Evaluation of Intel Memory Drive Technology Performance for Computational Astrophysics

In this paper, we present a new version of the AstroPhi code for the numerical simulation of relativistic astrophysics. We provide benchmark data for Intel Memory Drive Technology (IMDT), which was used as an extension of DDR4 memory on the computational node. We used this new generation of Software-defined Memory (SDM) based on Intel ScaleMP collaboration and using 3D XPointTM based Intel Solid-State Drives (SSDs) called Optane for numerical simulation of astrophysical problems. Modern astrophysical problems such as a variety of dynamic and general relativistic phenomena (mergers of binary neutron stars and black hole-neutron star binaries or stellar collapse and explosion) require a large amount of memory as well as a large number of computational nodes. IMDT gave the possibility of extending DRAM memory or extending the scratch drive for an operating system defined by the user. To put the performance of IMDT in comparison, we used two systems: DRAM and DRAM+IMDT nodes. The performance was measured as a percentage of used memory and analyzed in detail.