Event preperation

Based on the experiences gathered from previous overclocking events, I am absolutely confident that preparation is a vital part of a competition. Not only to have hands-on experience with the hardware, so that you know how everything behaves and what setting nets you better performance, but also because you have an indication of what scores are good and which aren't. As for this event, no competitors received any of the used hardware at home to prepare. Lucky for us, Massman still had his Gigabyte P55-UD6 mainboard lying around, the one used for the P55 release article here on Madshrimps. We were even more lucky, as he also had a Core i5 750 to spare, which gave us the best possible preparation combo possible.

Leeghoofd was in charge of preparing the PI session (if you are a power user, first rule of thumb is to obey the orders from the big boys, though the latter took some advantage of me for sure)

Things I found out: Running the CPU with only 2 cores was always slightly faster then with 3 or 4 cores. Differences weren’t huge, but a gain is a gain in such events. I tried 2 different OS : XP64 SP2, which is my daily OS, for initial testing. And XP SP3 to get some sort of reference time.

Since we didn’t have access to the Kingston rams to test them I used my OCZ Blades, though at conservative timings. I stuck with C8-8-8-22 TRFC 82 timings during the testing for 2000Mhz ram speeds. C9-9-9-24 was used for 2100Mhz. If you are smart (and all of our readers are) you noticed that I only tested 2 BCLK settings, namely 200 and 210MHz. This due to time constraints and to not waste someone else’s hardware (although this UD6 is the one with the infamous burned socket).

Without any major OS tweaks (just large system cache enabled and just a Copy waza via SPI tweaker) Killing explorer and set Real time affinity to SuperPI : I managed to set a time of 9 mins 9sec fairly easy.

Massman told me sub 8 mins 55sec was the target, I don’t think with my "let’s call it conservative ram and subsettings", I was ever going to breach 9 mins 5. He apparently pulled my leg (once again) as he reached 9 mins 5 with some more effort (aka tweaks).


Thomas aka Pt1t teamed up with BobbleMagnifique to represent France, here they are trying to figure out how to insulate the board

If ever there was going to be a tie; the Gigabyte referees would ask us to do an Everest Cache/bandwidth benchmark.

Enabling the C-states in the bios gives a big boost here. Yet contradictory to my findings, enabling them did nothing for Superpi nor 3DMark01, in fact it was most of the times even slower.

As for the 3DMark we mostly relied on what Massman had already figured out during one of his P55-3DMark01 benchmark sessions. These findings have been posted and described on different overclocking forums. In short, the key to a good 3DMark01 is:

Use the right LOD values

High uncore

Use correct run order

Just a couple of minutes before we left for the event, we did a quick 3DMark01 run and set our goal to 80k tot 81k.


For the Oriental fans, Massman impersonates Tom Cruise again