PRIMERJAVA UCINKOVITOSTI PAKETA WIEN PODATKI: dr. Igor Vilfan R. Krivec 04/04/09 ------------------------------------------------------------------------------------------------- From igor.vilfan@ijs.si Fri Apr 9 10:44:54 2004 To: rajmund.krivec@ijs.si Subject: primerjalni test ------------------------------------------------------------------------------------------------- Hitrost testnega programa na: (T_mercury/T)*(MHz_mercury/HMz) (relativna hitrost pod predpostavko MIPS (integer) aritmetike) ------------------------------------------------------------------------------------------------- Benderju: 0.88 Compiler options: -FR -mp -w Linker Flags: bbb.o -L../SRC_lib -L/opt/intel/mkl/lib/32 -Vaxlib -static R_LIB (LAPACK+BLAS): -llapack_lapw -lmkl_lapack -lmkl_p4 -lguide -lpthread (brez -llapack_lapw in brez bbb.o mi noce prevesti). ===> TOTAL CPU TIME: 318.1 (INIT = 2.7 + K-POINTS = 315.4) > SUM OF WALL CLOCK TIMES: 329.3 (INIT = 2.8 + K-POINTS = 326.6) Maximum WALL clock time: 330.154829025269 Maximum CPU time: 318.430000000000 Mercury: Compiler options: -Bstatic -fast -O5 -dalign -free Linker Flags: -L../SRC_lib -L../SRC_lib -fast -xlic_lib=sunperf R_LIB (LAPACK+BLAS): -llapack_lapw Benchmark: ===> TOTAL CPU TIME: 1679.7 (INIT = 9.3 + K-POINTS = 1670.4) > SUM OF WALL CLOCK TIMES: 1681.9 (INIT = 9.8 + K-POINTS = 1672.1) Maximum WALL clock time: 1686.113270998001 Maximum CPU time: 1681.77 1.00 Za primerjavo se nekaj vrednosti na drugih racunalnikih (http://www.wien2k.at/reg_user/benchmark/) At present an Itanium2 machine is the fastest machine in this list, followed by a much cheaper Intel P4 (3.2 GHz) with ifc compiler and mkl library (probably with the "goto"-library it would be even faster) and an (expensive) IBM Power 4+ . Compaq-Alpha 666 1092 sec Dec compiler 0.92 Compaq-Alpha EV68, 1GHz 574 sec -ldxml 1.2 G4 4877 sec /-O Absoft compiler G5(dual CPU64bits 2GHz) 690 sec Absoft/-O 0.49 G5 350 sec /xlf compiler /G5 64bits libraries /-O5 -qhot -arch=g5 Athlon XP3000+ (2.17 GHz) 1444 sec (PGI, PGI compiled BLAS) 0.21 Athlon XP3000+ (2.17 GHz) 705 sec ifc, mkl 0.44 Athlon XP3000+ (2.17 GHz) 541 sec ifc, ifc compiled LAPACK, Athlon 0.57 ATLAS) Athlon XP3000+ (2.17 GHz) 515 sec PGI, PGI compiled LAPACK, Athlon 0.60 ATLAS) P4, 2.5 GHz, dual channel mem. 347 sec ifc7, mkl6 0.78 P4, 2.5 GHz, dual channel mem. 328 sec ifc7, goto-library* 0.82 P4 dual-Xeon, 2.4 GHz 341 sec ifc7, mkl6 0.82 P4 dual-Xeon, 2.4 GHz 304 sec ifc7, goto-library* 0.92 P4, 3.0 GHz, dual channel mem. 288 sec ifc7, mkl6 0.78 P4, 3.2 GHz, dual channel mem. 258 sec ifc7, mkl6 0.58 P4, 3.2 GHz, 400MHz dual ch.mem. 228 sec ifort8.0, mkl6.1 0.92 AMD-Opteron, dual cpu, 1.8 Ghz 696 sec ifc7, mkl 0.53 AMD-Opteron, dual cpu, 1.8 Ghz 430 sec pgf90, ACML-library 0.86 AMD-Opteron, dual cpu, 1.8 Ghz 401 sec ifc7, goto-library* 0.93 AMD-Opteron, 1.6 Ghz 360 sec ifc7, goto_opt32-library* 1.2 AMD-Opteron, dual cpu, 2.0 Ghz 270 sec ifc7, goto_opt32-r0.92-library* 1.3 IBM p630 1.45GHz Power4+ 241 sec xlf 8.1.1,-q64 -O5,ESSL4.1 1.9 Itanium2(1.3GHz,SGI Altix 3700) 298 sec ifc7.1 + mkl6.0 1.7 Itanium2(1.5GHz 6Mb cache, HP) 190 sec HP f90 + mlib 2.4 ------------------------------------------------------------------------------------------------- * libgoto_p4_512-r0.6.so blas libraries are available from: http://www.cs.utexas.edu/users/kgoto/signup_first.html -- Igor Vilfan Phone : +386 1 477 36 70 J. Stefan Institute Fax : +386 1 477 37 16 P.O.Box 3000 E-mail: igor.vilfan@ijs.si SI-1001 Ljubljana, Slovenia URL : www-f1.ijs.si/~vilfan