The Hybrid Processing Units for Science (HPU4Science) academic team aims at tackling difficult scientific problems by developing novel computational methods that make good use of the HPU, Hybrid Processing Units, i.e. CPUs and GPUs, with an emphasis on Electron Paramagnetic Resonance (EPR) applications:
Our hardware tools
The first computational cluster comprises 7 desktop-type computers powered by Intel i7 or Core 2 Quad processors together with GPUs that range from the GTX 280 to Titan Z, through the Tesla C2090, for a theoretical peak performance of 40 TFLOPS, with a practical peak of around 30 TFLOPS for niche computational tasks.
That cluster, entirely home-assembled with consumer-grade parts, is located in the Laboratoire de Chimie de la Matière Condensée de Paris (LCMCP), a mixed CNRS/UPMC chemistry lab located at Chimie ParisTech that specializes both in the development of new materials and in the understanding and characterization of existing ones.
In 2016, we have changed gears and began investing in high end servers which we plug in the air-conditioned server room of Chimie-ParisTech also built that same year. Our first rack server is a DELL Poweredge 630 with 2 Xeon E5-2690, for a total of 32 cores (64 hyperthreads).
Our software tools
In our software environment, we commonly use Python, C, Julia and Matlab, together with specialized tools like PyCUDA, Thrust and CUV for the GPUs. In order to cope with the complexity of our code, our development follows the literate programming methodology as devised by Donald Knuth; we use Noweb coupled to Org-mode within Emacs, with Evil-mode enabled, of course. We commonly practice pair programming, and often switch between the keyboard and the marker.Our dwelling
We are academics and receive funding from CNRS, UPMC, Chimie-ParisTech (PSL), CNES, IRSN and ANR, be it indirectly for the infrastructure that hosts us or directly through specific programs. We haven't yet applied for --nor received-- any private funding.Join Us