Next: Analysis Up: Castep Performance on HECToR Previous: Node Usage Contents

Baseline

We decided to choose the Pathscale 3.0 binary, compiled with the recip_malloc_inline flags (see section 3.2.3) and linked against Cray's Libsci 10.2.1^3.2 and FFTW3 for our baseline, as this seemed to offer the best performance with the 4.2 Castep codebase.

**Figure 3.6:** Execution time for the 33 atom TiN benchmark. This calculation is performed at 8 k-points.
$\includegraphics[width=0.45\textwidth]{TiN.eps}$ $\includegraphics[width=0.45\textwidth]{TiN_log_time.eps}$ $\includegraphics[width=0.45\textwidth]{TiN_efficiency.eps}$

**Figure 3.7:** Scaling of execution time with cores for the 270 atom Al2O3 3x3 benchmark. This calculation is performed at 2 k-points.
[Execution time] $\includegraphics[width=0.45\textwidth]{Al2O3_time.eps}$ [Efficiency with respect to 16 cores] $\includegraphics[width=0.45\textwidth]{Al2O3_efficiency.eps}$

**Figure 3.8:** Breakdown of CPU time for 256 (3.8(a)) and 512 (3.8(b)) cores using 2 ppn, for Castep Al2O3 3x3 benchmark
[CPU time for Castep on 256 cores] $\includegraphics[width=14.0cm]{256_orig_craypat.eps}$ [CPU time for Castep on 512 cores] $\includegraphics[width=14.0cm]{512_orig_craypat.eps}$

**Figure 3.9:** The CPU time spent in the two dominant user-level subroutines and their children, for a 512-core (2 ppn) Castep calculation of the Al2O3 3x3 benchmark
[CPU time spent applying the Hamiltonian in Castep] $\includegraphics[width=14.0cm]{512_orig_craypat_applyH.eps}$ [CPU time spent preconditioning the search direction in Castep] $\includegraphics[width=14.0cm]{512_orig_craypat_nlpot.eps}$

Next: Analysis Up: Castep Performance on HECToR Previous: Node Usage Contents

Sarfraz A Nadeem 2008-09-01