Casual observations here -- I'm just beginning to test the OC limits of my E6600 and Striker mobo with Crucial Ballistix DDR2-1000 modules.
Voltage droop, as I understand it, is the difference between idle voltage and load voltage.
But there are some other things going on.
In another thread, I mentioned a forum post at "i4memory.com" -- a very methodical approach to testing the limits of the Striker board. The writer started at 1.525V to achieve a stable 3.6 Ghz setting of her E6600 processor at 1:1 and the stock multiplier of 9. She was using last year's Striker BIOS release, and the monitoring software showed a value of 1.42V.
I upgraded my BIOS, and at 1.475V ("set" value), the ASUS Probe monitoring software shows 1.44V. But there is no droop between idle and load -- the monitoring software shows the voltage as rock-steady and at that value of 1.44.
Other monitoring software coming from nVidia -- while posting real-time speed values and temperatures -- only reports the "set" value in BIOS (in this case, 1.475V).
Tests on the Striker and (I think) the P5N32-E board showed that the three PSU voltage rails are understated by the monitoring software, so for the 3.3V rail, the monitoring software may show 3.26, while application of a digital multitester shows something at, or just above -- 3.3V.
There are known discrepancies between the "set" values and the monitored readings on these -- and probably other boards. But this is different than "voltage droop," as I've explained here.
IF someone were to ask me, I'd say be careful about choosing your over-clock voltage limits, and see what data you can find in lab-test reviews so that you can calibrate your choices with the information about "true" voltages.
At least with the Striker and P5N32-E SLI boards, and assuming choice of high-performance memory, you can pretty much isolate the risk of over-clocking to the processor itself. For instance, if the Striker board specs allow for FSB at 800, 1066 and 1333, pushing the FSB to 1500 is only a 20% over-clock beyond the board's warranty limit, and less of a risk. Choosing 1333 is within the warranty limit. And similarly, if you buy DDR2-1000 modules and under-clock them to settings between 667 DDR and 800+ DDR, you're well within the memory-maker's warranty limits as long as you don't push the voltage beyond the specified limit -- for me, it's 2.2V. If the memory works at default settings at 1.85V, 2.05V is just high of being in the middle of the operational range.
That leaves the processor. Nobody knows yet how long before an E6600 would burn out at ANY settings over Intel's spec of 1.35V. But you KNOW they would choose a spec that minimizes their returns under warranty -- and therefore the cost of selling their product. They would apply a statistical distribution to determining what that limit should be -- for example, they would choose a warrantied range assuring that 99% of all E6600 processors would not fail when run in that range. When you go above that range, you are reducing the chances that your processor will last as long as the 99%, but your limit may correspond to 80% of all processors, or 70% etc etc.
Even so, with the small 65nm size of the cores, I'd say that wiggle room above some few hundreths of a volt above their warranty spec is smaller than it was for a 90nm processor.
An economics professor was making a joke during the 1980s about students focusing on econometrics (statistics), who were writing their own statistical software in the midst of the then new "microcomputer revolution." He joked that someday, there would be people sporting CPUs ornately mounted on stylish headbands.
I'm really not interested in spending three Franklins on a fashion statement, but the risk increases with over-volting.
Facebook
Twitter