Table 5. Performance Correlation between the SpecFP2000 Benchmark and GAUSSIAN 98 Results.
|
System Name |
Alpha500 |
P690 |
E7505(CL2) |
E7505(CL2) |
E7501-S |
E7501-T |
zx6000 |
zx6000 |
K8-32 |
K8-32 |
||||||
|
FORTRAN Compiler |
f90 |
xlf 8.0 |
ifc 7.1 |
PGI 5.0-2 |
ifc 7.1 |
ifc 7.1 |
efc 7.1 |
efc 7.1 |
ifc 7.1 |
ifc 7.1 |
||||||
|
architectural optimizations |
-tune ev6 -arch ev6 |
-power4 |
-tpp7 -axW |
-tp p7 |
-tpp7 -axW |
-tpp7 -axW |
-tpp2 |
-tpp2 |
-tpp7 -axW |
-tpp7 -axW |
||||||
|
numerical library and version |
DXML |
ESSL |
MKL 6.0 |
ATLAS 3.4.1 |
MKL 6.0 |
MKL 6.0 |
BLAS |
GOTO 0.7p |
ATLAS 3.4.7 |
MKL 6.0 |
||||||
|
sum of CPU time a |
2313 |
830 |
773 |
1118 |
912 |
857 |
820 |
791 |
728 |
733 |
||||||
|
inverse ratio of CPU time sum b |
0.35 |
0.99 |
1.06 |
0.73 |
0.90 |
0.96 |
1.00 |
1.04 |
1.13 |
1.12 |
||||||
|
geometric mean ratio b |
0.36 |
1.04 |
1.08 |
0.76 |
0.92 |
0.96 |
1.00 |
1.04 |
1.15 |
1.12 |
||||||
|
published SpecFP2000 base |
383 |
1202 |
934 |
1139 |
1093 |
|||||||||||
|
measured SpecFP2000 base c |
-e |
1044 |
941 |
1157 |
1007 |
|||||||||||
|
measured SpecFP2000 base ratio d |
0.33 |
0.90 |
0.81 |
1.00 |
0.97 |
|||||||||||
|
published SpecFP2000 peak |
422 |
1266 |
944 |
1139 |
1168 |
|||||||||||
|
measured SpecFP2000 peak b |
-e |
1083 |
947 |
1157 |
1043 |
|||||||||||
|
measured SpecFP2000 peak ratio d |
0.36 |
0.94 |
0.82 |
1.00 |
1.05 |
a The summation over CPU time includes all the test-files listed except the first part of test439.
b Inverse CPU time ratio computed by (Tref / T)
and geometric mean ratio calculated by 
, where the reference system
is zx6000/BLAS.
c SpecFP2000 benchmark done by the authors. The marks differ from the published values due to the variations of compilers, operating systems and environmental settings.
d All SpecFP ratios are relative to the measured values in the zx6000 system (base=peak=1157).
e Measurements of SpecFP2000 cannot be performed due to absence of KAP C and Fortran compilers.