Run Memtest for even 30 minutes at high Vdimm and frequencies and put your hand on the DIMMS - they get very warm.
Of interest may be my tests and observations:
Using a MAX3 and Mushkin lvl2 PC3500 at 2.8V 2-5-3-2:
Max of 272 FSB @ 5:4 max (218 mhz memory spd) - memtest stable 2.5 Hrs. repeated test #5
-----
Using an IC7 as above - same exact max. memory speed.
-----
Using a P4C800-E Deluxe, same settings, except this board has a tick more Vdimm available at maximum setting (2.85v) which measures 2.87 with my VOM.
Max of 279 FSB @ 5:4 (223 memory spd) - 2.5 Hrs. of repeated test #5 Memtest
-----
Case temp is a constant 29c, CPU is H20 cooled, fully loaded 32c.
For all tests
-----
Adding a 80mm fan, with a fanmate adjusted to 2000 RPM (virtually silent) mounted right above the DIMMS, blowing down on them, with no other changes, yields:
Max of 285 FSB @ 5:4 (228 memory spd) - 2.5 Hrs. + of repeated test #5 memtest.
Also raised heat in room to reach a 32c case temp for the 2.5 hours of Memtest - still bulletproof at 285 FSB.
Kill the fan, and back down to 279 as the maximum error-free speed.
Crank the fan beyond 2000 RPM, and 287 FSB will fly.
Virtual silence is important to me, but I realize alot of people around here have a half-dozen fans blowing in and out of their PCs. If this is the case, you may not see as much improvement. If its not, and you don't care as much about a little noise, it may just make up to 10 Mhz FSB of an improvement.
My CPU starts to require more than 1.55vcore beyond 283, so the fan made all the difference in the world, and is inaudible from 18" away (case is in a rack, H20 cooling box is remotely mounted away from the rack, HDDs are the loudest thing in the box.)
If there is some kind of H20 cooled DIMM coolers I would like to know about it, but the fan approach will net repeatable, solid, stable gains. It allowed me to max my CPU without a Vdimm mod, and creates less heat / stress on my Lvl2 Mushkin at the same time. YMMV.
- Chris
