dgh%dgh@Sun.COM (David Hough) (08/04/87)
I've had several requests for benchmark information about the Sun-4/260. Here are some results for a Sun-4/260 with a 60 ns clock, 16 Meg of memory, and running the Beta-2 version of the Sys4 software release, which has been released to our Beta sites. For comparison, I've listed comparable results for a Sun-3/260 with a 40 ns clock, a 60 ns FPA, 16 Meg of memory, and SunOS 3.2 software. In all cases except one optimization is at the highest available level. The exception is Spice-3 for the Sun-4, for which some modules won't compile at the highest level of optimiza- tion; they're compiled at a lower level. The benchmarks except Doduc's are discussed in the 3.2 version of the Floating-Point Programmer's Guide for the Sun Workstation. The Beta-2 software is preliminary. The final software may well be faster, except for the Linpack 100x100 rolled benchmark for which the generated code already seems to be about optimal. LINEAR BENCHMARKS (Fortran) Single Precision Double Precision Sun-3 Sun-4 Sun-3 Sun-4 Linpack benchmarks: KFLOPS 100x100 rolled BLAs 860 1600 480 1080 100x100 unrolled 4x BLAs 770 1250 440 860 1000x1000 any method 900 1650 600 1100 NONLINEAR BENCHMARKS (Fortran) Benchmark of Nhuan Doduc, real elapsed execution time in seconds 890 540 Spice 2G6, real elapsed execution time in seconds mosamp2 input deck 31 17 NONLINEAR BENCHMARK (C) Spice 3A7, real elapsed execution time in seconds mosamp2 input deck 42 23 David Hough ARPA: dhough@sun.com UUCP: {ucbvax,decvax,allegra,decwrl,cbosgd,ihnp4,seismo}!sun!dhough