Nice, but...if you use a proper Xeon, you have a 100W...but also crunching time is less then 1/5th of the Atom...

So that might be efficient...but only in energy consumption! ;)

Since it has a 24 pin ATX PSU connector you should be able to run 2 off 1 PSU. More than 2 maybe. I used to run dual caseless Asus and Gigabyte mobos off 1 PSU. Never had any problems on Intel 2600k CPUs. Could not get it to work on a 3770k with an MSI mobo. Never tried to figure out if it was the mobo or the CPU that kept it from working. All you need is a 24 pin splitter, a PSU with dual 4/8 pin ATX connectors (or a 4/8 pin splitter if it doesn't) if the mobo even uses them and start switches for each mobo. Even though all the fans start when starting 1 mobo you need individual start switches for each to start both.

Nanoprobe, it's interesting that your newer processors didn't work. Maybe they have some built-in controller function the older ones don't have, and as a consequence didn't work?

Do the start switches you mentioned attach to the 24-pin connectors? I've searched online and have only found this.


Oldchap, enels recently wrote that his Z3740 has a 1050 RAC when crunching UGMs with Windows. That would mean 262.5/core and with your example of 10 boards, each having an 8-core processor, it would come to 21000 points. But the C2750 has a 2,4GHz base frequency while the Z3740 only has 1,33GHz, so the question is how many more points this would result in? Assuming a linear correlation between the frequency difference and the RAC, it would mean 37900 points. An 80% increase, but that doesn't take into account the memory bandwidth, where the C2750 only has a 50% increase. So let's assume the C2750 core has a 65% better RAC than the Z3740. This would mean the 10-mobo combo would have a 34600 RAC.

My dual X5650 computer has a 9000 RAC and needs about 290W, which means 31,0 points per Watt. Even if the ten C2750 motherboards need 400W, which I don't think they need, the C2750 setup would have 86,5 points per Watt. The E5-2670v2 computer you linked to would probably need around 290W and would generate 78.7 points per Watt. Enels' tablet needs 5w from the wall so that means its 1050 RAC translates to 210 points per Watt - an incredible number, but true.

It would seem that the C2750 setup is the most power efficient of these three examples, not counting the tablett. But not by much, 10%. What's bothering me is: what's wrong with my computer?

That dual 10-core computer's 22828 RAC is MUCH higher than mine. I don't understand how the difference can be so huge. Can an E5-2670v2 really be 150% more effective than an X5650? I realize it has 10 cores instead of 6, but still. Something's off. Is it because X5650 doesn't have AVX ISA?

They attach to the motherboard just as if you had the boards mounted inside a case. I used these and they worked fine.

@ Nanoprobe

Thanks for the link. Now I understand.


@enels

I never read that comment. And there is also the reason why the C2750 is not performing as well as I thought it would.

BTW, are your points WCG or BOINC?


@ Oldchap

"Starting with RAC. A team mate of mine rented a c2750 with 8Gig and an ssd. The numbers I quoted are averages of a week crunching MCM1 and a week of FAHV."

That settles it. I thought you guesstimated the numbers.


So, what is the conclusion? If badge hunting is our game, we should go for many C2750s. If it's collecting points then we should go for Xeons. And science? Who cares!

All this is somewhat confusing. I think that work units is all that matters and should therefore be the only standard for measurement. That also means projects shouldn't be compared against each other. No permanent points of any kind. Badges should be given for achieving work unit milestones. WCG, and most other projects, seem to reward everything - except - work units. If we crunch one work unit then we should receive one temporary wu-point. Not be given badges for time spent crunching or complicate things with relative points based on what kind of CPUs we have. After we receive, let's say, a bronze badge, the temporary points should be reset back to zero.

Here is a discussion about how meaningless these points are. I think the whole point system ought to be scrapped. Now. The best system I've seen so far is GPUGrid's, were crunchers receive badges based on the percentile they contributed to a published science paper. I would like to see that on WCG, but with number of work units as a metric.

This, of course, will never happen.



OT EDIT:

I've had problems with high CPU temperatures since I bought my computers 6 weeks ago. They've been hovering around 75C and I didn't know what to do. For a while I considered changing the cooling paste and the fans. Now I discovered, almost by accident, that by lowering the processors maximum speed, from 99% to 79%, the temperatures dropped all the way down to 59C - AND - the power usage went down from 275-295W to 225W.

I know RAC will probably fall, but I'll have to accept that.


OT EDIT 2:

This article is interesting. They managed to substantially lower the computers energy consumption, with only a moderate drop in performance during full load. I don't know if the undervolting is possible with a Xeon, but I'll probably replace the HDD with a USB stick.

Workshop: making your PC as energy efficient as possible