Wbmw:
Get it through your rose covered glasses. AMD switched its definition after Intel pummeled them using a bad (to me) definition of TDP. The 939 document shows that the two used different definitions. Being old enough, I remember the original definition, Thermal Dissipation Power. I have documents defined that way.
Even the 3.51 version which TDP was changed to Thermal Design Power still used max current at typical voltages for all five voltage supplies (planes) at maximum die temperature.
That wasn't using "... commercially available software ..." as you claimed. That was Intel's definition in April of 2003 for TDP. AMD used "Total Thermal Output" in version 3.00 of the Opteron Datasheet, 23932.pdf. AMD's 40340.pdf version 2.04 had it "Maximum Thermal Dissipation Power."
But even in Version 3.51 of that document, family TDP was sum(Vmax * Imax) for all voltage sources. That was the number used by many Intel supporters as the number to use against Intel's looser at the time TDP maximum for some undefined suite of software programs at typical voltages and typical die temperature which could be exceeded. Of course impartial and AMD supporters correctly called foul for comparing apples and oranges given the quite different standards.
And they are still different. AMD maintains its TDP as an absolute upper bound of thermal output (and energy input) while Intel doesn't guarantee its TDP won't be exceeded, thus not an absolute upper bound at all, just one which is not exceeded most of the time.
As to your contention that TMA2 is not used to limit TDP within specification, " as long as the processor does not exceed Tj" it what allows it to go over Intel's TDP. I stated that as a required condition, that the HSF force the die temperature, Tj (Junction Temperature) to go beyond the trip point for any thermal output above the TDP specified. It does this by having a high enough thermal resistance to have TDP * X C/W > Tj - Ta (ambient) where X is the thermal resistance of the HSF. Also TMA2 has to be enabled for this to work, else nothing happens until Ttma is tripped. Not having TMA or TMA2 enabled is out of specification.
In real life, Ttma2 is higher than the specified Tj, so X must be larger to compensate for this margin between Tj and Ttma2 (about 105C IIRC). Ttma, the "drop dead, save the CPU" thermal trip occurs at about 110-115C, IIRC.
Your last few paragraphs prove my point that this TDP is never used. Even Intel uses a HSF much better than one that would do this. Tripping TMA2 runs the processor beyond the specified maximum operating junction temperature, so you have to push the CPU out of the SOA to get this effect. That is not what designers do. They take TDPmax and add additional margins just to be safe. Thus the HSF will keep the CPU cooler than the max Tj at thermal dissipations far above the TDP specified. Intel's own retail HSF does that on any CPU it sells.
If anyone actually used the TDP specified as the limit, you would get the dreaded "laptop burn", where many owners would sue. Which effectively eliminates TDP as a useful design specification.
And your characterization of me calling the Intel HSF, bad, is just poor reading comprehension on your part. I said that Intel's HSF was much better than their TDP would require. Thus they didn't use their TDP when it came to designing the HSF. They used a TDP much higher given the much lower thermal resistance. Its a "do as I say, not as I do" from Intel.
Pete |
