Buggy PENTIUM

Before I demand a recall of my pentium chip, does anyone know if it will interfer with the calculations in NONMEM?

I checked the output from my validation runs after installing NONMEM on a Pentium-based PC. All point estimates were exact to the 3 digits printed; estimates of standard errors were occasionally off in the third digit, but that is certainly not unexpected. I have a call in to the Intel folks and have been waiting for a week now for a response. Another Pentium owner in our lab has spoken with an Intel technician and has informed me that they are sending a replacement chip. Evidently the intensive calculations involved in nonlinear mixed effects modeling met Intel's requirements for replacing the chip.

Further discussion on pentii - a diagnostic that some of you may have seen: Perform the following calculation : (4195835/3145727) X 3145727 The answer should be 4195835. A defective machine will answer 4195579.

The Intel policy is "customer satisfaction", meaning any customer who is unsatisfied with their Pentium chip can get a replacement in a box. Who takes the old Pentium off the motherboard and puts the new one on is between you and your computer supplier, and not necessarily trivial: a fair number of motherboards get ruined by snagging another component while taking the Pentium out. I'm not entirely clear on who is responsible for supplying heat sinks, which cannot be removed easily once glued to the chip. So anyone who uses a word processor on a Pentium can get a replacement chip, and getting a replacement chip does not mean Intel admits that the division bug adversely affects word processors, or any other software. Intel has contact people scattered throughout the countries, but since there aren't more than 40-100 NONMEMers on this list, you might contact: Joe Brandenberg (503) 629 7701 joeb@ssd.intel.com If you have any curiosity about numerical analysis matters at all (such as, how could division, which we learned in grade school, have a bug in it? What software applications have been tested and what is their MTBF due to the division bug? Are any algorithms other than division affected?) then I would suggest you ask for the Intel white paper on the bug first. NONMEM has not been tested yet. If you want a bugless chip, Joe can handle that too - tell him and he'll mail you a Pentium in a cardboard box.