dank@eng.umd.edu (Daniel R. Kuespert) (10/06/90)
In article <48351@bigtex.cactus.org> james@bigtex.cactus.org (James Van Artsdalen) writes: >In <tim.655078682@ziggurat>, tim@ziggurat.gg.caltech.edu (Tim Kay) wrote: > >] this guy wants to offer a 386sx 16Mhz with a fast crystal installed >] to yield 20Mhz (he says Norton SI is 21). Is this a safe thing to do? > >| It is not safe. While you might get lucky [...] > >> This is incorrect. Typically, there is absolutely no difference between >> the fast and slow parts that a manufacturer ships, *except* that the faster >> parts happen to work at the faster speed. > >No, it was correct. The word "typically" was a good choice. While the dangers of pushing chips beyond their design speeds seems to be well known, I've never been sure exactly _why_ they fail. My original post (the "While you might get lucky..." above) focused on overheating. I was always told that minor flaws in the conduction channels provide a bit of extra resistance; these flaws were negligible for lower speeds but caused additional Joule heating, chip damage, and eventual failure at higher speeds. Is this the only reason for chip failure? One would also think that signal dispersion would play a role---when you start putting the pulses closer together, at some point the distinction between different peaks is lost. I'd imagine this could give you really weird and intermittent errors if you were operating slightly beyond the chip's limits. -- Daniel R. Kuespert | Disclaimer: Chemical Process Systems Lab | University of Maryland, College Park | "I lied; it's a vice I have." dank@eng.umd.edu |
james@bigtex.cactus.org (James Van Artsdalen) (10/06/90)
In <1990Oct5.171736.13039@eng.umd.edu>, dank@eng.umd.edu (Daniel R. Kuespert) wrote: > While the dangers of pushing chips beyond their design speeds seems to > be well known, I've never been sure exactly _why_ they fail. My > original post (the "While you might get lucky..." above) focused on > overheating. [...] Heating slows down the signal paths in a chip because the resistance is increased. For PCs, the most important factor is likely to be that the signals simply take X nanoseconds to traverse the logic in the chip and to be driven on the pins. It doesn't matter how fast the clock goes, it still takes X ns to do the work. Speeding up the clock makes it possible to use a smaller X better, but it doesn't change X. As a chip heats up, X usually becomes longer, which is why a 386 with a "fast" crystal might work for a little bit and then fail hard. I don't know if the most likely failures will be internal signal paths, or external signal timing. Probably, "depends"... -- James R. Van Artsdalen james@bigtex.cactus.org "Live Free or Die" Dell Computer Co 9505 Arboretum Blvd Austin TX 78759 512-338-8789