seamans@seaimage.nlm.nih.gov (James R. Seamans) (02/22/91)
I'm not really sure how to begin this thread of discussion, but I'll try anything twice. I'll try to kept it short. Here at the National Library of Medicine, we have a large Sun shop with a diversity of systems. In October of 1990, a requirement for five new Sparc systems was identified and analysis of the missions for these systems was started. Cut through the **** Two Sparc vendors were identified who could supply the needed hardware. Of course one was Sun Microsystems and the other was Solbourne Computer. Solbourne provided us a S4000 system to test our software and generally see the box. Sun had provided a Sparc 2 system which was in the building but we did not know it, since it was in another division. I later took our executable and related files to the local Sun office to be tested. System Architecture: Solbourne S4000 Sparc Panasonic MN10501 25.5 MIPS, 1.7 MFLOPS, 12 SPECmarks 16 MByte of memory 207 MB disk; running X window system. Sun Sparc 2 4/75 Sun Sparc 2 28.5 MIPS, 4.2 MFLOPS, 21 SPECmarks 16 MByte of memory 207 MB disk; running Openwindow system. The software was a locally developed correlation program for image processing. It has a significant amount of floating point operations and was originally developed on a Sun 4/260. The program was compiled without any optimization on the Sun 4/260 and the executable was run on each system without any modifications. Here are the timings for the different systems: Sun 4/260 3 hr 28 min Sun 4/75 1 hr 6 min S4000 1 hr 8 min Question #1: If the Sun 4/75 has significantly better floating point operation than the S4000, how come it only showed a 2 minute better time? When the Systems Engineer from Sun saw the timing, he immediately wanted and got a re-compile using the latest Sun un-bundled C compiler. He utilized several optimizations to re-compile the program. re-running on: Sun 4/470 35 mins (? 22 MIPS system ) Sun 4/75 30 mins Now, these were the timings I was expecting from the Sparc 2. Question #2: Does this mean that the specifications Sun quotes is based on the un-bundled C compiler and gives a false impression of the Sparc 2 system speed? Comparing comparable systems from both: 16 Mb mem, 19"Color, 207M disk ( you know a standard box) Solbourne is $4000 less Now, Jump to the system we want to purchase: Solbourne 19" 24bit color 1280 x 1024 Sun 21" 24bit color 1280 x 1024 16 Mb memory rest of the system about the same At current pricing, Solbourne is $21,500 less than Sun. Question #3 ....: Who is Sun kidding? Is the 2 extra inches worth $21500? Have I missed some hidden attribute of the Sun machine? What is going on at Sun? Are they changing their name to IBM? Caveat: No I have not had a chance to re-run on the Solbourne with an un-bundled compiler but it WILL happen. Jim Seamans, Senior Computer Scientist seamans@seaimage.nlm.nih.gov (130.14.1.73)