Path: utzoo!utgpu!attcan!uunet!lll-winken!lll-tis!helios.ee.lbl.gov!nosc!ucsd!ucsdhub!hp-sdd!hplabs!hpl-opus!hpccc!hp-sde!hpfcdc!hpfcmr!hpfcse!hpuecoa!bgphp1!rclark From: rclark@bgphp1.UUCP (Roger N. Clark) Newsgroups: comp.sys.hp Subject: Re: HP 9000/500 vs. Vectra with 386 CPU and 387 co-processor? Message-ID: <830016@bgphp1.UUCP> Date: 28 Jul 88 14:20:12 GMT References: <592@rocksanne.UUCP> Organization: U.S. Geological Survey, Branch of Geophysics, Denver Lines: 139 > Does anyone have any experience and/or comments on comparing > a HP 9000 Series 500 system with a Intel 386-based machine (perhaps like > the HP Vectra line)? > .... The assertions that the HP9000 series 500 is slower than the 300 and 800 series is not exactly true. Neither are the 386 machines faster in certain cases. Of course you can always find a benchmark that will show one machines strength. HP doesn't seem to know how good of a machine they have in the 500. The 500 excells in a number crunching multiuser (or multitasking) environment. The following benchmarks demonstrate some points. For example, the Byte sieve benchmark shows the 350 to be very fast: SYSTEM sieve.c rsieve.c fsieve.f cbench.c fbench.f dbench.f u+s u+s u+s u+s u+s u+s =============================================================================== HP9000 s835 0.11 0 0.11 0 0.17 0 0.05 0 0.04 0 0.05 0 HP9000 s825 0.14 0 0.14 0 0.26 0 0.17 0 0.10 0 0.17 0 ------------------------------------------------------------------------------- HP9000 s350 FPA 0.74 0 0.46 0 0.64 0 0.33 0 0.15 0 0.17 0 HP9000 s350 68881 0.74 0 0.46 0 0.64 0 0.33 0 0.15 0 0.21 0 ------------------------------------------------------------------------------- HP9000 s500 FPA 2.7 0.0 2.7 0.0 3.1 0.0 0.79 0 0.65 0 0.90 0 ------------------------------------------------------------------------------- Compaq 386/20 w80387 0.39 Compaq Portable III 12 MHz MS C 5.0 w/ 80287 1.1 IBM-AT MS F77 80287 4.7 4.8 4.9 ------------------------------------------------------------------------------- Cray-1 Fortran 0.11 ------------------------------------------------------------------------------- The sieve.c is the Byte benchmark, rsieve.c uses registers, fsieve.f is in fortran, and the cbench and fbench are my own invention that just does a bunch of +, -, *, /, sin, cos, sqrt, log, etc. On the face of it, it does look like the series 500 is slow. But the above benchmarks are separated into to categories: 1) array indexing (the sieve) and 2) compute power. I put the Cray in there to illustrate how benchmarks can be used to give a false picture: no one ine their right mind would say a Compaq 382/20 is less than 4 times slower than a Cray 1! Maybe that's indicated by the sieve, but just try doing some REAL problems! When the 2 are combined (array indexing and floating point computation) the picture changes: A Multitasking, CPU intensive Benchmark ("weird box filter" in Fortran) ----------------------------------------------------------------------- Number of Tasks System 1 2 3 4 5 7 10 12 ----------------------------------------------------------------------- HP9000/835 HP-UX:2.0 0.5 1.0 1.5 2.0 2.4 3.4 4.9 5.9 HP9000/825 HP-UX:1.2 1.9 3.8 5.7 7.6 9.5 13.3 19.1 22.8 HP9000/500 3 CPUs 5.9 6.0 6.3 8.4 10.5 14.7 21.5 27.8 HP9000/840 HP-UX:1.2 2.1 3.9 5.8 7.8 10.9 16.2 26.0 32.3 HP9000/350 FPA UX6.0 2.7 5.2 7.9 10.5 13.2 18.4 26.4 31.6 HP9000/500 2 CPUs 5.9 6.1 9.1 14.1 16.9 25.4 35.8 43.0 HP9000/320 no FPA 11.4 22.7 33.8 45.2 56.4 80.2 119.4 143 Compaq Portable III/40: 12 MHz True Basic 122. " with 80287 119. with MS C 5.0 7 IBM-PC AT +80287 7.2 - - - - - - - IBM-PC no 8087 1111. - - - - - - - ----------------------------------------------------------------------- The "weird box filter" does +, -, /, * on a 200 by 200 array and the result goes into a 200 by 200 array. In a compute intensive multitasking environment, the series 500 looks very good (at least for this benchmark). If you have one task to run, and it fits in the memory of a 386 machine, then it is probably the cheapest way to go. If you need multitasking then the 500 is impressive! Of course, if you are buying new machines, a 350 is probably cheaper. BUT WAIT... What if you have to do complex floating point? Here is a complex floating point benchmark: Complex BENCHMARKS SYSTEM Complex.f u s real %cpu =============================================================================== Cray XMP/48 3.9 ------------------------------------------------------------------------------- HP 9000/825 94.4 0.3 96 99% ------------------------------------------------------------------------------- HP 9040 Unix FPA 271.1 0.3 279 97% <--- 1 cpu only ------------------------------------------------------------------------------- HP 9000/350 (A) 423.1 0.2 424 100% HP 9000/350 (B) 442.8 0.1 443 100% ------------------------------------------------------------------------------- Now we see a single cpu 500 much faster than the 350, and in a multitasking compute heavy environment with 3 cpus, the series 500 would be similar to an 825 (e.g. to run 3 jobs, the 825 would take about 283 seconds, the 500 about 300 seconds)! Note that the 350 (A) is with a 68881 and (B) is with the FPA; the FPA is slower than using the 68881 alone!!!!! Both ran HPUX6.0. My conclusion (as I've posted to the net before): The HP9000 series 500 is a DAMN GOOD MACHINE. Its fast, and never crashes. This is borne out by many users doing varied compute, edit, compile, debug, nroff, etc. jobs. The 500 holds up very well under load. Try that on a compaq 386 running OS/2 or Unix: it will roll over an die long before the 500 does. If you have an application that many people need to run, do a careful cost-benefit analysis before you buy everyone PC's; I have often seen that it is cheaper to buy multiuser workstations/minis and still get the same response, plus you have MUCH easier time sharing data bases and peripherals. SUPPORT THE 500! (If you are interested in the actual benchmarks, I can post them or email them; the box filter benchmark was posted a few months ago). Finally: use your exact application when benchmarking a machine before deciding what to buy if speed is that important! If you have a 500, it might be worth keeping a while unless you need things like NFS (would somebody please put NFS on the 500!). I'm still pissed HP pulled the rug out from those who bought the 500. I can understand why, but it doesn't make it any easier these days when budgets are so tight. I just wish HP gave more for the 500 in the upgrade program to the 800 series. Roger N. Clark (stuck with a 500; not unhappy but I want NFS) bgphp1!rclark