Hi!
It all comes down to the amount of DCI's you wish to poll and the polling intervals.
We are using the following CPU and RAM for around 800 nodes with, at the moment, about 25 - 30 DCI's per node,
1x Intel(R) Xeon(R) CPU W3520
12GB RAM
And we are not close to maximizing the resources.
Hope this helps!
Edit: We would probably be able to handle the current load with half the memory, but as you never know what the future hold, and as hardware is relative cheap, it is better to oversize, than to not be able to monitor what you need to
Best regards,
Dan