This is a slight variation on the intent of JmBoullier's question, but here goes ...
I'm intending to buy new hardware this year, and would like to see benchmarks of various computer setups crunching WUs under BOINC before deciding what to buy.
It would be useful if the results were collected, some analysis done, and the stats made easily available on WCG website (rather than having to trawl thru these fora). Collecting Whetstone & Dhrystone results (BOINC >> Advanced >> Run CPU benchmarks) would be one way of generating the stats, but it would be better if all contestants could run the same test suite comprising one "typical" WU from each project (repeated according to the no of CPU cores), as the relationship between Whetstones, Drhystones and real crunching speed is not clear to me at least.

At this point in time, 4-core CPUs are becoming available at reasonable prices, and it would be interesting to see how well their total speed scales up with the increased no of cores. The shrinking of CPU features from 65nm to 45nm is also happening, and I'd also like to see how this new generation rates.
(Currently 1 x AMD 4200+ X2, skt 939, 2321MHz: 2349 Whetstones, 4233 Drhystones per CPU, ~90 WCG points/CPU-hr)

Hi Rick!
I have the time that you miss for reading and collecting information when I need some and that's what I did when I thought of building a machine dedicated to WCG. The result of these thoughts is an Intel Q6600, it is running since Sunday evening and I am amazed. It is the same clock speed as my "normal" machine, a Pentium 4HT 2.4 GHz and despite my hearing and saying during my career that "clock speed was not all" I never had such a demonstration so close in front of me. It seems that each core of the new beast is about twice as fast as the Pentium.

I am really happy with my choice which also happens to be the best cost/performance ratio currently, including electricity costs (the Q6600 is rated 95 W according to Intel's specs).

All the above is for a plain install out of the box with the standard heat sink which is sold with it. Now if you also consider the number of reports mentioning an amazing overclocking capability up to at least 3.0 GHz without any risk or particular difficulty (and more for OC experts) I wonder what you could want more for the current price...

Cheers. Jean.

Additional notes:
1. I have no particular link with Intel, past or present...
2. This machine is the first one I built myself
3. For the record its latest benchmark is
Number of CPUs: 4
--- 2359 floating point MIPS (Whetstone) per CPU
--- 5309 integer MIPS (Dhrystone) per CPU
4. I have never tried overclocking yet but i certainly will...
5. Sorry retsof...

BOINC and intense projects like HPF2 are the best O/C proving grounds available, even better than prime95. If an overclocked machine will run this, it will run anything. Don't try to bite off too much. Increase parameters gradually and watch the chip and mobo temperatures. On the other hand, try a few prime95 first so that you don't build up a long record of error workunits.

One of my AMD machines with a newer chip sits happily overclocked about 20%. Another one refuses to run faster than stock speed. The overclocked machine runs cooler than the old stock computer. A third is a from-the-store Compaq with Compaq windows. I can't even get to the BIOS at all. Compaq Windows is not to be recommended either since it seems to interfere with e-mail.

AMD had a bit of a stumble at the beginning of the quad core line. Intel won the marketing battle by strapping together a couple of dual cores.
Competition breeds improvement on both sides. Hyperthreading ... R.I.P.

The new X2 is running and contributing workunits..



After I get back from a short trip, I'll play with the overclocking a bit. It's a black edition, so the CPU multiplier is unlocked.

You are still going to see a few. The machine may run a word processor, but BOINC may or may not work at that speed. It could still take a few iterations with BOINC crashing and backing off the O/C bit by bit until it is stable. It should usually die within a few minutes of starting a WU.

I would also suspend any that have some time on them, so that comp time isn't wasted.

The little DDD-T workunits still work the best as a test project, from what we have here. They aren't identical, but they are fairly consistent.