Subj: Some Infos on Intel's Pentium Chip Debacle, known as "Intel's Waterloo". Last Update: 21.DEZ.1994 - 20.40 MEZ/CET - V 01.17 Sorry, the introduction is in German. Aus den aktuellen System-News des HAIRclusters vom 23.NOV.1994: ________________________________________________________________________________ %NEWS-A-PCwelt: >>>>>> W a r n u n g v o r d e m K a u f >>>>>> e i n e s P e n t i u m - P C s. ------------------------------------------------ Wie heute bekannt wurde, sind a l l e von der Firma Intel seit 1993 produzierten Pentium-Chips fehlerhaft. Der Fehler kann zu absolut falschen Rechenergebnissen fuehren. Ob Intel eine Rueckruf-Aktion startet ist noch unklar und bleibt abzuwarten. ------------------------------------------------ Quelle: CNN-International (Business News) - 20.23 Uhr MEZ/CET [20.45h]-23/11/94-khd ________________________________________________________________________________ Aus den aktuellen System-News des HAIRclusters vom 10.DEZ.1994: ________________________________________________________________________________ %NEWS-A-PCwelt: >>>>>> W a r n u n g v o r d e m K a u f >>>>>> e i n e s P e n t i u m - P C s. ------------------------------------------------ Defekte Pentium-Maschinenbefehle sind: o FDIV o FDIVP o FDIVR o FDIVRP o FIDIV o FIDIVR o FPREM o FPREM1 o FPTAN o FPATAN o FYL2X o FYL2XP1 ------------------------------------------------ V o r s i c h t ! Alle auf einem (defekten) Pentium-Computer er- zeugten Programme duerfen nicht auf anderen Platt- formen - auch nicht auf einem Power Macintosh - verwendet werden. Sie m¸ssen auf der Zielplattform n e u compiliert werden. Fazit: Der Pentium-Defekt, ein Arbeitsbeschaffungs- programm der Firma Intel auch fuer die gesamte Software-Industrie, und der Anwender muss die nun faelligen Ugrades "loehnen". --- --------------------------------------------------- Quelle: "comp.sys.intel" (UseNet). [22.45h]-10/12/94-khd ________________________________________________________________________________ Am 20. Dezember 1994 - 14.00 GMT (6.00 PST) erfolgte durch Intel der (Quasi-)Rueckruf aller bisher produ- zierten 6 Millionen Pentium-Chips. Intel tat damit einen ersten Schritt in die richtige Richtung einer etwas kundenfreundlicheren Politik, allerdings vier Wochen zu spaet! Intel wird nun alle defekten Pentium-Chips ersetzen, sobald sie dazu in der Lage sind, und nunmehr keine Fragen mehr stellen. Weiteres siehe in Intels Original-Presseerklaerung INTELS_PRESSREL_2.TXT auf diesem Ftp-Server. Intels kostenlose Pentium-Hotline ist in Deutschland: 0130/ 818 921 Noch ein wichtiger Hinweis: Die zur Zeit noch im Handel erhaeltlichen Pentium-PCs sind a l l e noch mit dem defekten Pentium-Chip ausgestattet. Das wird sich voraussichtlich erst im Maerz 1995 aendern. ________________________________________________________________________________ Im November 1994 - nur 8 Jahre nach Tschernobyl - war es soweit, auch die Computerindustrie hatte ihren ersten Super-GAU, den "groessten anzunehmenden Unfall". Es erwischte die Firma Intel voll. Intel ist die Nr.1 unter den Chip-Herstellern und besitzt einen Weltmarkt- anteil von 80 %. Intel hat sich einen eklatanten Konstruktions-Fehler ihres vielgepriesenen Pentium-Mikroprozessors geleistet. Seit 1993 haben sie etwa 6 Millionen Chips mit einem eingebauten Divisionsfehler verkauft, ohne es zunaechst selbst zu bemerken. Der Skandal war vollends da, als Intel nur die Chips von ausgesuchten Anwendern aufgrund eines Interviews ersetzen wollte. _________________________________________________________ Die weltweite Diskussion wird im Internet zur Zeit in der Newsgruppe "comp.sys.intel" gefuehrt. Wichtige Auszuege und Infos sind am Schluss zu finden. Den vollen Text des Artikels der NEW YORK TIMES vom Donnerstag, den 24. November 1994 zum Intel Pentium Bug findet man nun auch auf diesem Ftp-Server in der Textdatei PBUG_NY_TIMES.TXT. Bitte beachten Sie das Copyright des NY Times Services, NY, USA. Inzwischen hat auch die "Frankfurter Allgemeine Ztg" ueber Intels Probleme mit dem Pentium-Chip berichtet (29.November 1994 - Seite 25). Der Text ist in der Datei PBUG_FAZ.TXT (Copyright beachten!). Weiteres zum Echo in den Deutschen Massenmedien ist in der Datei PBUG_GERMAN_MEDIAS.TXT zu finden. Eine Stellungnahme des Intel-Praesidenten Andy Grove vom 27. November 1994 (19.31 GMT) befindet sich als Textdatei INTELS_PRESIDENT.TXT ebenfalls auf diesem Ftp-Server. Titel: "My Perspective on Pentium". In- zwischen liegt ein 31-seitiges Weissbuch der Intel Corp. vor (leider noch nicht elektronisch), in dem Intel weitere Fehler des Pentium-Chips einraeumen muss. Das Weissbuch ist datiert vom 30.November 1994. Ein Testprogramm von Terje Mathisen, Norwegen steht im Ftp-Server "ftp.grumed.fu-berlin.de" im Verzeichnis "PC" als komprimierte Datei P87TEST.ZIP zur Verfuegung. Auf dem Server liegt nun auch, frisch (8.12.94) vom Intel-Server aus Santa Clara kopiert, Intels eigene FPU Testsoftware MDIAG.EXE. Diese 'failed' auch bei einem defekten Pentium. Warum bleibt aber unklar. Erste Recherchen (am 24.11.) zum Divisionsfehler (FDIV- Bug) des Intel Pentium Chips ergaben, dass Intel-Deutsch- land und die Haendler (Vobis) die Sache runter spielen ("...da ist kein Fehler") oder sich ahnungslos geben. --- Dieses Verniedlichen ist sicher ein noch viel groesserer Fehler als der Defekt des Chips selbst. Irgendwie 'checken' die Firmen noch immer nicht, dass die Welt - Dank des Internets - zu einem sehr grossen Dorfmarktplatz geworden ist, auf dem sich Wahrheiten nicht mehr so einfach vertuschen lassen. Es ist zu erwarten, dass die Medien in diesem unglaublichen Fall von "Volksverdummung" keinen Pardon kennen werden. Dabei ist der im Internet schon Anfang November 1994 bekanntgewordene Fehler sehr ernst zu nehmen. Denn in den Wissenschaften - und nicht nur dort - muss man sich 100%-ig auf die Richtigkeit a l l e r Rechenoperationen eines Mikroprozessors oder Computers verlassen koennen. Wir wissen nun seit dem Report vom 23.November 1994 des globalen Fernsehsenders CNN-International (Business News), dass wir uns auf den Pentium-PC eben nicht verlassen koennnen. Oder, was ist von einem Computerchip zu halten, der bei der Berechnung von zum Beispiel z = x - (x/y) * y mit x = 4195835.0 und y = 3145727.0 nicht z = 0 erhaelt, sondern stolz z = 256.0 ^^^^^ ^^^^^^^^^ Der Pentium errechnet fuer x/y = 1.3337391, richtig waere aber 1.3338204. Bereits das 5.Digit ist falsch. Von falschen 4.Digits wird inzwischen auf dem Internet berichtet. Und solchen Mist berechnet der Pentium fuer sehr viele kritische Zahlenpaare (Einzelheiten siehe Datei PBUG_MODEL.TXT auf diesem Ftp-Server sowie in den wdv-notes Nr.327 in der File WNOTES_327.PS_ZIP). Wie weiss man, ob irgendein (gekauftes) Programm nicht in irgendeiner Zwischenrechnung einer Kettenrechnung irgendwo auf eine solch kritisches Zahlenpaar ("magic numbers") stoesst und dann falsch dividiert? Um es ganz klar zu sagen, mit dem fehlerhaften Pentium- Chip koennen ernsthaft keinerlei wissenschaftliche Aus- wertungen vorgenommen werden, geschweige denn die damit bereits produzierten Ergebnisse publiziert werden. Der Pentium muss sofort kostenlos durch einen voll funktionstuechtigen Pentium-Chip ersetzt werden, und das muss natuerlich weltweit mit allen Pentium-PCs, nicht nur fuer die priviligierten Anwender, passieren. Intel hat kein Recht, irgendwelche Fragen zu stellen. Es muss ausreichen, dass man einen defekten Pentium-Computer besitzt. Vom Neukauf eines Pentium-PCs muss aus der Sicht des computererfahrenen Ingenieurs sehr eindringend abgeraten werden: "Don't buy a Pentium-PC!" Vorschlaege im Internet fuer eine neue (der Wahrheit entsprechende) Intel-Werbung: o Intel inside ---> Insel in tide (Aus: The Economist) ^ ^ ^ ^ o Intel inside - but can it divide? o Intel inside - can't divide! o The Intel Pentium - Right "most" of the time! o The Intel Pentium - The Chip that redefines Math! Auf unserem kleinen Ftp-Server stehen inzwischen weitere Infodateien zur allgemeinen Verfuegung. Darunter ist u.a. auch eine formatierte Kurzdokumentation (2 Seiten) im PostScript-Format (komprimiert). Zum Drucken benoetigt man einen PostScript-Drucker. Die Datei heisst WNOTES_307.PS_ZIP. Einzelheiten zu den Infodateien entnehmen Sie bitte der aktuellen Textdatei AAA_READ.ME im Verzeichnis "PC". Ich will mich bemuehen, moeglichst viel Information zu diesem Intel-Skandal zusammenzutragen. -- ||||||| Karl-Heinz Dittberner | Arnimallee 22 | | | | Scientific Data Processing | D-14195 Berlin, Germany | | | | Free University Berlin | Phone: (+4930) 838 2531 |w|d|v| E-Mail: wdv@vaxser.grumed.fu-berlin.de (Internet) -------------------------------------------------------------------------------- Es folgen nun die Internet-Infos: ================================ Subj: Intel's official policy statement. The following is the full text of document #9788 from Intel... Intel Pentium(TM) Processor: Floating Point Unit Information INTEL INFORMATION ON THE SUBTLE FLAW OF THE FLOATING POINT UNIT OF THE PENTIUM(TM) PROCESSOR There has been a lot of communication recently on the Internet about a floating point flaw on the Pentium(TM) processor. For almost all users, this is not a problem. Here are the facts. Intel detected a subtle flaw in the precision of the divide operation for the Pentium Processor. For rare cases (one in nine billion possible divides), the precision of the result is reduced. Intel discovered this subtle flaw during ongoing testing -- after several trillions of floating point operations in our continuing testing of the Pentium processor. Intel immediately tested the most stringent technical applications that use the floating point unit over the course of months and we have been unable to detect any error. In fact, after extensive testing and shipping millions of Pentium processor-based systems, there has only been one reported instance of this flaw affecting a user, to our knowledge. In this case, a mathematician doing theoretical analysis of prime numbers and reciprocals saw reduced precision at the 9th place to the right of the decimal. In fact, extensive engineering tests demonstrated that an average spreadsheet user could encounter this subtle flaw of reduced precision once in every 27,000 years of use. Based on these empirical observations and our extensive testing, the user of standard off-the-shelf software will not be impacted. If you do this kind of prime numbers generation or other complex mathematics, call 1-800-628-8686 (International 916-356-3551). If you don-t, you wont encounter any problems with your Pentium processor-based system. if ever in the life of the computer this becomes a problem, Intel will work with the customer to resolve the issue. 