Superpi 1M

First up was to explore the new MSI Fuzion board. This board has got the Lucid Hydra chip onboard. Allowing to mix card of different brands (ATI and NVIDA cards mixed) and still get a performance boost similar to SLI and Crossfire technology.

Time to explore Bclock. CPU was kept around 10°C, multi lowered for the CPU and rams. Straight to 220Bclock, but no post, this is going to be heavy if the CPU or board is Bclock limited.

After fiddling some in the bios the board needed some reboots to correctly apply all the correct tables before correctly posting. So one altertion at a time was required. We easily went up to 230Bclock, booting straight into Windows 7. But stability wasn't there. Trying to up PCIE clock didn't sort out the instability issues, more CPU volts, VTT, nope didn't help us to get further either. PCH voltage to 1.25 did the trick. 235 was finally bench stable.

Bart had the brilliant idea to ask for some cheapo GPU's to be used during the Pi runs, as everyone was modding their GTX 480 cards to be put under LN2. Luckily MSI repair service had loads of PCIE cards and network cards ( internal joke lads, no further comments) available for us. They also fetched the LN2 all the time for us ( no Dewars allowed inside the building ) thanks for that lads !



From the Gigabyte Tweaking event we knew that 2 cores were slightly faster than one, 3 or 4. Sadly our fuzion board wasn't so keen on freeing the multies ( and there's no 3 core option at this moment). The 24 Multi ( 2 cores ) also seemed very unstable on this board, so we opted to boot with 4 cores and 21 multi. More cold (-55°C) brought us up to 245 sort of Pi 1m stable, but once we tried to go colder (-60°C) the rig became unstable. Massman explained again to me ( he's both my mentor and godfather ) the issues with Lynnfield and LN2. Quick sum up :

Lynnfield + extreme cold = coldbootbug, coldbug (typical for Intel CPU's) but also a coldinstabilitybug

This didn't seem to good at start, but once we heard Thomas was having issues already at -30°C, our CPU wasn't that bad at all. Same ordeal for Pascal and Walter, thinking at start more in the direction of corrupt Hard drives, bad RAM. But in fact it was the cold that made their rig not even able to boot into Windows.

A quick look at the Hwbot 2D rankings , showed us that our sample was pretty good, topping around 5180mhz. So we got a good feeling again about our setup. So the quest for the lowest Piscore began.

An 8 second + “something” score was directly achieved without too much trouble. Messing with rams and co didn't bring much improvement, it was pure CPU power that would take home the victory. Knowing Massman and his philosophy behind Superpi 1Mb : If it freezes at 10 loops, insist on it, it will pass sometime. And yep he was right (again) as we got an almost 8 second flat. Trying to boot at high Bclock ( read over 240 ) usually resulted in a BSOD, reboot when booting into windows. 230-235 Bclock was used to boot into the OS, then upping slowly via the OC Dashboard in windows7. Lot's of trial and error finally gave us this 247Block run and Sub 8 secs Pi run. 5th score on the bot with only single channel and on Windows 7... There’s a bit more to gain on XP and good rams...



After having loads of issues with his USB keyboard, Viss and Pete managed to get 2nd spot for Pi with 8.348 secs. Pascal and Walter 3rd spot with 8secs 486 and Thomas with his Ûber crap CPU ( running at a mere 4.8Ghz ) at 8.58secs.