Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
![]() |
World Community Grid Forums
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 3
|
![]() |
Author |
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
See this Thread for the full discussion http://www.worldcommunitygrid.org/forums/wcg/...599_lastpage,yes#lastpost The discussion went a little off-topic there and so I'm creating this new thread.
About two weeks ago or so, I bought one of those kill-a-watt meters. Nearly all of my machines are on a single power strip, I put the meter in-between the strip and the wall to see what the power consumption was for the machines calculating WCG. On the meter, there is a setting for VA (volt-amps) that is instantaneous, the other settings only accumulated over time. Other meters might have different types of readings but I wanted something simple that calculated power (VA and Watts are similar and related) and it had to be instantaneous. Note the WCG average-points are also "instantaneous" too -- if you wait long enough. With all 11 of my machines on it, the meter read a peak of 1080 VA. The WCG average points value at that time was 69,190. So from those two numbers I calculated 64.0 points per VA. This is the baseline. It gives me the amount of electric power I'm using to get my machines to calculate a certain number of points. I built a new hex-core machine. I then permanently shut down 7 of the current machines (criteria was any one generating less than 5,000 average-points). And then I started up the new hex-core. It took the last two weeks for the average-points to creep up on the new machine, and it's still rising although very slowly. Now, I have only 5 machines in all that use up 830 VA and generate 79,513 average-points. This is 95.8 points per VA. The conclusion is clear: use more CPUs and fewer boxes. This spreads the power consumption overhead of the fan, power supply itself, memory, hard drive (I used SSD), etc to as many CPU cores as possible. The result is lower power consumption and higher average-points over all. Hope that helps, John PS Here's the output of my Ruby script: arrizza@john8:~/projects/boinc$ ruby getall.rb |
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Thanks arrizza
![]() Lawrence |
||
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
The conclusion is clear: use more CPUs and fewer boxes. This is correct up to a point. If we consider single socket machines, then yes no doubt. A six physical core CPU is better than 3 dual core machines or at the extreme much much better than 6 singl core machine. First because we have a technology gap and second because you have a lot of lost energy to power, disks, RAM, motherboard, PSU losses etc. etc. Now comes the issue single socket, dual socket. My past experience and it was long debated with Movieman and Dataman long ago, is that in terms of point/watt two single socket top machines would beat a dual socket machine. Today with new 10 core Xeons maybe this has to be checked again, but then we enter the point/US$ ratio and that is also a very important issue. ![]() ![]() |
||
|
|
![]() |