22 November 1994 Intel FaxBack # 9788 Page 1 -------------------------------------------------------------------------------- Subj: Intel Pentium Bug: Scary Business? Xref: math.fu-berlin.de comp.sys.ibm.pc.hardware.chips:14937 comp.sys.ibm.pc.hardware.systems:9026 comp.sys.ibm.pc.hardware.misc:14914 comp.sys.intel:20413 Newsgroups: comp.sys.ibm.pc.hardware.chips,comp.sys.ibm.pc.hardware.systems,comp.sys.ibm.pc.hardware.misc,comp.sys.intel Path: math.fu-berlin.de!zrz.TU-Berlin.DE!CNB.CompuNet.DE!netmbx.de!Germany.EU.net!wizard.pn.com!satisfied.elf.com!news.mathworks.com!hookup!olivea!news.hal.COM!decwrl!tribune.usask.ca!quartz.ucs.ualberta.ca!news.ucalgary.ca!bauwens From: bauwens@acs.ucalgary.ca (Luc Bauwens) Subject: Intel Pentium Bug: Scary Business? Message-ID: Date: Sat, 12 Nov 1994 00:37:15 GMT References: <3a0spg$gn4@nntp.Stanford.EDU> Organization: The University of Calgary, Alberta Lines: 51 Folks, if what I read here is right, that is, *if it's true that* 1. Intel knew about this since three or four months ago; 2. They decided to keep their mouth shut and; 3. They stealthily updated their CPU, hoping that nobody would notice Then the message is clear: that Intel products are intended as toys, and that no responsible professional can afford to use them for any work that he has to assume responsibility for. If the bridge or the building collapses, who is going to go to jail? You or me, or Mr. Andy Groves (or what's his name?). Bugs happen. But it would seem to me that any responsible manufacturer, in that situation, would have an obligation to go out of their way to make sure that their customers become aware of the problem *and stop using a dangerous device* ASAP. But what does Intel? This is unbelievable; what world are they living in? They try ducking the problem, indulge in wishful thinking and keep their finger crossed that nobody will find out and that the problem will go away. (And in the end, of course somebody does find out. Thank god.) We have been paying a premium price for Intel products. Presumably because Intel is a reliable, trustworthy company. Now this :-(. And for the folks who think this is an esoteric issue that won't have any practical consequences, they are quite wrong. It's just the opposite. If the bug were elsewhere and not in the FPU, then it would merely make your computer crash at random or something like that. People would have the right to be pissed off, but there would be no serious consequences. In the FPU, however, the problem goes unnoticed, but quite possibly a substantial volume of work was performed using Pentium CPUs that is potentially quite wrong and in some cases, potentially life-threatening. If I were using a Pentium to design potentially structures of devices which can potentially fail and threaten lives, I would definitely find myself in a position whereby I would feel obliged to *redo and check* all the work done on the faulty CPU. Will Intel pay for the billions of Dollars of lost work? Folks, this is a major issue, and it's going to be interesting how Intel will now deal with damage control. How has Wall Street reacted so far? LB -------------------------------------------------------------------------------- Subj: Re: Intel Pentium Bug: Scary Business? Xref: math.fu-berlin.de comp.sys.ibm.pc.hardware.chips:14947 comp.sys.ibm.pc.hardware.systems:9037 comp.sys.ibm.pc.hardware.misc:14925 comp.sys.intel:20428 Newsgroups: comp.sys.ibm.pc.hardware.chips,comp.sys.ibm.pc.hardware.systems,comp.sys.ibm.pc.hardware.misc,comp.sys.intel Path: math.fu-berlin.de!cs.tu-berlin.de!zrz.TU-Berlin.DE!CNB.CompuNet.DE!netmbx.de!Germany.EU.net!EU.net!howland.reston.ans.net!agate!darkstar.UCSC.EDU!news.hal.COM!decwrl!tribune.usask.ca!quartz.ucs.ualberta.ca!news.ucalgary.ca!bauwens From: bauwens@acs.ucalgary.ca (Luc Bauwens) Subject: Re: Intel Pentium Bug: Scary Business? Message-ID: Date: Sat, 12 Nov 1994 06:40:18 GMT References: <3a1kvd$r7r@nntp.Stanford.EDU> Organization: The University of Calgary, Alberta Lines: 65 In article <3a1kvd$r7r@nntp.Stanford.EDU> ibrahim@leland.Stanford.EDU (Nabeel Robert Ibrahim) writes: >Luc Bauwens wrote: >>In the FPU, however, the >>problem goes unnoticed, but quite possibly a substantial volume of >>work was performed using Pentium CPUs that is potentially quite wrong >>and in some cases, potentially life-threatening. > >OH, COME ON!!! This is getting ridiculous... >Every damn Intel hater is now going around the net spreading >lies like the error was "life-threatening." There is no >evidence to support this blatant fabrication. >Furthermore, the idea that this bug may have affected >a "substantial volume of work" is plain wrong. Why don't >you read a little more before you go spouting some BS across >the net? Cool down, OK. I am not an "Intel hater". I don't really care about them either way. I just think this is a very serious matter. And that Intel blew it big. (About the potential magnitude and scope of the bug, see below.) >>If I were using a Pentium to design potentially structures of devices >>which can potentially fail and threaten lives, I would definitely >>find myself in a position whereby I would feel obliged to *redo and >>check* all the work done on the faulty CPU. > >No self-respecting company would use a PC to design >large structures...they have workstations for a reason, >you know. Perhaps. But small structures collapse too, you know. [...] >The only thing that Intel has to deal with are arrogant Intel >haters who enjoy making it sound as if a very minor, obscure >bug will bring about the end of the world. I think you grossly underestimate the seriousness of the bug. Consider the following, pulled from the Intel newsgroup: >From: jgm@doug.uucp (James MacKinnon) >Subject: Re: FDIV bug in Pentium chips! >Date: 9 Nov 1994 20:55:31 GMT >Organization: Department of Economics >Lines: 50 >NNTP-Posting-Host: doug.econ.queensu.ca >Summary: more about Pentium double precision divide bug > >The FDIV bug in Pentium chips appears to be more severe than some >previous posts have indicated. It appears to occur whenever the >numerator is a power of two (including 2^0 = 1) and for a large number >of different