FUD Snuffing AMD JPEG Issue

Posted By Van Smith

Date: July 25, 2001

With sad predictability an Internet sub-community of message board posters has attempted to magnify a news report made here Monday regarding a minor manufacturing defect in a small number of AMD's processors.


Misinformation War For Profit and Position

It is an ugly Internet secret that members of the investment community visit high traffic message boards posting responses made to drive down a company's stock value so that the poster can profit by shorting the affected stock. 

Also known is that corporate insiders do much the same thing in efforts to aid their company's fortunes while undermining rival companies.  Additionally, it is a possibility if not likelihood that some of these posters are paid to undertake such covert guerilla tactics judging by the sheer quantity of corporate FUD (Fear-Uncertainty-Doubt) dumped on some of the more influential message boards.  

The true extent of these practices is not known since the Internet can afford a level of anonymity, but it appears to be widespread according to remarks I have gathered from a few webmasters.  If you have any evidence supporting such conjecture, please contact me.


The AMD JPEG Issue

AMD became the target of such actions yesterday regarding the minor flaw in a few processors that would lead to distorted image rendering and sound problems in narrowly specific conditions.

According to AMD, the so called "JPEG issue" was a manufacturing test problem and empirical evidence supports this assertion.  Even though the gap in AMD's testing procedures allowed a small number of faulty processors to trickle out over time, the fact that the vast majority of processors were fine despite this oversight, suggests that the problem is easily correctable.  By implementing more specific testing procedures, effectively all chips exhibiting the JPEG problem can be caught without impacting yields with any numerical significance.

AMD asserts that it has implemented such measures to their manufacturing test process.  Anecdotal evidence seems to support this, as no claims for newer Athlons having this flaw have been publicly made.

The gentleman who made the JPEG problem widely known via a report on his family's site, Ron Gordon, states up front:

This defect has been observed in only a small proportion of AMD microprocessors. The majority of AMD microprocessors operate correctly.

Later on in this webpage he states:

This problem does not appear to be caused by a fundamental design flaw with the AMD-K6-2 microprocessor, and the vast majority of these microprocessors operate properly. Evidently, certain sequences of instructions on some AMD-K6-2 microprocessors do not yield the correct results at the rated speed.


Propaganda of Misleading Comparisons

A few FUD posters, after testing the waters, have even tried to compare the apparently isolated AMD JPEG problem with the Pentium floating point division bug of several years ago.  The Pentium defect was a design flaw in its division lookup tables that impacted *all* chips manufactured over a certain period of time and, unlike claims to the contrary by revisionist historians, the problem was an issue for *everyone* using the chips -- it was a matter of Russian Roulette if and when the problem would bite you.  Furthermore, the problem could occur in programs as innocuous and commonplace as Windows Calculator or spreadsheets.

For instance any Pentium with the defective stepping would incorrectly execute the following division in Windows Calculator:

4195835 / 3145727  = 1.333 820 449 136 241 000  (Correct value)
4195835 / 3145727 = 1.333 739 068 902 037 589  (Flawed Pentium)

Intel was not forthcoming on this issue.  Even though the company became aware of the problem when the design was undergoing validation, the chipmaker chose not to publish the errata or even disclose it to its closest OEMs.  After the company finally admitted to the design flaw, it then initially required each customer to give Intel a good reason why the customer needed a new chip before the Santa Clara, California, company would send a replacement part.

Intel's actions were far from magnanimous.  Only under intensifying customer outrage combined with IBM's threat to stop shipping Intel's chips did the company back down and offer replacements for all defective Pentiums.


P4 Throttling a More Apt Analogy

A much better comparison can be made between AMD's JPEG issue and the Pentium 4 chip that I witnessed throttling.  I believe that the individual 1.7 GHz Pentium 4 that throttled (measurably slowed down under load) while running Quake III was a defective chip that eluded the chip giant's quality control measures for the still young design.  I do not believe that this is a widespread problem and it almost certainly doesn't warrant a recall. 

I do believe strongly, however, that other Pentium 4 chips exhibiting the same problem under the same circumstances probably do exist in small numbers.  I trust that Intel, like AMD, will honor its warranties and replace these defective chips at the owner's request once these problems are spotted.

AMD, for its part, advises any users who have CPUs impacted by the JPEG flaw to utilize its technical support site or call them at (408) 749-3060.  Mr. Gordon has a procedure for testing any chip you think might have this bug.  If you are running Internet Explorer 4 or 5, simply go to his main AMD page and view the test images on his site.


Just Say "No" To Corporate Trolls

People who post on Internet message boards with the deliberate intention to incite responses usually for insincere or calculated reasons are known in Web parlance as "trolls."

Please keep perspective on the AMD JPEG issue and beware the corporate trolls.