Finally, after countless and tens of tried cedars I managed to get some score

First tried some ocing of retail celery 347

D347 lapped, then IHS-removed and polished
Xtremelabs.org MAGNUM Legend Edition copper pot
DFI Lanparty UT P35-T2R
Some fast evil Nidec fans
OCZ 1GB DRAM
Radeon 7000 VE PCI
ENERMAX Revolution85+ 1250W

http://www.xdevs.com/images/ln2_cel/celrig.jpg

For temps and voltage control measurements I've used Fluke 87V, also with scope Tektronix 2246 for PWM tweaking.

Initial test on DFI P35 showed better results, than on usual P5B DLX, so i left DFI for rest of testing. On day before had tested celery and got 7.9G with lapped CPU.

To get few MHz more, decided to make some insane step forward - remove IHS from CPU. As we know, all latest Intel CPU's have soldered IHS to die with indium metal. it have +170°C melting point. IHS is quite thick, near 1.5-2mm and is bit a barrier for high-end LN2 cooling, where each degree is a need. Thanks to my cpu, it have no coldbugs even with all pot full with boiling LN2.

Removing IHS will render cpu die very sensetive to damage, and I had to manage extreme caution to put cooling flat and firmly. Die size is near 1x1 cm, not like huge i7 cores .

But no win for those who afraid, lets forget any risks and push silicon
Cutted resin glue between IHS and cpu package, and started heating to +170°C with thermal-controller heating gun. Minute spent.....two....nothing happens, lid don't go....added temp to 180...nothing.....190....nothing still....at +200 CPU jumps at few cm and drops it's self on floor. Whoa?

First thought - we have another dead bug now. IHS don't go still .
Heating again, this time fixed CPU firmly. After few secs of heating IHS drop off.



Then cleaned up rest of indium solder on CPU with sharp blade.



One more issue also arrive, head-less CPU lost it's ability to be cooled with stock coolers, because retention bracked is now higher that cpu die, and cooler will just lay on socket, w/o contact to CPU.

So I opened socket and remove bracket.



Not usual to see working system with open socket, lol

Then time to prepared DFI mobo to go.

Before mod tried different capacitor under subzero temp, to see which work on 77K of LN2 temp, which not. Panasonic and Sanyo OSCON SEPC work good, only dropping 10-15% of it's capacity. As many of overclockes know, Volterra VRM's don't have electrolytic caps, coz of fast switch clocks. But on extreme when every uF help to keep stability, so I've added five extra caps.

http://www.xdevs.com/images/ln2_cel/dfimod1.jpg
http://www.xdevs.com/images/ln2_cel/dfimod2.jpg

After system assy spend few hours to get good settings for DFI. Those who ever oced with DFI mobos know, that there are a lot of magic bios settings, which slightly affect stability. Voltage on CPU was ramped up to maximum vid 1.6000V and offset set to +130%. This is maximum, allowed by mobo. This will result 2.12-2.15V by DMM reading on CPU caps. During boot there was quite loud swirling and whizzling sounds from DFI volterra VRM. Under this music I've started to push F7 more that human hand can survive On most interesting moments I've putted my nikon on tripod near rig to make photos every second, not to lost WR moment

As for temps, booted straight from -170°C and under windows dropped temp to lowest -186..-190° possible. more than 2kg MAGNUM pot helped much, temps was stable and I had to put LN2 not continuosly.



http://www.xdevs.com/images/ln2_cel/celln3.jpg

Have validations on 8203MHz, but CPU-Z validator don't accept them, saying File ok, but corrupted .

http://www.xdevs.com/images/ln2_cel/cel8203.jpg

Have photos of 8237MHz, with halted CPU

http://www.xdevs.com/images/ln2_cel/celhalt.jpg

Good valid still a WR



In final system started to non stable boot, sometimes loosing memory, and reporting C1 on POST led. So I stopped action and disassembled rig.

http://www.xdevs.com/images/ln2_cel/celncd.jpg

Not needed parts of socket

http://www.xdevs.com/images/ln2_cel/celncf.jpg

Thanks to whole Xtremelabs.org team for support.
Thanks for T0lsty for LN2 supply and help with 8800GTX benchies other day.

Russian-language report about WR here