values of the denominator. In particular, the denominator >0.824633702441E+12 that was posted here is not an isolated one. The bug >also occurs for every whole number from 0.824633702430E+12 (and probably >lower) through 0.824633702449E+12 (but not for 0.824633702450E+12). If you do iterative methods, this sort of thing could throw you off to outer space, you know. Give results that are totally unrelated to the problem that you are really trying to solve. Luc -------------------------------------------------------------------------------- Subj: Re: FDIV bug in Pentium chips! Path: math.fu-berlin.de!cs.tu-berlin.de!zrz.TU-Berlin.DE!netmbx.de!Germany.EU.net!EU.net!howland.reston.ans.net!torn!news.ccs.queensu.ca!doug!jgm From: jgm@doug.uucp (James MacKinnon) Newsgroups: comp.sys.intel Subject: Re: FDIV bug in Pentium chips! Date: 12 Nov 1994 17:01:17 GMT Organization: Department of Economics Lines: 46 Sender: jgm@doug.econ.queensu.ca Message-ID: <3a2scu$crj@knot.queensu.ca> References: <39rd03$re9@knot.queensu.ca> In a previous article, referring to the Pentium FDIV bug, I wrote >>This appears to be a very serious hardware error in all Pentium chips. In response, Ron Guilmette (rfg@segfault.us.com) wrote > Oh come on now. Very serious? As another poster noted, you would have > to have some mighty specialized code even to be able to NOTICE the problem. I stand by my original statement (except that the "all" may no longer be true; see below). It is undoubtedly true that many Pentium users will not be affected by this bug. But anyone who does floating point calculations for which it is necessary to use real*8 variables in Fortran, or doubles in C, could be very seriously affected. I have lots of programs which work just fine when certain variables are real*8 and produce garbage when those variables are real*4. This bug means that, at some critical point in one of my programs, such as when doing a pivot step in a matrix inversion, a result that should be accurate to 15 digits may only be accurate to 8 or so. I cannot possibly trust the results of such a program if it is run on a buggy Pentium computer. Several posters have reported that Intel claims to have fixed this bug in recent versions of the Pentium. Can anyone independently verify this claim? That is, has anyone run Terje Mathisen's test program on a recently bought Pentium system and had it report that the bug is *not* present? It is present in every Pentium that I have tested. Frankly, I am very disappointed by the way Intel is responding to this problem. Their public statements about how rare the problem is may seem reasonable to people who know nothing about floating point calculations, but they provide no reassurance at all to those of us who know something about the issues, and they smack of deliberate disinformation. Dell Canada, from whom we have bought two P-90 machines specifically for doing number crunching, says that Intel has not yet come up with any way for Dell to fix the machines it has sold. I have informed Dell that we will not purchase any more Pentiums, from them or anyone else, until the existing machines have been fixed and we can be assured that all new machines operate correctly. -- James G. MacKinnon Department of Economics phone: 613 545-2293 Queen's University Fax: 613 545-6668 Kingston, Ontario, Canada Internet: jgm@qed.econ.queensu.ca K7L 3N6 -------------------------------------------------------------------------------- Subj: Intel fixes minor flaw in new chip Path: math.fu-berlin.de!zib-berlin.de!ceres.fokus.gmd.de!nntp.gmd.de!xlink.net!howland.reston.ans.net!news.sprintlink.net!malgudi.oar.net!infinet!alancz From: alancz@infinet.com (Alan Czarnecki) Newsgroups: comp.sys.intel Subject: Intel fixes minor flaw in new chip Date: 24 Nov 1994 13:51:56 GMT Organization: InfiNet - Internet Access (614/224-3410) Lines: 57 Message-ID: <3b25ps$alh@rigel.infinet.com> An article in my local newspaper this morning by the Knight-Ridder news service states "Intel fixes minor flaw in new chip". The article quoted verbatim reads: More than 2 million of the Pentium chips shipped by Intel Corp. contain a flaw, but the company said it affects only the calc- ulations of theoretical mathematics. The glitch, which has been fixed since its discovery this summer, has created a stir on the Internet, the worldwide communications network widely used by the scientific community. Howard High, a spokesman for the Santa Clara, Calif.-based chip manufacturer, said Tuesday that the bug in the chip's ability to calculate extended mathematical problems was discovered - and fixed - this summer. In creating the Pentium's circuitry - which contains more than 3 million transistors - designers made a small flaw in a floating point unit, which performs math functions. As a result, the Pentiums can make errors in equations involving calculations to the ninth digit on the right side of the decimal point. Even most engineers and financial analysts require accuracy only to the fourth or fifth decimal point, High said. Spreadsheet and word processor users need not worry. High said Intel repaired the error at a cost of perhaps several hundred thousand dollars. At the time, Intel had shipped about 2 million of the estimated 5 million or more Pentiums sold so far. But the glitch has been the subject of hundreds of messages on the Internet, many of which criticized the company for failing to acknowledge the bug. Others questioned the rarity of the problem. "I know it's buzzing all over the 'net'. But there are maybe several dozen people that this would affect," High said. "So far, we've only heard from one. It's