After a short time-out I decided to return to the last stable frequency of 475MHz. All this time the chipset North Bridge voltage was set at the maximum possible 1.51V. With this particular voltage setting the board was running stably for the longest period of time at 425MHz and passed the stability check at 475MHz. All my experience with DFI LANPARTY UT NF680i LT SLI-T2R so far showed that the higher was the NBcore setting, the better, bust starting from 1.54V the board would shut down immediately.
I decided to check out if the NBcore can be reduced, because the temperature was still too high despite the additional air-cooling. To my surprise, at 475MHz the board only failed OCCT stability check once, when I dropped NBcore to 1.32V. But even in this case it could at least boost the Windows OS. So, it turned out that high NBcore wasn’t helping overclocking: it was hindering it! After that it was all a matter of experience: the mainboard did manage to overclock the CPU to 490MHz FSB with the NBcore raised to just 1.39V.
Nvidia chipsets are known to be more flexible when working with the memory than Intel chipsets. During our test session the memory was working in synchronous mode with higher timing settings. We failed to set Command Rate to 1T, but S&M utility showed that the memory can work just fine at 980MHz with 4-4-4-12 timings. I have never managed to achieve this result on any other Intel based mainboard before. With the memory working with 4-4-4-12 timings at 980MHz and the CPU overclocked to 490MHz FSB the system passed a 30-minute OCCT stability test and over a dozen of performance tests in different applications.
When we almost finished our exhausting overclocking experiments and I was about to get to the next part of our test session, one more unexpected thing happened. We got to testing in 7-zip utility, which refused to work. It reported buffer overflow, decoding error or just froze without reason.
At first I thought that memory subsystem was the one to blame, but neither higher memory timings nor higher memory voltage settings helped. NBcore increase also had no effect as well as reinstallation of the application from scratch. In the nominal work mode, however, all tests were passed without any issues. The solution to the problem turned out the CPU VTT Voltage parameter that needed to be increased quite significantly. Z-zip started working fine as soon as we pushed this voltage setting from the nominal 1.21V to 1.34V, but maximum stability was achieved only at 1.44V CPU VTT Voltage.
Very interesting effect. I usually use SuperPi test (4-8- million decimal digits) or 15-minute OCCT test for the preliminary stability check of the overclocked processor. If I manage to confirm the efficiency of 7-zip utility on other chipsets as well, I might switch to it eventually, because it reports the first results in 15 second, or doesn’t report anything at all.