Errata articles are the current rage

RECENTLY, THERE have been a couple of articles on errata making the rounds, the first on Intel, the second on AMD. They both make my head hurt. A lot. No, not the errata, but the explanations, and lack of understanding of some simple concepts. When the first one, Intel, came out, I was teetering on the verge of breaking out the cluestick, but thought the better of it. When the conspiracy theory second article came out, well time to spring into action, two days late as usual.

Lets start with the Intel one, the basic gist of the article is that Yonah (hardcore or whatever it is marketeered as now) has an errata list, and that list is 34 long. This is somehow supposed to mean something, and the author bandies about words like chilling and criminal. The list, half as long as the one on the P4 is simply business as usual for Intel, and on par with other chipmakers with devices of similar complexity..

The reason these are published is that people who make things that could possibly be affected by an errata will know about potential problems. That means low level people, BIOS developers, board makers, and OS developers can implement workarounds or simply not use the features listed. It is a living document, with each passing day, more gasp bugs are found, and with each revision, some are removed. This is how the semiconductor industry has worked for decades, and will work for the next few. There is no conspiracy.

As someone who has programmed for game consoles in the past, the list that they have is far longer, far more serious, and almost never gets fixed. General purpose CPUs have a much greater likelihood of putting a fixed CPU's functionality into service, so they tend to get a lot more attention and updates. Again, no conspiracy, and to suggest so is to simply not have a clue about the process of semiconductor design and manufacture.

Then on to the AMD one, again, this one focuses on the alleged conspiracy of AMD to hide errata, and was the one that caused me physical pain. There are two premises in the article not worth mentioning, but I will anyway. It is kind of like picking at a scab, the 'missing' numbers and the 'old data sheet conspiracy'. Sigh. Ow.

Let's start out by saying that the AMD errata numbers are at 136 according to the article, more than twice that of the P4 and four times that of softcore (or whatever it is), and note the distinct lack of screeching. If 34 is chilling and almost criminal, wouldn't 136 be child molester territory, certainly at least schoolyard meth dealer?

So another fundamental problem is that the author does not understand what a family is, and what a superset and subset are. Both boggle the mind, and lead to errors in the article. They also give me a stabbing pain in the temples.

First, the superset business, the word itself is defined here for those who don't get it. Basically if you release a technical document, and then release a new revision, that is fine. If the new one fixed errors in the old one, great, the world is a better place, birds sing and little furry things come out and play with children in the forest. If the new one rewrites things, changes history and glosses over problems, that is a conspiracy. Furry things bite the little rugrats, while according to David Kanter, the birds peck out their eyes, supping on the sweet jellies therein..

If the new doc adds to the old doc, putting new information out without changing the old, that is called a superset. It contains all the old stuff, and new stuff. Birds again sing, furry things play with the kiddies, and candy rains from the sky. The thing is, the author simply can't grasp this concept.

So, the new ones have what the old ones do and more. There are some bugs that are not discovered for years, and may be debilitating, cause hair loss, and turn simple forest creatures mean. They won't be in the old sheets, but will be in the new ones, along with all the new info. If AMD left these sheets out there, there is a good chance that people can and would get the wrong information. In technical terms, this is called either protecting the user or ass-covering, depending on the side you are looking at it from. There is however, no conspiracy. I repeat, there is no conspiracy.

The reason AMD probably does not want this information out is because they don't want people slinging outdated things about them around. It will only cause them problems and benefit no one. Far from being a conspiracy, they are just trying to protect people who can't grasp simple concepts about how the system works.

Then comes the funniest one of the bunch, the missing numbers. There are two problems here, first is that there are holes on the list. The reason for this is simple, several revisions of the chips never see the light of day. The A0 stepping of Woodcrest had a massive FP bug that was fixed in A1, and said bug was on an errata list. No A0s made the fanbio circuit, and precious few trusted insiders got A1s. By the time they get released they will probably be in the Bs somewhere, and most of the Ax bugs will have been fixed. The ones that are deemed not worthy to bother with won't get fixed. This will leave holes in the numbering. 'Oh noze, teh men in blak R coming'.

What about the parts left out? This comes from a fundamental non-grasp of the term revision. The data sheets being hidden by the vast left/right/centrist conspiracy at AMD are, well, not really worth bothering with. Each revision of a CPU gets a data sheet, complete with errata lists and whatever else AMD thinks should be on there. The Athlon had data sheets, as did the Athlon XP, MP and so do the Opterons and Athlon 64s. These sheets are not the same for each processor, and when a CPU is fundamentally changed, AKA a revision, it gets a new data sheet. The old one tends not to pertain any longer and the document essentially gets forked.

If you have 90nm CPU1, and it has 10 errata, the sheets should reflect that. If you shrink it down to 65nm, fix four bugs, and still call it CPU1, it probably has s different data sheet, it is a different CPU. Say then you revise the core and add seven new bugs, two of which get fixes in the first few internal steppings, along with two older bugs. The result is called CPU2, is 65nm, and has a new data sheet.

CPU1/90 will have 10 errata listed, CPU1/65 will have six and dear sweet non-denominational higher being, will have, shudder, holes in the list! Call DHS, call the attorney general, they are out to get us!!!! To make matters worse, CPU2 has nine listed, several are new, and more holes. I sense an alien mothership about to land.

Let's make it more complex, say a year after CPU2 is released, there is an obscure bug found in CPU1/90. The datasheets for CPU1/90, and subsequent CPUs if applicable, get updated. The new one contains all the old bugs, and the new one, for a total of 11. The old data sheets for CPU1/90 is now wrong because it does not contain bug 11, so they are replaced with the new sheet. If they left the old one up, it would pose problems for designers, imagine if you were making a heart/lung machine that triggered that bug, serious no birds singing territory.

If AMD left the old ones up, a slip of the mouse, a click on the wrong line, or a drunken download-fest could cause them serious problems. Lawyers are like mosquitoes, horny, morally deficient mosquitoes, and will actually sue a large company over a user's stupidity. That is why they don't keep wrong info on line, it is ass-covering.

Calling Intel and AMD bad things over things like this is almost irresponsible. It shows a basic lack of understanding of the industry, it's common practices, and simple logic. The fact that these articles were spread so far and wide damages the industry and spreads FUD and misunderstanding among those trying to learn about the real problems. It is flat out sensationalism, with facts thrown in for questionable reasons. This type of logic belongs in a political 'debate', but not on a technical web site. µ