FPGA board with best cost/CLB ratio?

For a university project we are looking for FPGA boards which offer the best possible cost/CLB ratio. We are interested in realizing a system with a large number (possibly hundreds) FPGAs running in parallel. Both single FPGA or multiple FPGA boards are an option as long as the price per logic function is "optimum".

Thanks a lot, Christof

Reply to
Christof
Loading thread data ...

Hundreds of FPGAs in parallel is cool. You should consider to design your own boards ;-)

Martin

PS.: That's is not a joke. In that range it makes sense to build your own solution (or let it build).

Reply to
Martin Schoeberl

Except for the time it'd take (and for college students, money), I'd agree with you, Martin. When I first read the posting, I assumed they would just build a proof-of-concept. Or maybe that was me just being hopeful that they aren't about to spend the money it would take to buy

*that* many demo/eval/prototyping boards.

Anyway, they need to find a board that is stuffed with as many Spartan

3 or Cyclone parts as possible. They will cost less if they are in a smallish package, and at least with Xilinx, the top device or two in each family may not be the most cost effective (i.e., for Xilinx, I'll bet either the XC3S2000 in the 676 pkg or XC3S1500 in the 320 or 456 pkg will be the most cost effective). It's mostly dependant on which part is running the highest volume.

But as we started out discussing, the per/device cost is going to disappear due to using pre-made boards. In this case, I'm afraid they're just going to have to get quotes from each place and figure up the cost/logic. There are just too many variables: profit margins, exchange rates, number of FPGA's per board, and exactly which FPGA is on that board. Not to mention finding boards that are easy to tie together (using a backplane? cables?).

Marc

Reply to
Marc Randolph

In addition to the other advice offered so far, it seems to me that unless your hundreds of FPGAs (or boards) are operating in complete or even relative isolation from each other (which is very unlikely), then suitability for interconnection should be a high-priority criteria in your search.

In fact, it should possible be higher even than your stated $/LUT metric. If your architecture is as scaleable as you think it wil be, you can always just buy more boards. What you can't do is redesign the IO on the 100 boards you've already bought...

Regards,

John

Reply to
John Williams

Christof

For Uni's, Xess boards are great as they are cheap well specified and well supported. Also you can get a Diligent board free from Xilinx. Cost per Flip Flop is not usually the critical issue.

Tom

Reply to
DAVID BINNIE

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.