Independent testing labs
"Perhaps you could have an Intel colleague or two with the requisite knowledge design a "burn P4" program that would create maximal thermal stress. Then we could distribute it together with that die temperature thing, and see what people manage to turn up."
There's a big _disconnect_ between the extensive thermal and test coverage tests that companies like Intel, AMD, IBM, and Motorola do and the haphazard, onesy-twosy, "anecdotal" tests that some of the newly popular Web sites do.
I hate to keep going back to my Intel experiences, which ended in 1986, but in "my day" we had thermal imagers to map the die temperatures as a function of test vectors being applied, we had extensive measurements of chip and package temperatures, and we did exhaustive tests. (Paul Engel even designed a thermal test chip, circa 1975.)
I assume the chip vendors are still doing a lot of work on this. By all evidence, and anecdotal reports, they still are.
However, by the time the Athlons and P4s and whatnot get out, the reports turn in to "Rashad at his site just reported that the Athlon gets, like, rilly rilly hot when playing Quake! Anyone know about this problem?"
There used to be some outside testing labs. Mos-Aid, in Canada, did extensive characterizations of RAMs. Microprocessor Report used to sometimes do tests (besides their usually excellent architectural analyses). A testing lab in Arizona also did reports.
My point is that if you folks want to resolve reports you are hearing from "Rashad's Site" or "Tom's House of Processors," you'll need to start asking for actual measurements. These are not necessarily hard to do. Thermal imagers have come way down in price, and probably there are ways a clever experimentalist could modify consumer-grade cameras for simple imaging. (Imaging the device surface is harder. We used to use a Barnes Thermal Imager and we could of course look at chips exposed to the open.)
This costs money, even if "found technology" is used as much as possible. It costs money to buy several CPUs from each vendor being tested, it costs money to do the tests carefully, and it costs money to put reports together.
But it beats the hell out of anecdotal reports that a particular CPU must be having thermal problems because "Rashad's Hardware Site" felt a chip getting really hot when Photoshop was crunching some images.
And "is so! is not! is so!" battles between advocates of AMD vs. Intel don't resolve anything.
--Tim May |