reasonably rare." ....end of article... IMHO the approach Intel is taking to this problem is incredibly arrogant in that they must believe they can use the above mis-information to shape public opinion and minimize negative images of their company. Intel Corp. must know that the Engineers and Scientists are recommending "no buy" advice to their purchasing agents for Pentium based PC's, but don't care. Obviously, the profit for the company is seen to come from other than Engineering and Scientific applications. -------------------------------------------------------------------------------- Subj: Re: comp.intel.andy-grove.head.ostrich.sand Path: math.fu-berlin.de!zib-berlin.de!ceres.fokus.gmd.de!nntp.gmd.de!xlink.net!news.urz.uni-heidelberg.de!orion.mgen.uni-heidelberg.de!andreas From: andreas@orion.mgen.uni-heidelberg.de (Andreas Helke) Newsgroups: comp.sys.intel Subject: Re: comp.intel.andy-grove.head.ostrich.sand Date: 24 Nov 1994 12:08:00 GMT Organization: University of Heidelberg, Germany Lines: 21 Message-ID: <3b1vn0$le6@sun0.urz.uni-heidelberg.de> References: Richard Krehbiel (richk@netcom17.netcom.com) wrote: : In article firefly writes: : > 5505001/294911 = 18.66600093 (pentium) : > 5505001/294911 = 18.666651973 (powerpc) : > : > you be the judge. : 5505001/294911 = 18.66665197 (HP 41C calculator) The floating point unit of a 486 and the software emulation of the Borland Pascal compiler agree with the Powerpc. 5505001/294911 = 18.6666519729681 I don't have a Pentium to test it there. Andreas -- * Andreas Helke, Institut fuer molekulare Genetik, Universitaet Heidelberg ** Im Neuenheimer Feld 230, 69122 Heidelberg, Germany *** orion.mgen.uni-heidelberg.de has a ftp server with Unix tools for DOS **** and with the HFM V. 4.03 DOS file manager and archive program shell -------------------------------------------------------------------------------- Subj: FDIV bug: Attempted summary and new p87test Path: math.fu-berlin.de!zrz.TU-Berlin.DE!netmbx.de!Germany.EU.net!news.dfn.de!swiss.ans.net!gatech!swrinde!pipex!sunic!trane.uninett.no!eunet.no!nuug!telepost.no!hydro.com!usenet From: Terje.Mathisen@hda.hydro.com (Terje Mathisen) Newsgroups: comp.sys.intel Subject: FDIV bug: Attempted summary and new p87test Date: 13 Nov 1994 14:39:59 GMT Organization: Hydro Data, Norsk Hydro (Norway) Lines: 136 Message-ID: <3a58g0$75m@vkhdsu01.hda.hydro.com> References: <39rd03$re9@knot.queensu.ca> <39uv8e$s0k@hustle.rahul.net> <3a0b2r$jka@agate.berkeley.edu> <3a3kb4$ke5@nntpd.lkg.dec.com> Reply-To: Terje.Mathisen@hda.hydro.com (Terje Mathisen) Since I started this thread, I'd like to summarize my current views on this bug: - *If* Intel have know about this problem for several months, and *decided* not to warn users about it, those responsible for that decision should be fired. There is no excuse for knowing about a silent bug like this, and not tell about it. - An error that results in a hardware crash is much less serious than this one which returns wrong answers. - Since the bit-pattern that fails seems to be about 18 bits long (..11110111000001...), the odds of hitting it should be about 1/2^18, or 1/6E7. With 8 significant digits in the answer, this means that the average (for truly random data) precision on a Pentium is less than 16 digits. - I have nothing against x86 cpu's, rather the opposite, since I love writing asm code for them. The company I work for will still buy new Pentium machines, but as soon as I first heard and verified the problem, I specified that new machines would only be aquired from dealers who promized to exchange the cpus as soon as Intel makes the fixed p5s available. Since I first posted p87test, I've gotten email from people who tell me that it thinks their Pentium-90 is a 386. This will happen if you are running under a V86 monitor which virtualizes the eflags register and disallow toggling the AlignmentControl (bit #18) and/or CPUID (bit #21). The new version will perform the test for FDIV precision independently of cpu version, as long as a numeric coprocessor seems to be present. I have also found a bug in the program, where I demanded that the result after FDIV/FMUL should be identical to 1.0. I have changed this to allow a small ulp error. To make the program more useful, it will now also report (and document) the 'secret' CPUID feature bits, on a Pentium or new-model 486. -Terje Mathisen (include std disclaimer) "almost all programming can be viewed as an exercise in caching" -------------------------------------------------------------------------------- The Pentium test "p87test.zip" is now available from in directory [.PC]. -------------------------------------------------------------------------------- And an Intel employee tells us: Subj: Time to think about using another processor. Newsgroups: comp.sys.intel Path: math.fu-berlin.de!zib-berlin.de!ceres.fokus.gmd.de!nntp.gmd.de!xlink.net!howland.reston.ans.net!ix.netcom.com!netcom.com!rlaw From: rlaw@netcom.com (reginald law) Subject: Time to think about using another processor Message-ID: Keywords: Pentinum Bug Organization: NETCOM On-line Communication Services (408 261-4700 guest) Date: Fri, 25 Nov 1994 06:48:10 GMT Lines: 6 With these problems with the intel chip, it might be time to think about using a MIPS, Alpha, or a Power PC platform. With NT being ported to these platform and 32bit applications available it might be time to get of the intel platforms. Truly amazing that we ship a faulty product. ^^^^^^^^^^^^^^^^^^^^^^^^^^ -------------------------------------------------------------------------------- Subject: FDIV bug: Where are the good chips? Path: