CP2K PERFORMANCE ON ARCHER Fiona Reid Overview •  ARCHER / HECToR architecture comparison •  Benchmarks and results •  Performance hints and tips •  Conclusions Architecture comparison Feature HECToR ARCHER Processors AMD Interlagos 2.3GHz Intel Ivy Bridge 2.7GHz Cores per node 32 (4× 8-core NUMA) 24 (2× 12-core NUMA) Memory per node 32 GB (1 GB/core) 64GB (2.66 GB/core) 128GB (5.33 GB/core) Nodes 2816 (90,112 cores) 3008 (72,192 cores) Interconnect Cray Gemini Cray Aries Topology 3D Torus Dragonfly Post-processing Nodes (None) 2 Nodes: 48 core SandyBridge 1TB Memory Benchmarks •  The performance of the code has been evaluated using five benchmarks which are indicative of the different types of calculations that can be run with CP2K. •  As both HECToR and ARCHER charge per node, our tests utilise full nodes. •  The PSMP (i.e. MPI/OpenMP) version of CP2K has been used with appropriate combinations of MPI processes and threads tested. H2O-64 •  Short molecular dynamics simulation (10 steps) in NVE ensemble at 300K •  64 water molecules (192 atoms, 512 electrons) in 12.4 Å3 cell •  Quickstep DFT, LDA functional, TZV2P basis set and 280 Ry cut-off Performance comparison for the H2O-64 benchmark Performance comparison of the H2O-64 benchmark 1000 ARCHER HECToR ) s MPI d n )o s dc ne cos MPI e( 100 MPI s (e e m m Tinti 1.87 MPI 2TH MPI u 2TH 4TH r MPI K 2TH 2 1.84 P MPI C MPI 6TH 1.84 6TH MPI MPI 1.69 2.03 2.73 2.03 2.56 10 1 10 100 Number of nodes used Number of nodes used kAU used by the H2O-64 benchmark kAUs used by H2O-64 benchmark 1 0.1 0.94 d de es usu 1.27 sU U’ A A 1.01 kk 1.26 0.01 1.52 1.40 1.40 1.37 ARCHER HECToR 0.001 1 10 100 Number of nodes used Number of nodes used Fayalite-FIST •  Short molecular dynamics simulation (1000 steps) in NPT ensemble at 300K •  28000 atoms (103 supercell, 28 atoms per unit cell) •  Fe SiO (Iron silicate a.k.a. fayalite) 2 4 •  Classical potential (Morse with hard-core repulsive term, cutoff 5.5 Å), plus long-range electrostatics with SPME summation (Smoothed Particle Mesh Ewald) Performance comparison for the Fayalite-FIST benchmark Performance comparison of the fayalite-FIST benchmark 1000 ARCHER HECToR 2TH 2TH 2TH 4TH 2TH 4TH ) 4TH 4TH s 4TH d n )o s dc ne MPI o cs e( s (e 4TH e m m Tinti 1.97 MPI u 6TH r 6TH 6TH K 1.91 6TH 6TH 6TH 2 P 2.02 C 2.06 2.16 2.02 2.09 2.23 2.04 100 1 10 100 Number of nodes used Number of nodes used kAU used by the Fayalite-FIST benchmark kAUs used by fayalite-FIST benchmark 10 1 1.26 d de es su 1.15 u sU 1.23 U’ A A kk 1.28 1.19 0.1 1.25 1.28 1.34 1.31 ARCHER HECToR 0.01 1 10 100 Number of nodes used Number of nodes used