My contract for cheap electricity will be ending soon
I decided to cut down on power usage
out of 11 boxes running 7 of them are single cores averaging 120W a piece
the worst offender was my first Boinc cruncher an Athlon XP 2800+ drawing 170W full load

so I'm shutting down the single core boxes
but because I'm a credit whore I needed something to replace them

with Intel quad core 6600 being so cheap I had to give it a try

tore a part the Athlon 2800+ water cooled rig
threw in Q6600 on a ASUS P5K-VW mobo with 2gigs of Crucial Ballstix PC6400 memory
after some research about running temps and seeing how the stock heatsink sat on the chip I decided to use the water kit too

now I'm getting some base line data for stock setup before I start overclocking

with all four cores running
using only 135W
with CPU temp @ 55C

I'm letting run stock for a few days then I'll have to play

I can replace 7 single cores using 800+ watts with 2 quad cores using less than 300 watts and still be a credit whore

Kickin on yoyo@home
Frank Encruncher

I finally got around to OCing my new box: an E6750 on a GA-P35-DS3R. I'm currently testing at 3.2ghz up from the stock 2.67ghz. 40 minutes into Prime95 we are still stable as expected. Reviews I've read indicate an expected load increase of 7 watts idling and 25 watts under load

Been playing with this stuff for awhile now with dissapointing results. I can easily bump my 2.67ghz up to 3.2 ghz, and I do see a corresponding improvement in 3DMark scores. However, my BOINC wu's slow down¿ For example, I do Einstein@home in about 41k to 43k seconds (333mhx * 8 ), but when overclocked (400mhz * 8 ), I jump to 68k seconds! My guess is that a 333.33mhz bus speed moves math data more efficiently than a 400mhz bus speed, but I can't find any information on BOINC boards or from Dr. Google. The highest multiplier I can get is 8x, so I'm tempted to try 666mhz * 5 for a 3.33ghz speed.

Any thoughts?

OK, I decided to go back to square one and redo everything in a controlled manner. I am concluding that my mobo and chip get along very well at optimized default settings and it will be difficult to beat without a serious cost in voltage, heat, and likely lifespan.

I noticed while running cpu-z that my core speed shows 2000 mhz at stock (6 * 333). I know I can make this run at 2667 mhz by disabling C1E and EIST (Speedstep). I decided to start off with this config. I did this and established a baseline for processing an Einstein wu and for 3DMark. I was careful to ensure the some processes were running for every test. I then started BOINC back up to 100% and did temperature readings with speedfan, upped by 15 degrees c per some unfindable post.

OC Setting - stock with C1E and EIST disabled:
3DMark - 4155
CPU time per tick - 68 secs
Core Temp - 43c

Then I re-enabled C1E and EIST and ran the same tests.

OC Setting - Default settings:
3DMark - 4886
CPU time per tick - 41 secs
Core Temp - 40c

So by disabling these two settings I get a 15% degrade in 3dMark performance, a 66% degrade in Einstein performance, and an 8% increase in heat output. WOW! So after reading up on this stuff it sounds like it's a pro-con tradeoff game. Enabling them may restrict you from setting other voltages and may cause instability at lower speeds Hmmm ... Sounds like another test.

OC Setting - 8 x 400 with C1E & EIST enabled
3DMark - 5083
CPU time per tick - 34.67 secs
Core Temp - 55c

Running at 3200mhz with speedstepping gives a 15% Einstein boost, a 4% 3DMark boost, but costs an extra 10 degrees c or a 38% hit, and I have a decent 3rd party heatsink. I'm very leary of running it this much hotter for only this much performance boost. For the odd game it might not be bad, but to run it 24/7 for science when accuracy is more important, I don't think so.

I'm going to work at it more. I've done several other tests with consistently bad results. I'd like to understand everything. I need to explore memory timings and other voltage boost impacts, but I seriously doubt I'll be overclocking my puppy