On Thursday 17 February 2005 19:09, Paul Leyland wrote: > On Thu, 2005-02-17 at 18:33, Jesse Scott wrote: > > Since the thermal energy of a one core CPU running P95 is so great how > > will a system having a dual core CPU react? Will intel still have the cpu > > down throttle? > > I suspect you are making possibly unwarrented assumptions there. > > First, you are assuming that each core in the forthcoming processors > will produce at least half as much as the single core in present chips. > If each produces less than 50%, and other things being equal (which they > may or not be) then the future chips will not take so much power.
Sure. But Intel do appear to be having problems - the last few fab types have all ended up being shipped with higher power consumptions than the originally released engineering spec, despite reductions in operating voltage, mask resolution etc. etc. Probably the easy way to reduce power consumption is to remove or at least reduce the L2 cache (which accounts for most of the transistors & can't easily have the "inactive parts" drop to standby mode) but that would really hammer the performance, and is therefore ruled out by marketing considerations. > > Second, you are assuming that the cooling arrangements are no better > than those specified for current chips. It's possible, though possibly > too expensive, to dissipate much more power from a cpu than is commonly > done in mass-market machines. The supercomputer manufacturers of yore > were specialists in plumbing and air conditioning. Now it's probably > unnecessary to go to the extremes that CDC, Cray and IBM used to go to, > but there is a great deal more can be done in the cooling area than is > presently done. Ten years ago there were people building PC systems with a refrigerator in the case. Worked fine, though you needed to be deaf, or really dedicated, to put up with the mechanical noise from the compressor etc. The problem with P4s is the sheer density of the power output - well over one watt per square millimetre of chip surface even if you cool both sides (which mechanical design of the socket makes hard to do properly). This is about 100 times the energy output received by a black surface at right angles to solar radiation in an equatorial desert. With due respect, Cray & Co were dealing with lower energy densities than that (though the total waste energy volume was obviously massive, with thousands of processor units in parallel). In any case, no-one has yet repealed the Second Law of Thermodynamics, so active coolers, being (usually much) less than 100% efficient, will generate extra waste heat in themselves, resulting in more heat to shift out of the system box even if the problem areas are kept to more manageable temperatures. Intel's BTX standard goes some way to fixing the problem with P4 thermal dissipation, though it does not make for slimline PC cases - a BTX processor heat sink is the size and weight of a housebrick. With the exception of hardcore gamers, few if any PC users need more than a few percent of the processor power of current systems. It's as though mini-size cars were being sold fitted with Mack truck engines. Unnecessary and probably bad for the planet, even if a few selfish power junkies like us are grateful. The really odd thing is that Transmeta seem to be exiting the processor market - just at the time when the arguments seem to be turning their way! Regards Brian Beesley _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
