to program both GPUs and FPGAs? What kind of precision do you need in your solution, how big will the problem be, do you need low-latency early results which can be refined to be more precise later, or do you just need a final answer?