Anything below TJmax is fine if you don't mind your CPU's expected lifetime to be "on the other side" of the 3yr warranty period. (meaning 3.1yrs or some such)
The QA engineering that goes into the process development itself is all geared towards ensuring this much just from a
minimization of extenuating liabilities standpoint to Intel's books.
There is no single temperature threshold above which your cpu's lifetime suddenly diminishes. It is a continuous function that is dependent on an exponential of the operating temperature thanks to the physics that underlie the
Arrhenius equation.
As a rule of thumb, for every 10C higher your operating temps the expected lifetime of your CPU is reduced by 50% (think of half-life). Whatever your expected cpu lifetime is if operating at 50C, call it X years, you can expect that lifetime to be cut in half if you operate your cpu at 60C, so X/2 years, and again cut in half once more if you operate it at 70C, so X/4 years, etc.
That may sound dire but understand the lifetime is engineered into the IC from the "top-down" in terms of the thermal specs. Meaning your thermal spec was set for your chip with the desire to minimize the number of in-field fails that would occur under warranty.
So making the assumption that your CPU has an expected lifespan equal to (really we should assume greater than as Intel would not be silly enough to make the mean of the distribution equal to their warranty period and then have to deal with the entire left-hand side of the distribution failing under warranty) the standard warranty period (3yrs) when operating at TJmax is a reasonable assumption. Then for every 10C below TJmax you operate the chip you should double the expected lifespan.
If TJmax is 90C and you operate at 80C then a very reasonable lower-estimate of your CPU's expected lifespan would be 6yrs (2 x 3yrs warranty period). If you operate at 70C then 2x2x3yrs = 12yrs expected lifespan.
What is the basis for my arguing this? At TI we required our process technology to be developed so as to enable the minimum lifetime requirement of 10yrs operating at max spec'ed operating voltage and max spec'ed operating temps in continuous 24/7 operation. It is SOP for the industry.
Now where you can really cook your goose (cpu) is over-volting
and running hot. It doesn't take much to be operating your CPU in a voltage/temperature regime that in combination the two factors contribute to lowering the expected lifespan of your CPU to something <1yr.
Not too mention there is always a distribution to the lifespan and your particular chip could have some intrinsic weakness/flaw in it that puts its expected lifespan at a value below the mean of the distribution and by operating at elevated temps and volts it is destined to fail substantially sooner than the warranty period. (I killed my QX6700 in something like 18 months, never operated above TJmax or above Vccmax, but had lapped the IHS so no warranty replacement for me)
Facebook
Twitter