Date: Mon, 29 Sep 1997 14:20:20 -0500 (CDT) From: Tony Kimball <Anthony.Kimball@East.Sun.COM> To: tom@sdf.com Cc: michaelv@MindBender.serv.net, freebsd-chat@freefall.freebsd.org Subject: Re: supermicro p6sns/p6sas Message-ID: <199709291920.OAA25886@compound.east.sun.com> References: <199709291814.NAA25663@compound.east.sun.com> <Pine.BSF.3.95q.970929112604.4432A-100000@misery.sdf.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Quoth Tom on Mon, 29 September: : : The Pentium division bug was fixed long ago. And Intel gave free : replacements to everyone. Yes. Similarly, by all reports, the K6 bug is fixed, and free replacements are available. I have to wonder whether Intel would have offered the free replacements if the division bug had not been so well publicized, but can only speculate and/or compare past vendor behaviour. : What? The "make world" problems were VERY serious. Simple operations : in gcc were being preformed incorrectly sometimes, causing core dumps. : Such failures appeared in all kinds of other software as well. Well, not all kinds. I understand that the various flavors of Windows are not known to demonstrate the bug. Seriousness in real-world terms means loss of life/limb/property. Wasted time is one form of partial loss of life, and certainly having to type 'make world' again is a waste of time, but a floating-point error in an embedded system could crash your airliner or slam your missile into a hospital. Again, relatively weighting the seriousness of the bugs in practice, I'd have to say that the major losses incurred in each case were those of the manufacturer. Certainly Intel lost more money on the division bug, but then they made more on the sales in the first place. The whole issue seems pretty subjective/hypothetical: No actual airliners ever used a pentium in a critical component to my knowledge, or if they did (hah!), they didn't crash because of it, thankfully. Of course an integer error is not intrinsically less likely to cause a control failure than a floating-point error, in favor of your point. I guess the issue is kind of like benchmarking -- unless you benchmark performance against your specific application, results are unreliable; similarly, correctness. Since are larger proportion of the Pentium user base (DOS/Windows users) were affected by the division bug, one might reasonably chose to argue on those grounds that the division bug was more serious.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709291920.OAA25886>