It's only got 32MB of DDR1 RAM. That isn't enough to run cgminer, is it?
It runs fine, but I have not tested with multiple devices. I don't think it'll be a problem, I'll hook up some FPGAs and let you know.
Once you clear all the shit out of linux, you will be surprised at how light it can be (<1mb).
Main issue is going to be the interfaces bottlenecking then hammering the CPU interrupt threads, potentially with ASICS you should be seeing a full nonce cycle time in about 30-35ms, so there will be an enforced work requirement at the end of that , plus any overhead of "found" nonces during that period.
It all depends how these bloody suppliers are going to configure the ASICS.
A. Each ASIC sha256 internal protocol engine is given a full range.
B. Each ASIC sha256 internal protocol engine is given a partial range
C. buffering of work within the engine?
D. buffering of final results?
I.E with a full scan range of 30-35ms, it makes sense to return a SINGLE buffer of all valid results for that nonce at the end of the scan, that way work scheduling becomes very easy and less likely to collide on the communication bus.
Once you allow individual reporting of nonce hits, you basically convert your scheduling of jobs to a random function, as returning nonces collide with potential job allocations on the communication busses.
HC