Benchmarking GCC and ifort compilers on new hardware

Compilers and CPU benchmarks

Data

First number is timing in seconds (lower number is better)
Second number is factor relative to the best number (in red) for each of A, B, C, D, JAC regardless of the platform
number in brackets <> is speedup over single CPU timing
OS is GNU/Linux-2.4.2X, various distributions
CHARMM is c31a2, includes 12 DEC 2003 (R2) version of GAMESS for QM calculations
pref.dat was used
Altix (ia64): 16 CPUs
Pentium4 (ia32): P4 3.2GHz, 8 boxes (CPUs), GigE
AMD Opteron (x86_64): 2 X Opteron 244
MDGRAPE-2S: GRAPE (ia32 after MDGRAPE line is the time for no cutoff on the host only)

NOTE: None of the relative performance factors are set yet

1 CPU
machine	compiler	A	B	C	D	JAC	SHC5
x86_64-2.2GHz	gcc-3.4	37.1,1.00	65.1,1.00
ia32-3.2GHz	gcc-3.4	45.8,1.00	89.5,1.00	515.3,1.00	2592.4,1.00	707.6,1.00	804.6,1.00
ia32-3.2GHz	ifort-8.0	40.9,1.00	83.1,1.00	399.7,1.00	2208.0,1.00	672.1,1.00	768.5,1.00
ia64-1.4GHz	gcc-3.4	99.7,1.00	146.3,1.00	1061.0,1.00	7832.5.7,1.00	1406.4,1.00	1372.9,1.00
ia64-1.4GHz	ifort-8.0	78.7,1.00	107.2,1.00	698.3,1.00	2769.3,1.00	619.0,1.00	1120.7,1.00
x86_64-1.8GHz	gcc-3.4	48.0,1.00	81.9,1.00	452.3,1.00	2725.2,1.00	772.3,1.00	779.5,1.00
x86_64-1.8GHz	pathf90-1.3	51.7,1.00	76.4,1.00			702.7,1.00
x86_64-1.8GHz	pgf77-5.1	48.7	85.2	RT/E	RT/E	712.9,1.00	786.9,1.00
x86_64-1.8GHz	ifort-8.0	54.5	97.5	465.6		897.3,1.00	950.5,1.00
Mac-G5-2.0GHz	xlf-8.1		99.9
Mac-G5-2.0GHz	gcc-3.4		114.3
IBM-Pwr4-1GHz	xlf-8.1	100.5,1.00	159.9,1.00			1267.9,1.00	1282.5,1.00
IBM-Pwr4-1GHz	gcc-3.2(64)	150.9,1.00	248.1,1.00
IBM-Pwr4-1GHz	gcc-3.2(32)	164.7,1.00	251.7,1.00
MDGRAPE-2S	ifort-8.0	33.8,1.00		N/A	N/A		2294.9,1.00
ia32-3.2GHz	ifort-8.0	712.1,21.07		N/A	N/A		60148.4,26.21
2 CPUs
x86_64-2.2GHz	gcc-3.4	18.7,1.00<1.98>	33.0,1.00<1.97>
ia32-3.2GHz	gcc-3.4	24.3,1.00<1.88>	52.1,1.00<1.72>	270.8,1.00<1.90>	1286.1,1.00<2.01>	429.5,1.00<1.65>	410.0,1.00<1.96>
ia32-3.2GHz	ifort-8.0	21.8,1.00<1.88>	48.3,1.00<1.72>	207.0,1.00<1.93>	1141.5,1.00<1.93>	407.9,1.00<1.65>	385.0,1.00<2.00>
ia64-1.4GHz	gcc-3.4	50.5,1.00<1.97>	74.6,1.00<1.96>	537.7,1.00<1.97>	4050.1,1.00<1.93>	728.6,1.00<1.93>	699.5,1.00<1.96>
ia64-1.4GHz	ifort-8.0	39.8,1.00<1.98>	53.6,1.00<2.00>	354.4,1.00<1.97>	1458.6,1.00<1.90>	331.9,1.0<1.87>	580.0,1.00<1.93>
x86_64-1.8GHz	gcc-3.4	24.6,1.00<1.95>	44.1,1.00<1.86>	244.0,1.00<1.85>	1376.7,1.00<1.97>
Mac-G5-2.0GHz	gcc-3.4		63.5,1.00<1.80>
IBM-Pwr4-1GHz	xlf-8.1	51.8,1.00<1.94>	83.3,1.00<1.92>			658.0,1.00<1.93>	657.8,1.0<1.95>
MDGRAPE-2S	ifort-8.0	18.5,1.00<1.83>		N/A	N/A		1161.1,1.0<1.98>
ia32-3.2GHz	ifort-8.0	360.3,19.48<1.98>		N/A	N/A		30348.0,26.09<1.98>
4 CPUs
x86_64-2.2GHz	gcc-3.4	9.6,1.00<3.86>	17.4,1.00<3.74>
ia32-3.2GHz	gcc-3.4	14.0,1.00<3.27>	32.1,1.00<2.79>	133.7,1.00<3.85>	656.2,1.00<3.95>	274.7,1.00<2.58>	219.0,1.00<3.67>
ia32-3.2GHz	ifort-8.0	12.6,1.00<3.25>	30.4,1.00<2.73>	106.4,1.00<3.76>	578.0,1.00<3.82>	264.8,1.00<2.54>	200.9,1.00<3.83>
ia64-1.4GHz	gcc-3.4	26.0,1.00<3.83>	38.2,1.00<3.83>	275.4,1.00<3.85>	1997.6,1.00<3.92>	379.4,1.00<3.71>	362.1,1.00<3.79>
ia64-1.4GHz	ifort-8.0	20.3,1.00<3.88>	28.2,1.00<3.80>	182.0,1.00<3.84>	719.2,1.00<3.85>	176.1,1.00<3.52>	295.5,1.00<3.79>
x86_64-2.2GHz	gcc-3.4	18.7,1.00<1.98>	33.0,1.00<1.97>
IBM-Pwr4-1GHz	xlf-8.1	27.3,1.00<3.68>	44.1,1.00<3.63>			362.6,1.00<3.50>	408.9,1.0<3.14>
MDGRAPE-2S	ifort-8.0	11.1,1.00<3.05>		N/A	N/A		593.2,1.00<3.87>
ia32-3.2GHz	ifort-8.0	184.8,16.65<3.85>		N/A	N/A		15077.0,25.18<3.99>
8 CPUs
x86_64-2.2GHz	gcc-3.4	5.6,1.00<6.63>	11.3,1.00<5.76>
ia32-3.2GHz	gcc-3.4	9.0,1.00<5.09>	23.2,1.00<3.86>	71.2,1.00<7.24>	350.4,1.00<7.40>	204.5,1.00<3.46>	125.3,1.00<6.42>
ia32-3.2GHz	ifort-8.0	8.3,1.00<4.92>	22.2,1.00<3.74>	58.5,1.00<6.83>	301.5,1.00<7.32>	198.1,1.00<3.39>	119.8,1.00<6.41>
ia64-1.4GHz	gcc-3.4	13.5,1.00<7.39>	20.8,1.00<7.03>	143.7,1.00<7.38>	1031.9,1.00<7.59>	211.6,1.00<6.65>	188.0,1.00<7.30>
ia64-1.4GHz	ifort-8.0	10.8,1.00<7.29>	16.0,1.00<6.70>	94.9,1.00,<7.36>	369.3,1.00<7.50>	107.7,1.00<5.75>	154.9,1.00<7.23>
MDGRAPE-2S	ifort-8.0	7.7,1.00<4.40>		N/A	N/A		315.3,1.00<7.29>
ia32-3.2GHz	ifort-8.0	93.7,12.17<7.60>		N/A	N/A		7533.0,23.29<7.98>
16 CPUs
ia64-1.4GHz	gcc-3.4	7.7,1.00<12.95>	13.1,1.00<11.17>	78.1,1.00<13.59>	519.8,1.00<15.07>	135.4,1.00<10.39>	105.3,1.00<13.04>
ia64-1.4GHz	ifort-8.0	6.4,1.00<12.30>	10.8,1.00<9.93>	50.8,1.00<13.75>	191.3,1.00<14.46>	85.0,1.00<7.28>	88.4,1.00<12.68>

