apeNEXT News

In February 2004 a first full board of apeNEXT has been put into operation and tested successfully. apeNEXT is the newest generation of APE special purpose high-performance computers developed by INFN, DESY and University of Paris-Sud. A single board with 16 processors operates at a peak speed of about 20 GFlops. In autumn 2004 the APE collaboration plans to commission two prototype machines with 512 processors each at a price of 0.5 Euro/MFlops peak speed. This will be the next milestone on the path towards commissioning large multi-TFlops apeNEXT installations.

APE computers are developed to satisfy the huge amounts of compute power in theoretical particle physics in a cost-efficient way. Since many years these kind of massively-parallel computers have played an important role for simulations of Quantum Chromodynamics (QCD), the theory of strong interactions.

The APE machines are based on the simple but efficient Single Instruction Multiple Data (SIMD) architecture. The custom designed VLIW processors are application optimised. Due to a powerful 3-dimensional communication network the APE architectures are highly scalable.

Here a brief overview on the various generations of APE machines:

• APE1 had a peak performance of 1 GFlops and started physics production in 1987.
• The second generation of APE computers, APE100, with a peak performance up to 100 GFlops came into operation 1994 and is still used.
• Large APEmille installations have been installed at various European sites since 2000 with an overall peak performance of 2 TFlops.
• First multi-Tflops apeNEXT installations are expected to be commissioned 2005.