math.fu-berlin.de!zib-berlin.de!ceres.fokus.gmd.de!nntp.gmd.de!xlink.net!howland.reston.ans.net!europa.eng.gtefsd.com!news.mathworks.com!newshost.marcam.com!zip.eecs.umich.edu!newsxfer.itd.umich.edu!nntp.cs.ubc.ca!torn!uunet.ca!uunet.ca!io.org!nobod From: boychuk@io.org (Dennis Boychuk) Newsgroups: comp.sys.intel Subject: FDIV bug: Where are the good chips? Date: 24 Nov 1994 19:00:21 -0500 Organization: Internex Online (io.org) Data: 416-363-4151 Voice: 416-363-8676 Lines: 27 Message-ID: <3b39el$odj@ionews.io.org> Summary: Conspiracy theories. The New York Times article suggested that the first 2,000,000 of the existing 5,000,000 chips are faulty. Given that many recently delivered chips are reported to be faulty, this is doubtful. If there is a fix and good chips have been produced since September, where are they? In the absence of sound information from Intel, it's time for some conspiracy theories. If there was a fix, has Intel been giving them to "important" customers, perhaps with a non-disclosure agreement? Has Intel been *reselling* the bad chips to other buyers to this day? Flame: Consipacies aside, Intel's whitewash has been totally disgusting. They have destroyed their good will. X,000,000 high-end users are the wrong people to ·´“orip off. If Intel is so slimy as to refuse me a free replacement, I will have to buy a good chip -- preferably a clone -- and sue them for the cost. For my next machines, I will choose among alternatives: clones, or the PowerPC. I am, of course, advising my clients and collegues not to buy a pentium. When this is over, it should be apparent that it would have cost them less to do the right thing fom the start. Flame off. Have a nice day. -- Dennis Boychuk boychuk@io.org -------------------------------------------------------------------------------- Subject: Re: Intel Pentium Bug: Scary Business? Xref: math.fu-berlin.de comp.sys.ibm.pc.hardware.chips:15865 comp.sys.ibm.pc.hardware.systems:9818 comp.sys.ibm.pc.hardware.misc:15902 comp.sys.intel:21934 sci.engr:10678 sci.engr.civil:4639 Path: math.fu-berlin.de!zib-berlin.de!zrz.TU-Berlin.DE!netmbx.de!Germany.EU.net!howland.reston.ans.net!gatech!bloom-beacon.mit.edu!news.bu.edu!ppp-82-7.bu.edu!user From: mek@acs.bu.edu (Mark E. Kern) Newsgroups: comp.sys.ibm.pc.hardware.chips,comp.sys.ibm.pc.hardware.systems,comp.sys.ibm.pc.hardware.misc,comp.sys.intel,sci.engr,sci.engr.civil Subject: Re: Intel Pentium Bug: Scary Business? Date: 26 Nov 1994 17:31:28 GMT Organization: BU Law Lines: 35 Message-ID: References: <3agjvn$hmc@Ra.MsState.Edu> <3al4ok$a2l@clarknet.clark.net> <3arfum$4b9@news.primenet.com> <3atbor$pln@mark.ucdavis.edu> Intel has handled this rather badly. Although bugs are not unheard of in microprocessors, Intels actions have been pretty arrogant and high-handed: 1) They kept it quiet, without alerting engineers to the problem. 2) They failed to distinguish the fixed chips from the defective ones, hoping you wouldn't notice. 3) They refuse to replace (even at discounted cost) your CPU unless you meet some nebulous, unpublished critere -, even though you paid for a working chip in the first place. 4) They are spreading misleading statistics to the media in order to silence this group with big money and media contacts. 5) The official statement contains absolutely no apology, no remorse, no "thank you for bearing with us while we sort this out" message. Instead, it downplays the bug and by implication, the reputation of those who have complained. If you just sit here, Intel will weather this, and it will soon be forgotten. Then, Intel will go on treating you as they have in the past - taking you for granted, and telling you what you should think. You need to send a clear message to Intel. There are only a few ways to do it. The best way, would to call you P5 system vendor, and demand *loudly* that they take the system back. After all, you paid the premium for a P5, and you want the high-end features you paid for to *work*. It doesn't matter if they take the system back or not. Your call will be duely noted, and if enough calls come in, they will really put the heat on Intel. They have the clout to do it, while individual posters do not. Markus -------------------------------------------------------------------------------- Subject: Lot Number & FDIV Path: math.fu-berlin.de!zib-berlin.de!zrz.TU-Berlin.DE!netmbx.de!Germany.EU.net!EU.net!uunet!haven.umd.edu!cville-srv.wam.umd.edu!kpla From: kpla@wam.umd.edu (Kok Ping Liew) Newsgroups: comp.sys.intel Subject: Lot Number & FDIV Date: 27 Nov 1994 05:57:49 GMT Organization: University of Maryland College Park Lines: 58 Message-ID: <3b974t$bdr@cville-srv.wam.umd.edu> Here's how to read the CPU lot number according to Intel's faxback (CPU Warranty): eg. L4125833 L = Plant Code 4 = Last digit of year when it was produced 12 = the work week of year when it was produced Besides physically reading the lot number off the chip, are there software that can read off similar info? I work part-time as a computer technician for a small IBM-compatible manufacturer, and having just found out about the FDIV bug in the Pentium I am quite concerned. Fortunately, we haven't sold any Pentium system to date since we generally encourage our customer to stay away from newly released and often buggy products and to buy 486dx2/66 for its price/ performance ratio. But we have just received a large order from our corporate client for P5/90s. You can imagine the dilemma we are facing now. The batch of P5/90s we acquired was dated on the 42nd week of this year. puts around September. We are all keeping our fingers crossed here. Once they are