My company is planning on buying something along these lines: http://store.digilentinc.com/netfpga-sume-virtex-7-fpga-development-board/ for some high speed image processing. We are hoping to receive approximately 18Gbit/s of raw data (over two 10Gbit connections) from each of two cameras, perform a bunch of processing, and send a reduced stream (perhaps ~10Gbits) of data over PCIe to other processing elements. This is a big enough technical problem, but before we get there we have a nontechnical one: we have no people on staff with real FPGA experience.
From everything we've read and what people have told us about what 'should' be possible with FPGAs, we are finding no reason why our proposed architecture wouldn't work, but this is much weaker than the confidence we would like to have, where we could know for sure that it could work and getting there is "just engineering".
I am thinking we will need a consultant to weigh in, perhaps two or three to make sure we aren't wagering the whole program on one opinion. Ideally, I would like for some of our people to gain some familiarity to at least understand the constraints to be able to interpret the consultants' answers to some depth, but I am not sure if this is realistic.
Can anyone offer advice on how I might go about answering this question of whether an appropriately sized FPGA PCIe NIC can solve my problem? My current thought is there is no substitute for real experience, and I would do more harm than good if I were to count DSP blocks, logic cells, RAM bandwidth etc. and try to predict whether the current algorithm will fit. But I still need to answer the question one way or another.
I realize this is not a real technical question, but I'd still be interested in your thoughts on how I might move forward.
Question
Jamie
My company is planning on buying something along these lines: http://store.digilentinc.com/netfpga-sume-virtex-7-fpga-development-board/ for some high speed image processing. We are hoping to receive approximately 18Gbit/s of raw data (over two 10Gbit connections) from each of two cameras, perform a bunch of processing, and send a reduced stream (perhaps ~10Gbits) of data over PCIe to other processing elements. This is a big enough technical problem, but before we get there we have a nontechnical one: we have no people on staff with real FPGA experience.
From everything we've read and what people have told us about what 'should' be possible with FPGAs, we are finding no reason why our proposed architecture wouldn't work, but this is much weaker than the confidence we would like to have, where we could know for sure that it could work and getting there is "just engineering".
I am thinking we will need a consultant to weigh in, perhaps two or three to make sure we aren't wagering the whole program on one opinion. Ideally, I would like for some of our people to gain some familiarity to at least understand the constraints to be able to interpret the consultants' answers to some depth, but I am not sure if this is realistic.
Can anyone offer advice on how I might go about answering this question of whether an appropriately sized FPGA PCIe NIC can solve my problem? My current thought is there is no substitute for real experience, and I would do more harm than good if I were to count DSP blocks, logic cells, RAM bandwidth etc. and try to predict whether the current algorithm will fit. But I still need to answer the question one way or another.
I realize this is not a real technical question, but I'd still be interested in your thoughts on how I might move forward.
Thanks,
-Jamie
Link to comment
Share on other sites
5 answers to this question
Recommended Posts
Archived
This topic is now archived and is closed to further replies.