GRAPE-6 Emulation Library

Performance tests of our GRAPE-6 emulation library. We have integrated a Plummer sphere for 0.5 timeunit with a blocked timestep code. We have measured the elapsed time for each model and computed the effective performance of the runs. In all runs, we used 1 GPU board per system.

System configuration & Summary

System	CPU	clock	Ncore	Pmax	note
Aizu 5870	Core i7	2.66GHz	4	335	dn64
Kiev 5870	AthlonX2 4850e	2.5GHz	2	232	dn64
Beijin 4870	Opteron 2376HE	2.3GHz	4	143
Aizu 4850	Core2	3.16GHz	2	159

Results

Supported functions

// standard functions (implemented)
int g6_open(int i);
int g6_close(int i);
int g6_initialize_jp_buffer(int i, int n);
int g6_set_ti(int dev, double t);
int g6_npipes(void);
int g6_set_j_particle(int clusterid, int address, int index, double tj, double dtj, double mass,
                      double a2by18[3], double a1by6[3], double aby2[3], double v[3], double x[3]);
void g6calc_firsthalf(int clusterid, int nj, int ni, int index[], double xi[][3], double vi[][3], 
                      double fold[][3], double jold[][3], double phiold[], double eps2, double h2[]);
int g6calc_lasthalf(int clusterid, int nj, int ni, int index[], double xi[][3], double vi[][3], 
                    double eps2, double h2[], double acc[][3], double jerk[][3], double pot[]);

// standard functions (implemened but stub)
void g6_set_tunit(int newtunit);
void g6_set_xunit(int newxunit);
void g6_set_overflow_flag_test_mode(int aflag, int jflag, int pflag);