Feb 95 Viewpoint
|Column Tag:||The Editors Page
The Editors Page
By Scott T Boyd, Editor
Learning From Others Mistakes
If the world didnt know about Intels Pentium processor before, they sure do now. Intel spent millions last year flying their logo around TV sets to establish their brand image. If only theyd known how much cheaper it would be to have Andrew Grove, Intel President and CEO, apologize on the Internet. Never mind that many on the net took offense that Intel would choose when and whether to replace a customers chip. Intel not only seemed reluctant to own up to their problem (they did, after all, wait several months until someone else discovered it independently, to mention it), but much of what they said sounded more like they were upset that they got caught.
Those of us Macintosh supporters smugly rubbing our hands together with glee and thinking, Goody, goody! Intels in trouble, may have another thing coming. Intel hurt their reputation, right? Was it just my imagination, or was that a full-length Pentium ad shown on MacNeil/Lehrer, Night Line, and every other major news program? Will the market remember Intels mistake? Or will they remember over a full week of coverage? Will they remember the names Intel and Pentium? and that Intel eventually took care of their customers? Even with a chargeback of tens of millions of dollars to pay for the chip replacements, Intel may have scored an advertising coup like weve never seen before.
For the Morbidly Curious
After reading Intels white paper (http://www.intel.com), IBMs position paper (http://www.ibm.com), and Intels rebuttal to IBMs paper, there seems to be no disagreement about the source of the bug (missing entries in a lookup table caused by a flawed script written to download entries into a hardware PLA (Programmable Lookup Array)). They also seem to agree about the worst-case impact on any single operation (inaccuracies can occur starting with the 4th significant decimal digit).
How politics affected the presentation of the data shows in the assumptions each party made. Intel assumed that the average spreadsheet user about 1000 floating point divides on any given day, and that numbers are uniformly distributed. They also used spreadsheets from all around Intel to determine the probability of occurrence for bad number combinations.
IBM, on the other hand, figured that the average spreadsheet user would spend about 15 minutes a day recalculating, and would get one divide per 16,000 instructions when recalculating. At 90MHz, thats about 4687 divides/second, or 4.2M per day. They also assert that all bit patterns are not equally probable. They created random numbers in a variety of common decimal patterns, and used them to create numerators and denominators. Their observations led them to believe that one out of every 100 million divisions might lead to bad results.
Intel says one error every 27,000 years. IBM says every 24 days. Do you hear the axes grinding? Bottom line? Intel takes a $35M-$70M charge against earnings to give customers what they want (and now they know whats inside and that they want one).
Hungry and Sleepy?
I recently got a Connectix QuickCam (the $99 all-seeing eyeball that plugs into a Macs serial port). I plugged it in, installed some software, and all of a sudden my modem was a problem. Not that theres anything technically wrong with the modem or their software - perhaps I should back up and explain a bit.
The camera works with pretty much any QuickTime software. One particularly interesting application is CU-SeeMe, a free application from Cornell University which supports real-time multi-party videoconferencing on the Internet. Even a 14.4 connection will get you video, but itll leave you hungry for more - much more bandwidth, that is. Four of us in three different parts of the country got online with three different cameras and a VCR. Even though we didnt do much more than watch each other smile at the camera, we had so much fun that we were all wondering how to beg, borrow, or steal more bandwidth (know any internet providers who want to trade Macintosh code for a frame relay connection?). We dont know whether it will help our virtual businesses (it sure didnt help our sleep patterns), but we have little doubt that videoconferencing will one day look no more surprising in a home office than a copier, fax machine, or a Macintosh.
Were certainly not going to replace your Pentium chip just so you can play Doom!
- Intel Pentium hotline staffer
Maybe you ought to consider a Macintosh this Christmas.
- Wall Street Journal 15 Dec 1994
I suppose it is the corrected chip that will be called RePentium.
- Peter G. Neumann
Intel - changing the way people think about floating point.
- excerpt from a speech, originally intended as a compliment,
as reported by Jörg Brown
Food For Thought
Right in the middle of a MacNeil/Lehrer News Hour montage of Pentium clips, I saw Apples Graphing Calculator spinning a 3D parabolic equation. They didnt realize they were showing Apples Power Macintosh, not a Pentium box. Do you think they might have been able to see the bugs had it really been running on a Pentium?