	apeNEXT: the future	Schifano S. Fabio - INFN

Let consider the following data about PCs and custom processors

PC APE
Performance 2Gflops 1,6Gflops
Efficiency 30% 50%
Remote BW 100 Mb/s (Myrinet) 400 (1200) Mb/s
Power diss 30W / node 4W / node
Power/Tflop 45KW 5KW
Vol / Tflops 10 m^3 1 m^3

	PC	APE
Performance	2Gflops	1,6Gflops
Efficiency	30%	50%
Remote BW	100 Mb/s (Myrinet)	400 (1200) Mb/s
Power diss	30W / node	4W / node
Power/Tflop	45KW	5KW
Vol / Tflops	10 m^3	1 m^3

It's our opinion that:

system with O(100) nodes can be realized using PCs
system with O(1000) nodes needs custon processors
we have a real experience about assembling thousand of nodes
it's more fun to design a dedicated QCD processor instead of design the most powerful cooling system of the world.

Both solution need a fast and low latency interconnection structure.
The target of out project is useful for both alternatives.

apeNEXT: the future