Navigation and service

High-temperature superconductivity

High-temperature superconductivity

The understanding of superconducting mechanisms in high critical temperature (HTc) materials remains one of the most challenging topics in condensed matter physics. Using Quantum Monte Carlo simulations we aim to study realistic materials in order to determine whether high-temperature superconducting properties can be quantitatively understood and therefore predicted within an ab-initio approach.

The TurboRVB package developed at SISSA (Scuola Internazionale Superiore di Studi Avanzati di Trieste) allows approaching this problem with massively parallel simulations. The TurboRVB is well suited for the DEEP Architecture, since the embarrassingly parallel nature of QMC applications can benefit from the large amount of cores in the Booster.

The code, however, is a prime example for the need of code modernization: Being a pure MPI application, threading is limited to the BLAS library, which, depending on the size of the system simulated, might or might not be enough. Attempts to improve this met limited success due to the structure of the code. At the same time, having a very large number of MPI processes per node increases the burden of collective communications.

In DEEP, OmpSs proved its flexibility, by allowing a single process running in the Cluster to act as the driver for the whole simulation, offloaded to the Booster.

These simulations are done by CINECA, Italy.


"Quantum Monte Carlo is intrinsically parallel, and benefits directly by increasing the number of Monte Carlo walkers. Therefore, we need as much parallelism as possible. The massive amount of threads able to run on the Booster nodes, with their communications, is highly attractive to the QMC community. Traditionally, the parallelism available in BlueGene class machines fits very nicely our needs. Now we believe that the Booster in the DEEP project is a firm first step towards proving an alternative platform with a greater performance per watt and equally good scalability." - Fabio Affinito, CINECA