Re: Pentium bug and Oracle

In article <3bnuls$olh_at_crcnis3.unl.edu> nolan_at_helios.unl.edu (Michael Nolan) writes:
>From: nolan_at_helios.unl.edu (Michael Nolan)
>Subject: Pentium bug and Oracle
>Date: 2 Dec 1994 20:05:15 GMT
 

>For folks running Oracle on a Pentium system (such as the NCR/AT&T 3000
>series), is there any reason to be concerned about Intel's Pentium FPDIV
>bug?
>---
>Michael Nolan, Sysop for the DBMS RoundTable on GEnie
>nolan_at_notes.tssi.com, dbms_at_genie.geis.com
>(posted from nolan_at_helios.unl.edu)

Well, I read about the bug and immediately went over to my oracle server, connected and typed:
select (4195835/3145727)*3145727 - 4195835 z from dual and got the correct answer: 0

Am I worried about Oracle messing up? No. It seams that Oracle 7 SVR4 for Unixware seems not to use the math hardware capabilities on our Pentium 90. I guess this is good ;)

Am I worried about all my other applications on my server and on my desktop machines? You bet.

Here's what I did...
I called Intel at 1-800-628-8686 and said I was working on a complex financial database server and had a P90 with the FDIV bug. After a few call backs and some questions from an Intel engineer or two I was approved for a replacement P90 in my server, but not for any of the other machines in our department. (I didn't tell them that Oracle doesn't use the Floating point hardware)

To summarize the current state of the Floating Point bug in Pentium chips: 1) Bug free chips are NOT shipping to any vendor at this time (except in sample quantities).

2) Intel will not replace your pentium unless they are convinced you will be hurt by the bug. (I suggest you do what ever it takes to convince them.)

3) Contrary to what Intel says, many credible people believe that you will eventually run into the bug on your Pentium, and it will cause you some sort of problem. (Just the risk of hitting the bug is enough to make me worry)

If you would like to test for the bug on your pentium you can run any number of applications like MS Word, Excel, Access, Windows Calculator, the XCalculator that ships with Unixware, or any program that is compiled to take advantage of the FPU and use double precision on a 486 or Pentium. (Don't write an app and compile it with the equation above, if you do, your compiler's optimizer may pre-calculate the correct answer for you) Get to a point where you can type an equation and type the one mentioned above. If you get the answer of 256, you have a buggy Pentium. The answer is 0 on all the other machines in my office:)

For lots more info on this bug try comp.sys.intel.

P.S. Thanks Oracle for using your own dedicated math libraries. Andy
Andrew McAllister -- mcallister_at_neptune.grad.missouri.edu Office of Research, University of Missouri-Columbia The views above are not those of my employer. Received on Sun Dec 04 1994 - 18:17:11 CET