tested for FDIV, I will post the results. Subsequently, if they are FDIV positive we will most certainly recommend to our client against buying any Pentium until the FDIV-bugfree Pentiums are made available. This incident of Intel's stubbornness to replace the affacted chips upon request has certainly made the decision to consider other alternatives to the Pentium such as K5, M1 and NexGen much easier. Because of our effort to give our customer the best-valued computer system, 50% of all our 486 systems are equipped with AMD's 486 instead of Intel's. If Intel continues to stonewall its customers, this will provide the openning needed to the K5, M1 and NexGen chips. Especially in the small reseller industry, the willingness to try other alternatives is great. With the PC industry being so competitive, services and tech support are fast becoming the prime determining factor in buying decision. And what kind of services can a company claim to render if a known defective product (regardless of frequency) can't be replaced? A customer deserves what he/she pays for. Intel by refusing to replace defective chips (upon request) is taking away the warranty service competitive advantage from the small resellers. I don't have exact figures, but I would guess that the total sales volume of all the small resellers is at least equal if not more than the combined sales of all major PC manufacturers. That is a hugh market to risk alienating. I think Intel is too confident about its "Intel Inside" media compaigne. Often all an ordinary customer cares about is price, services and Windows. Seldom do they care about the chip used inside if Windows can run fine on the outside. And without sufficient technical background they rely heavily on the recommendation of the salesperson and technician. With the salesperson and technician being more computer-literate, Intel can't get away with their excuse of "it doesn't really affect the average user, so we won't replace it". This message is not intended as a flame toward Intel. It merely offers certain perspective on the FDIV issue. Good day. -------------------------------------------------------------------------------- Subj: Re: NEWS: SF Chronical Reports HUGE Error in Pentiums Possible Path: math.fu-berlin.de!zib-berlin.de!prise.nz.dlr.de!news.dfn.de!darwin.sura.net!gatech!swrinde!hookup!news.moneng.mei.com!uwm.edu!alpha1.csd.uwm.edu!ggraef From: ggraef@alpha1.csd.uwm.edu (Gerald Luther Graef) Newsgroups: comp.sys.intel Subject: Re: NEWS: SF Chronical Reports HUGE Error in Pentiums Possible Date: 3 Dec 1994 19:09:20 GMT Organization: University of Wisconsin - Milwaukee, Physics Department Lines: 26 Message-ID: <3bqfp0$qgo@uwm.edu> References: <3bp4gp$130@yuggoth.ucsb.edu> In article <3bp4gp$130@yuggoth.ucsb.edu> schalit@rot.ucsb.edu (Matthew Schalit) writes: >The San Francisco Chronical had a fascinating article on the FDIV bug. >It is an artical that appeared in print 12/1/94, Thursday in the Business >section. It mentions a Mark Edelstone, analyst with Prudential >Securities Research who gives an error result in the fourth digit of a >four significant digit number. This is the LARGEST error I've heard >reported. Lets contact him and get a posting to this group. Anybody >have any connections to Prudential? Maybe, but does it matter? Intel releases have already stated that the FDIV error can be this large: I> However, Intel has run trillions of different random divides to validate I> that the occurrence is less than one in nine billion: the precision of I> the result ranges from the 4th to the 19th significant digit (or the I> 4th to the 19th place after the decimal point in numbers expressed in I> scientific notation, an example of which is 0.12345 x 105). Statistically, I> for these values, the 4th significant digit is the least likely position I> for an error. Importantly, intensive testing has shown that no known I> physical constant causes the reduced precision. Once they opened up, Intel has been very forthcoming about the bug: we don't need to add to the propaganda by implying that the bug is worse than anyone thought. (The above information is available on the intel www server, http://www.intel.com). -------------------------------------------------------------------------------- Subject: Re: Pentium - Intel White Paper Reveals More Problems Newsgroups: comp.sys.intel Path: math.fu-berlin.de!zib-berlin.de!prise.nz.dlr.de!news.dfn.de!darwin.sura.net!jhunix1.hcf.jhu.edu!mozart.amil.jhu.edu!europa.eng.gtefsd.com!uhog.mit.edu!news.media.mit.edu!rnewman From: rnewman@media.mit.edu (Ron Newman) Subject: Re: Pentium - Intel White Paper Reveals More Problems Message-ID: <1994Dec4.045038.15398@news.media.mit.edu> Sender: news@news.media.mit.edu (USENET News System) Organization: Massachvsetts Institvte of Technology References: Date: Sun, 4 Dec 1994 04:50:38 GMT Lines: 28 In article , wrote: > >I was faxed a copy of Intel' November 30th 1994 white paper entitled: > >" Statistical Analysis of Floating Point Flaw in the Pentium (TM) Processor" > >The paper was authored by H.P. Sharangpanl and M.L. Barton, Ph.D. of Intel > > >The paper reveals that Intel has NOT been telling us the complete story. >Buried in this (31 page) report of confussion is the fact that other functions >of the floating unit are effected by the FDIV "flaw" (BUG). Here's a qutoe >from the white paper: "Since this divide operation is used by the Divide, >Remaindering, and certain Transcendental Instructions, an inaccuracy introduced >in the operation manifests itself as an inaccuracy in the results generated by >these instructions. Post as much of this as