Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: Elmer who wrote (25797)	11/11/1997 12:54:00 AM
From: Petz	Read Replies (1) \| Respond to of 1579715

Elmer, you are right, the FP errata has not been corrected yet, I must have typed the URL incorrectly, which is why it said "document not found on server." As you know, SI doesn't hotlink a URL unless it is preceded by http://. But please read this: "The erratum has not been observed to occur in integer, single precision, double precision or software implementations of extended precision applications, which represent the majority of applications on the market. AMD has identified and tested commercially available applications which specifically make use of the numeric processor in the K6. No failures have been observed. AMD has tried to force failures by choosing conditions and operands known to aggravate the erratum, and no failures were observed." The Intel FPU bug, OTOH, could be demonstrated in 1-2-3 or Excel and had at least a 1 in 224 chance of occuring with random divide calculations. The AMD erratum only occurs in an extended precision floating point mode unique to the K6, which is why AMD had to test "commercially available software which specifically makes use of the numeric processor of the K6.*" The reason they couldn't force the bug to occur in any application is that the un-normallized floating point numbers are first normallized if typed in to the program, and any numbers which are the result of calculations within a program are also normallized before being stored. Therefore it is impossible to get the bug to occur other than by defining a floating point number as a hexadecimal constant in an assembly language program. Another section of the writeup says that if the erratum occurs an interrupt is always generated. Therefore AMD can and will write software to detect the occurence of the erratum for the truly paranoid. How they ever found this bug is beyond me and I'm amazed that they'd even consider, let alone promise* fixing it. The K6 still has a grand total of six errata, and three of these are hardware related and the other two can only occur as the result of a bug. OTOH, on the Pentium II, Intel has 18 errata they refuse to ever fix, 11 that they claim they will fix but haven't yet, 3 they actually fixed and 9 errata they eliminated by respecifying the processor. Of course this doesn't count the disastrous bug that totally halts ANY program under ANY operating system causing total system failure with just a single instuction! Petz