The computer consists of 32 nodes, each of which contains a GRAPE-6A board ("mini-GRAPE") in a Peripheral Component Interconnect (PCI) slot.[1] The GRAPE boards use pipelines to compute pairwise forces between particles at a speed of 130 Gflops.
The on-board memory of each GRAPE board can hold data for 128,000 particles, and by combining 32 of them in a cluster, a total of four million particles can be integrated, at sustained speeds of 4Tflops.[2]
