- Board or Cube
- Geometry: 2 x 2 x 2
Communication: onboard
Topology
HPC and Data for Lattice QCD
Topology
Topology
An APEmille machine is a 3D grid of processing nodes with hardware data links between the 3D first neighbouring nodes. The smallest machine is the single Processing Board (PB) whith 8 processing nodes with a 2x2x2 topology, (cube).

Communication: onboard, backplane (z)

Communication: onboard, backplane (y,z)

Communication: onboard, backplane (y,z), external cabling (x)

Communication: onboard, backplane (y,z), external cabling (x)
Global Architecture
In each APEmille installation, CPCI PC is used as I/O server for up to
four Processing Boards. If the geometry of the APEmille machine exceeds
one unit (2x2x8 nodes) these PCs themselves are connected via the
Host-Network
to the control process (APEos) and run only a stripped down
(Slave-) version of the APEos. Since the control process can also manage
the whole disk-I/O of the application program, network bandwith
should be designed sufficiently large. For the installation at Zeuthen
this bandwidth is provided by a high-speed network with ring-topology
(FLink).
