Table 8

Table 8. Throughput Correlation between the SpecFP2000rate Benchmark and GAUSSIAN 98 Results.

System	E7505 (CL2)									zx6000									K8-32
Compiler	ifc 7.1						PGI 5.0-2			efc 7.1									ifc 7.1
Arch. opt.^a	(a)						(b)			-tpp2 -O3									(a)
Library	BLAS			MKL 6.0			GOTO 0.6p			intrinsic BLAS			MKL 6.0			GOTO 0.7p			ATLAS 3.4.7
Duplication	Single	1st	2nd	Single	1st	2nd	Single	1st	2nd	Single	1st	2nd	Single	1st	2nd	Single	1st	2nd	Single	1st	2nd
sum of CPU time^b	800	1067	1079	773	1075	1073	1110	1405	-	820	852	852	801	824	823	791	812	813	728	740	736
inv. CPU time ratio^c	1.03	0.77	0.76	1.06	0.76	0.76	0.74	0.58	-	1.00	0.96	0.96	1.02	0.99	1.00	1.04	1.01	1.01	1.13	1.11	1.11
geometric mean ratio ^c	1.05	0.81	0.81	1.08	0.83	0.83	0.79	0.62	-	1.00	0.95	0.96	1.03	0.99	0.99	1.04	1.01	1.01	1.15	1.11	1.12
processors used	1	2								1	2								1	2
published SpecFPrate2000 base	10.8	14.7								13.2	23.9								12.7	23.3
measured SpecFPrate2000 base	11.0	13.0								13.6	24.6								11.7	22.0
measured base SpecFPrate2000 ratio^d	0.81	0.48								1.00	0.90								0.86	0.81
published SpecFPrate2000 peak	11.0	14.9								13.2	23.9								13.5	24.9
measured SpecFPrate2000 peak	11.0	14.1								13.6	24.6								12.1	22.7
measured peak SpecFPrate2000 ratio^d	0.81	0.52								1.00	0.90								0.89	0.83

^a The global options of ifc is set to "-unroll -O2", plus: (a) "-tpp7 -axW -ipo -ipo_obj". The global optimization to pgf77 is set to "-Munroll -O2", plus: (b) "-tp p7 -Mvect=cachesize:524288".

^bThe summation over CPU time includes all the test-files listed except the first part of test439.

^cInverse CPU time ratio computed by (T_ref / T) and geometric mean ratio calculated by , where the reference system is zx6000/BLAS.

^d1 processor: (measured SpecFPrate2000 mark)÷13.6 ; 2 processors: (measured SpecFPrate2000 mark)÷13.6÷2