In this paper, a new hydrodynamics code to simulate astrophysical flows on Intel Xeon Phi KNL and Intel Xeon Scalable processors is presented. A new vector high-order accuracy numerical method implemented in the form of a program code for massively parallel architectures is proposed in detail. The details of code verification are described. A detailed description is given and a parallel implementation of the code is made. The performance of 173 gigaflops and a 48-fold speedup are obtained on a single Intel Xeon Phi processor. A 97% weak scalability is reached with 16 Intel Xeon Phi 7290 processors. We also provide a short description of vector instructions that were used for code optimization. Through this optimization, we achieved 200% performance growth in comparison with the nonvectorized code.
25 May 2018 in Lobachevskii Journal of Mathematics
In this paper, a new approach to vectorization of algorithms of computational fluid dynamics to simulate the dynamics of astrophysical objects is presented. A co-design of a computational model, from...
1 November 2018 in Lobachevskii Journal of Mathematics
In this paper, a new hydrodynamics code called gooPhi to simulate astrophysical flows on modern Intel Xeon Phi processors with KNL architecture is presented. A new vector numerical method implemented...
A new parallel code based on models of special relativistic hydrodynamics is presented for describing interacting flows. A new highly accurate numerical method is considered and verified. A parallel implementation of the method by means of Coarray Fortran technology and its efficiency are described in detail. The code scalability is 92% on a cluster with Intel Xeon 6248R NKS-1P with 192 Coarray Fortran images. Different interacting relativistic flows are considered as astrophysical applications.
Intel Optane technology is a cost-effective solution to create large non-volatile memory pools, which is attractive to many scientific applications. Last year Intel® OptaneTM DC Persistent Memory in DIMM (PMM) form-factor was introduced, which is capable of being used as main memory device. This technology promises faster memory access comparing to Intel® OptaneTM DC P4800X SSD with Intel® Memory Drive Technology devices available previously. In this paper, we present new benchmark data for the Intel Optane DC Persistent Memory. We studied the performance of scientific application from domains of quantum chemistry and computational astrophysics. To put performance of the Intel Optane DC PMM in comparison, we used two memory configurations: DDR4 only system and Optane DC SSD with Intel Memory Drive Technology (IMDT) that is another option for memory extension. We see that PMM is superior to IMDT in almost all our benchmarks, to no surprise. However, PMM was 20-30% more performant in quantum chemistry and only marginally better for astrophysics simulations. We also found, that for practical calculations hybrid setup demonstrates on the same order of magnitude performance as DDR4 system.