bauman@shell.com (Evan G. Bauman) (08/07/90)
How fast is a SPARC 330/370 in floating point calculations? This will sure confuse the issue. Just got my July/90 issue of README, the Sun User Group newsletter. Inside is a table entitled "Performance Benchmark Results: Workstations". Looking over the values for the DP Linpack test, I see that the SPARC 1+ is rated at 1.7 MFlops. This isn't too far off the mark. I measured it at about 1.5. The SPARC 330 and 370 are rated at 2.7 MFlops. Now I find this incredible. We have a 330 and a 370 and my Linpack test only yields 1.7-1.8 MFlops. In fact, here's the verbatim results directly from Mr. Dongarra's benchmark. This is from a SPARC 370 workstation, with FORTRAN 1.2, compiled with f77 -O. norm. resid resid machep x(1) x(n) 1.67005097E+00 7.41628980E-14 2.22044605E-16 1.00000000E+00 1.00000000E+00 times are reported for matrices of order 100 dgefa dgesl total mflops unit ratio times for array with leading dimension of 201 3.800E-01 1.000E-02 3.900E-01 1.761E+00 1.136E+00 6.964E+00 4.300E-01 1.000E-02 4.400E-01 1.561E+00 1.282E+00 7.857E+00 3.700E-01 2.000E-02 3.900E-01 1.761E+00 1.136E+00 6.964E+00 3.990E-01 1.200E-02 4.110E-01 1.671E+00 1.197E+00 7.339E+00 times for array with leading dimension of 200 3.800E-01 2.000E-02 4.000E-01 1.717E+00 1.165E+00 7.143E+00 3.800E-01 2.000E-02 4.000E-01 1.717E+00 1.165E+00 7.143E+00 3.900E-01 1.000E-02 4.000E-01 1.717E+00 1.165E+00 7.143E+00 3.860E-01 1.300E-02 3.990E-01 1.721E+00 1.162E+00 7.125E+00 Does anybody know what's going on here? Do we have a defective floating point chip or is Sun just cooking the numbers to make themselves look good against DEC and IBM? I asked our sales rep., but he couldn't come up with an answer. Thanks. Evan G. Bauman Shell Development Company - Westhollow Research Center PO Box 1380; Houston, TX 77251-1380 bauman@shell.com or for the uucp types: {sun,bcm}!shell!bauman
iarrobin@wookie.east.sun.com (Mike Iarrobino - Systems Engineer) (08/10/90)
In article <1990Aug9.022829.7925@rice.edu>, bauman@shell.com (Evan G. Bauman) writes: > >The SPARC 330 and 370 are rated at 2.7 MFlops. Now I find this >incredible. We have a 330 and a 370 and my Linpack test only yields >1.7-1.8 MFlops. In fact, here's the verbatim results directly from Mr. >Dongarra's benchmark. This is from a SPARC 370 workstation, with FORTRAN >1.2, compiled with f77 -O. I believe these performance numbers were obtained using the latest FORTRAN compiler 1.3 at the highest optimization level (-fast -O4). The default for -O is -O2 which is a compromise between optimization and compile time. Michael V. Iarrobino EMail: mike.iarrobino@East.Sun.COM