Notes:

Compile options:

gcc-ia32: g77 -malign-double -O3 -march=pentium4 -mmmx -msse2 -mfpmath=sse -fomit-frame-pointer -fschedule-insns2 -fno-backslash -fugly-complex -fno-globals -Wno-globals
gcc-ia64: g77 -fno-backslash -fugly-complex -fno-globals -Wno-globals -O3 -minline-float-divide-max-throughput
gcc-x86_64: g77 -O3 -msse2 -mmmx -mfpmath=sse -fno-backslash -fugly-complex -fno-globals -Wno-globals
ifort-ia32: ifort -O3 -tpp7 -132 -axW -w95 -cm
ifort-ia64: ifort -O2 -tpp2 -132 -ftz -WB -w95 -cm -i8
pgf77-x86_64: pgf77 -fastsse -tp k8-64
pathf90-x86_64: pathf90 -O2
xlf-Mac-G5: xlf -O5 -qarch=g5
xlf-IBM-pwr4: xlf90_r -O3 -qfixed -qalign=4k -qarch=auto -qtune=auto -qmaxmem=-1 -q64 -qintsize=8 -qposition=appendold
gcc-IBM-pwr4: g77 -O3 -fno-globals -Wno-globals

N/A Either the method or instruction set not available
RT/E Runtime Error
A Spherical cutoff method molecular dynamics, MbCO+3830 waters(14026 atoms), 100 steps
B Periodic boundary method (PMEwald) molecular dynamics, MbCO+4985 waters(17491 atoms), 100 steps
C HF/6-31G quantum mechanical calculation of 36 atom system (nanotube model), RUNTYP=GRADIENT; files can be found here
D The same as C only B3LYP/6-31G DFT method is used; files can be found here
JAC The Joint Amber Charmm benchmark JAC1000(23558 atoms)
SHC5 + water (130711 atoms)

See also the older page and Spatial decomposition benchmarks

Milan Hodoscek

Last modified: Sun Feb 20 09:47:27 CEST 2005