Petaflops simulations for the chemistry of Alzheimer

Thanks to the exceptional computational resources placed at our disposal these last two years by GENCI and the PRACE European consortium, our group has been able to demonstrate the applicability of our simulations for systems difficult to reach by the usual methods of the field. The applications studied are related to the chemistry of Alzheimer’s disease, which brings into play interactions that are particularly tricky to reproduce.

It is known that Alzheimer is associated with a degeneration of the brain connected with the appearance of plaques made up of aggregates of molecules (peptides) known as β-amyloids. Understanding why and how these molecules aggregate is one of the major challenges in understanding the most fundamental mechanisms at the origin of the appearance of this major disease. In collaboration with the experimental biochemistry group of Pr Faller (LCC, Toulouse), we have been able to show that the chemical accuracy needed to describe these systems quantitatively could be attained by using the CURIE computer. Fig. 4 illustrates how the random trajectories of the electrons spread out in the case of the simulated chemical systems.

Fig. 4 – Measuring parallel efficiency resulting from the different strategies combined.

During these simulations, we could measure a real sustained performance of about 1 Pflops (960 single and double precision Tflops to be exact). Even if much remains to be done, our first probabilistic simulations on a massively parallel architecture open promising perspectives. Although the necessary computational loads – millions of CPU hours – are still too important to make QMC a routine method, the forthcoming exaflops machines will probably transform these emerging techniques into a simulation tool accessible to the broadest scientific community.