Wednesday, September 14, 2011

FPGA acceleration v.s. GPU acceleration

this's from 
comp.arch.fpga group

"I was an FPGA engineer before and I think high performance computing 
based FPGA will lead to a bright future. However through my recently 
projects I found GPU will be more appropriate when there is a 
acceleration need. 

In embedded system, FPGA co-processing plan: 
Intel E6x5C 

and GPU co-processing plan: 
AMD APU (with opencl support) 

and in desktop system, FPGA co-processing plan: 
Full custom design, mostly will be based on PCIe fabric 

and GPU co-processing plan: 
nVidia CUDA (with opencv basically support) 

If I choose FPGA co-processing, the algorithm will be specifically 
optimized and R&D time will be very noticeable. If I choose GPU plan, 
algorithm migration will cost little time(even the original one is 
Matlab code), and the acceleration performance will also be quite 
well. 

As a conclusion, the FPGA acceleration only suits some certain and 
fixed application. However in the real world , many projects and many 
algorithms are very uncertain and arbitrary. With same power 
consumption, GPU plan  may lead better results. For a concrete 
project, I will consider GPU or DSP, and FPGA at last. 

Do everybody agree? "

so comment on it........

No comments:

Post a Comment