There's something called the "recommended maximum." It is the maximum volting of the modules that would still be covered under the limited life-time warranty.
There was an extensive discussion on another thread about CPU voltage -- "what was safe" -- issues about both heat and electron migration. A poster who insisted that the voltage tolerances were higher noted something with which I would agree -- that we don't have any technical information about Intel's design, heat and voltage tolerance in the form of lab-test statistics. [That's how I interpreted his last sentence, anyway.]
But it boils down to this. They provide 3-year warranty on the retail-box processors, and the memory-makers provide "limited life-time" warranty. If they specify a "recommended maximum voltage," it is to insure that that the probability of returns under warranty have a probability of 0% or near-0%.
This is the interface between marketing, business practice, common-sense finance, and the engineering aspects. Their legal and sales people would ask the engineering people how they would warranty these things, and what terms and conditions to use. The engineering people, quality-control and lab-testing folks would tell them -- "Don't warranty applications which volt the product above this level."
If you buy a power-supply and it has a 1-year-warranty, you will find it likely that you, personally, will experience a noticeable number of failures after that period. If the PSU is warrantied for 5 years, you're much less likely to experience those failures.
So while some people may take a cue from the spec provided, and over-volt memory by 0.1 or 0.2 volts, or over-volt the processor to well above 10% of the "maximum spec" or the "default [idle]" spec, they're making an expensive hobby even more expensive -- as a matter of probability. You might actually FIND some modules that handle those excessive voltages, or you might find a processor that will handle those higher voltages. But for a sample of modules or processors all set at excessive voltages, a noticeably larger number will fail over some short, specified time.
We've also seen that you can push memory to its max in Mhz, tighten the latencies way down to just short of "too tight" and generating load-test error, and volted to the recommended limit. That's really pushing it, because if they were spec'd for maximum voltage of 2.2 (the warranty limit), and speed of 1000 Mhz, and latencies of 5,5,5,15 -- you're stretching them even further. The company might replace them (whether or not they could detect whether they were over-volted or not), and you might be honest in saying that you never ran them outside the voltage spec. But you might have been safer to run them at a lower speed with tight latencies, or the higher speed with looser latencies, if you choose to run them at 2.2V.
In closing here, I can tell you that this is all about trade-offs. You can run memory at lower speed than the advertised/marketed speed rating, but you can also tighten the latencies more at lower speed. You will find that IN A LOWER SPEED RANGE, with really tight latencies, an increase in speed will run up against a threshold where you have to loosen the latencies -- if you had already increased the voltage to maintain the existing settings. And you can run the memories at their spec speed and spec latencies, but if you want to tighten the latencies there, you need to increase voltage.
Either way, there are FSB settings that are low or "lackluster" with tight latency values that require maybe only the median voltage. There are higher FSB settings that require either more voltage, or loosened timings. The higher you go, the looser the timings needed, or at those same new timings as you take FSB higher, you will also need more voltage.
Facebook
Twitter