Compilers and CPU benchmarks (see new results)

Summary

• For benchmark A: AMD is better than P4 if the clock frequency is adjusted (AMD=1.6GHz;P4=2.4GHz)
• For benchmark B: P4 is the best platform, even if adjusted for clock frequency
• For benchmark C: AMD is the best platform, regardless of the clock frequencies used in these benchmarks
• For benchmark D: AMD is better than P4 if the clock frequency is adjusted
• On AMD gcc is 5% faster for A, 5% slower for B , 4% slower for C and 8% slower for D benchmarks
• On P4 gcc is 8% slower for A, 2% faster for B, 6% slower for C and 14% slower for D benchmarks

Data

First number is timing in seconds (lower number is better)
Second number is factor relative to the best number for each of A, B, C, D, regardless of the platform
OS is GNU/Linux, Gentoo distribution (www.gentoo.org)
gcc-3.2.1, libgc-2.3.1, linux-2.4.19-gentoo-r10
CHARMM is c30a1 which includes 20 June 2002(R1) version of GAMESS for quantum calculations
AMD: Asus motherboard, 512MB PC2100, MP1900+ dual box, only one CPU used in this benchmark
P4: Gigabyte motherboard, P4 2.4GHz/533, PC3200/333
Compile options:

• gcc: g77 -malign-double -O3 -march=i686 -fschedule-insns2 -fno-backslash -fugly-complex -fno-globals -Wno-globals
• gcc-p4: g77 -malign-double -O3 -march=pentium4 -mmmx -msse2 -mfpmath=sse -fomit-frame-pointer -fschedule-insns2 -fno-backslash -fugly-complex -fno-globals -Wno-globals
• gcc-amd: g77 -malign-double -O3 -march=athlon-mp -mmmx -m3dnow -mfpmath=sse -fomit-frame-pointer -fschedule-insns2 -fno-backslash -fugly-complex -fno-globals -Wno-globals
• ifc6-p4: ifc -axW -O3 -w
• ifc6-p3: ifc -axK -O3 -w
• ifc7-p4: ifc -xW -O3 -march=pentium4 -mcpu=pentium4 -w
• ifc7-p3: ifc -xK -O3 -march=pentiumiii -mcpu=pentiumpro -w

executable	AMD MP1900+ (1.6GHz,1.00)				P4 2.4GHz,1.50
	A	B	C	D	A	B	C	D
gcc-min	73.0,1.31	202.4,1.77	N/A	N/A	69.4,1.24	163.5,1.43	N/A	N/A
gcc-gms	73.3,1.31	205.4,1.79	630.7,1.07	4558.4,1.50	69.9,1.25	162.6,1.42	740.5,1.26	3914.6,1.29
gcc-ful	93.6,1.67	184.3,1.61	623.8,1.06	4551.9,1.50	115.9,2.08	146.7,1.28	736.4,1.25	3830.8,1.26
gcc-min-p4	N/A	N/A	N/A	N/A	60.5,1.08	129.1,1.13	N/A	N/A
gcc-gms-p4	N/A	N/A	N/A	N/A	60.7,1.09	128.4,1.12	687.3,1.17	3518.5,1.16
gcc-ful-p4	N/A	N/A	N/A	N/A	78.5,1.41	114.6,1.00	669.5,1.14	3475.1,1.14
gcc-ful-amd	94.1,1.69	189.2,1.65	615.1,1.04	4463.4,1.47	115.3,2.07	146.2,1.28	708.1,1.20	3720.8,1.22
ifc6-min-p3	77.0,1.38	205.1,1.79	N/A	N/A	73.6,1.32	182.9,1.60	N/A	N/A
ifc6-gms-p3	77.2,1.38	205.9,1.79	RT/E	RT/E	73.6,1.32	181.1,1.58	RT/E	RT/E
ifc6-ful-p3	142.9,2.56	188.7,1.65	RT/E	RT/E	94.2,1.69	168.7,1.47	RT/E	RT/E
ifc6-min-p4	78.7,1.41	191.6,1.67	N/A	N/A	70.0,1.25	142.9,1.25	N/A	N/A
ifc6-gms-p4	78.8,1.41	191.3,1.67	593.8,1.01	4179.9,1.38	69.9,1.25	143.4,1.25	RT/E	RT/E
ifc6-ful-p4	88.4,1.58	175.2,1.53	588.7,1.00	4115.7,1.35	76.5,1.37	135.5,1.18	RT/E	RT/E
ifc7-min-p3	85.1,1.52	210.0,1.83	N/A	N/A	73.6,1.32	176.9,1.54	N/A	N/A
ifc7-min-p4	N/A	N/A	N/A	N/A	55.8,1.00	128.7,1.12	N/A	N/A
ifc7-gms-p4	N/A	N/A	N/A	N/A	55.9,1.00	129.0,1.13	632.5,1.07	3039.7,1.00
ifc7-ful-p4	N/A	N/A	N/A	N/A	88.0,1.58	116.8,1.02	662.3,1.13	3315.6,1.09

Notes:

• N/A Either the method or instruction set not available
• RT/E Runtime Error
• A Spherical cutoff method molecular dynamics, MbCO+3830 waters, 100 steps
• B Periodic boundary method (Ewald) molecular dynamics, MbCO+4985 waters, 100 steps
• C HF/6-31G quantum mechanical calculation of 36 atom system (nanotube model), RUNTYP=GRADIENT
• D The same as C only B3LYP/6-31G DFT method is used
• gcc Compiled with gcc-3.2.1
• ifc6 Compiled with ifc 6
• ifc7 Compiled with ifc 7
• min Minumum CHARMM compiled: good for benchmarks A, B (pref.dat)
• gms Minumum CHARMM compiled: good for benchmarks A, B, C, D (pref.dat)
• ful Full (all methods) CHARMM compiled (pref.dat)
• gcc-xxx-p4 compiled with g77,optimized for P4
• gcc-xxx-amd compiled with g77,optimized for AMD
• ifc6-xxx-p3 compiled with ifc version 6, optimized for PIII
• ifc6-xxx-p4 compiled with ifc version 6, optimized for P4
• ifc7-xxx-p3 compiled with ifc version 7, optimized for PIII
• ifc7-xxx-p4 compiled with ifc version 7, optimized for P4

---

Milan Hodoscek

Last modified: Wed Jan 15 18:31:54 CET 2003