Author of AMD JPEG Site Talks to VHJ
Posted By Van Smith
Date: July 30, 2001
Ron Gordon, the man who made the AMD JPEG image anomaly well known, has told VHJ that he agrees with our recent analysis and estimates the problem is "one in a million, plus or minus an order of magnitude."
Mr. Gordon is author of a web page that gained widespread attention recently by claiming that a fraction of AMD K6-2, Athlon and Duron microprocessors contain a flaw that, under specific conditions, causes incorrect JPEG and MPEG rendering. The problem is also associated with certain audio problems.
Several days ago AMD confirmed the issue to VHJ as an oversight in its manufacturing test process that the company has since addressed. According to the surging chipmaker, a gap in its testing procedures allowed a very small number of affected chips to trickle out over the course of several months.
Despite the extremely minor and ordinary nature of the issue, a firestorm of criticism hailed down on many Internet message boards. Much of this negative attention was geared towards painting the problem -- nothing more than an insignificant reality of chipmaking since it is impossible to catch every defective chip in testing -- as if it was comparable to an ubiquitous design defect like the Pentium floating point division flaw of several years ago that served as a major setback for Intel.
We stood by AMD's assessment of the problem, and we applaud the company for it forthrightness. Furthermore, we believe the message board attacks were an all to common FUD (Fear, Uncertanty, and Doubt) campaign delivered to undermine the number two CPU manufacturer's upwards market momentum and stock price.
On the other side of the coin, Mr. Gordon reports that he has been the target of the occasional hate mail accusing him of "just AMD bashing."
Mr. Gordon reported to VHJ his assessment of the JPEG issue in light of AMD's clarification and our analysis:
Surely the number of processors affected must be small. I agree with the overall analysis that this is a manufacturing flaw, and that only a very small percentage of chips passed the standard set of tests, but still possessed this particular flaw. Most chips that are defective fail one of the other tests in their suite. Although the coverage of the test isn't 100%, it is obviously pretty good.
Mr. Gordon also states that for those few having this problem "AMD has always replaced the chip" even in cases when the warranty has expired, which he claims is the period when a majority of the problems appear to be diagnosed. Despite the exceedingly infrequent occurrence of the flaw, Mr. Gordon does lament the lack of official documentation to aid in troubleshooting the issue. This lack of information has made diagnosis typicaly more difficult, he stresses.
Mr. Gordon concludes:
But you are correct, my guess would be that the total number [of affected processors] must not be more than 1 in a million or so, plus or minus an order of magnitude.