you feel like typing in. See if you can get an electronic copy to put on a Web server somewhere. I'd *very* much like to see an example of an inaccurate FPREM, FPREM1, FCOS, FSIN, or the like.... Is this paper available from Intel's "Fax-back" server? -- Ron Newman MIT Media Laboratory rnewman@media.mit.edu -------------------------------------------------------------------------------- Subject: Pentium, what really is the truth ? Path: math.fu-berlin.de!zib-berlin.de!prise.nz.dlr.de!news.dfn.de!swiss.ans.net!howland.reston.ans.net!pipex!sunic!news.funet.fi!news.csc.fi!nokia.fi!tnclus!autiokar From: autiokar@tnclus.tele.nokia.fi (TIMO AUTIOKARI, P: INT+358 0 511 5280, FAX: INT+358 0 511 6496) Newsgroups: comp.sys.intel Subject: Pentium, what really is the truth ? Date: 4 Dec 94 13:07:42 EET Organization: Nokia Telecommunications. Lines: 54 Message-ID: <1994Dec4.130742.1@tnclus> It seems now, that Intel has unknowingly released a chip with a _systematic_ functional failure in it. My question is, is this at all possible ? And if it is, what can be concluded from that. During the design phase of any new integrated circuit, the design undergoes many extensive test phases. To prevent bugs to propagate, tests and simulations are _not_ drawn from the desing-data. This is common practice. The engineering samples will be tested thoroughly to (and over) the limits of the chip in concern. The Outgoing Inspection Test is created, independently from the desing-data. Now, a systematic functional failure is clearly the easiest type of failure to detect. Among the failures that component users sees, this type of failure is next to non-existent. There are a lot of other failure types involved with integrated circuits. These are, by their nature, much more difficult to detect. To mention _only_ a few: 1. open connections between the different layers of the conductors on the chip. These may be seem to operate at some suitable speed, due to the small capacitance that is formed over the submicrometer gap. But eg. disturbance in the electro-magnetic field will produce a random functional failure. 2. Parametric drifts that are induced by change of temperature may cause a failure. Parameter may be DC-parameter (current, voltage etc) or an AC-parameter (timing). 3. Leakage current caused by an impurity particle that is buried into the chip at the process time, may prevent a node to operate properly.The impurity particle comes from the air or from some material used in the production. It is a clear a manifestation of really week test coverage and quality control, when device with systematic functional failure goes out from a manufacturer. The example 3 is classical type of RANDOM functional failure. Even that kind of difficult to find failure will be detected in most cases at the Outgoing Inspection Test performed by the manufacturer. I find it rather tough to believe, that Intel would have so week test-coverare, that the simplest failure-type can pass the Outgoing Inspection. Not to mention the many pre-release test phases. So, either the test coverage really is poor (this makes the overall quality and reliability of the product very questionable) or the systematic functional failure was known somewhere in the very beginning. Timo --- My views are mine, only mine and no-one else but mine. -------------------------------------------------------------------------------- Subject: Re: Why didn't Intel test the Pentium FDIV *tables*? Xref: math.fu-berlin.de comp.arch.arithmetic:608 sci.math.num-analysis:16158 comp.sys.intel:30464 Path: math.fu-berlin.de!zib-berlin.de!news.mathworks.com!europa.eng.gtefsd.com!library.ucla.edu!agate!news.Stanford.EDU!Sunburn.Stanford.EDU!pratt From: pratt@Sunburn.Stanford.EDU (Vaughan R. Pratt) Newsgroups: comp.arch.arithmetic,sci.math.num-analysis,comp.sys.intel Subject: Re: Why didn't Intel test the Pentium FDIV *tables*? Date: 21 Dec 1994 16:08:45 GMT Organization: Computer Science Department, Stanford University. Lines: 29 Distribution: inet Message-ID: <3d9jud$sm5@Radon.Stanford.EDU> References: <1994Dec14.200603.28449@wdl.loral.com> <3d84mj$44v@news1.deltanet.com> In article <3d84mj$44v@news1.deltanet.com>, David K. Every wrote: >As I understand it... it is not testing the entries that causes >problems... but you have to test various SEQUENCES through the tables. >Which gets you back to the first problem... lots of potential >sequences. A simple but effective check on one's program is a test suite that exercises every branch of the program. This does not constitute complete verification by any means, yet it catches a remarkable number of bugs, and is among the most cost effective software tests. For hardware the simplest failure mode is for the power cord to be left unplugged. The second simplest is for wires to be stuck at 0 or 1. Any manufacturer planning to sell more than a billion dollars worth of a chip, that does not include among its test suite for that chip a test that exercises every wire to verify that it is stuck at neither 0 nor 1, does a gross disservice to its stockholders. In the case of the Pentium division flaw, such a test would have <<<<<<<<<<<|| immediately found all five missing table entries. <<<<<<<<<<<|| <<<<<<<<<<<|| This is in hindsight. I am amazed that it takes a problem of the <<<<<<<<<<<|| magnitude of the Pentium division error to wake us all up to such <<<<<<<<<<<|| simple realities. <<<<<<<<<<<|| -- Vaughan Pratt FTP: boole.stanford.edu:/pub/FDIV/README 4.999999/14.999999 has 6 3's, not 4 http://boole.stanford.edu/boole.html --------------------------------------------------------------------------------