Help with DPAD running on Xeon server

[compudog]

I posted earlier about acquiring three Xeon 2U servers. I have had one up and running for some time now. This thing has two 3.0 GHz Hyper threaded CPUs with 4GB of DDR RAM. The only thing running on it is DPAD. When I open taskmgr (running Windows Server 2003 Standard) it shows 4 CPUs running at 100%. I ran manualsend.bat this morning and sent in 44 results (this is after having run for 36 hours or so.) I tried again this evening and it had less than 10 results completed. Am I wasting electricity (other than being a space heater) or do I need to configure something differently? Processor Affinity is all four "CPUs" checked, and I just switched Priority for DPAD to "Above Normal" from "Low." Any ideas or would this machine be better suited to a different project. I will continue to run it at 100% as long as the weather here is cool/cold.

 

[waffleironhead]

FWIW the runs being done by dpad at the moment take forever on systems of that age. So after a few days and only that many results is not uncommon. A real idea would be the points that were accumulated by those results.
As a side noteI stopped running dpad awhile ago due to the long turn around time on my p4 3.0 prescott. The output was being overshadowed by current quad cores so by the time I had any results to turn in they were quite outdated. In any case, the science is still valid and appreciated by stephen i'm sure.

You could try running muonbench and compare the output you get to the charts in the moun1 forums. I figure your xeons should be comparable to my prescott give or take a bit. It should still be in the chart there unless stephen updated it and weeded out the low end processors.

 

[compudog]

Thanks waffleironhead, I am getting about 50 results every 2.5 days on a 1.8 Ghz Opteron dual core. I figured I'd get more out of two 3.0GHz hyper-threaded Xeons. If this is the average production, I think I may just save a little electricity...

 

[waffleironhead]

dpad has always been quite a bit faster on AMD systems, at least the K8 to P4 days. So your output is not too out of line.

 

[petrusbroder]

I have a Celeron @ 1.4GHz and it turns in approx 50 results every 9 - 10 days, 44 results would need 7.9 days. You have a two processors, running approximately double the speed, so that would mean 44 results in 1.9 days = you crunch better ... may be the effect of hyperthreading or the fact that I am using the comp approx 4 hours a day for other stuff such as writing, surfing, calculations ...

I do not know if hyperthreading really helps: IIRC you crunch twice the amount in twice the time ...



... my 2 cents...

 

[compudog]

At the rate of production on DPAD, I can't justify running the server 24/7. It's far too noisy, makes too much heat. I just thought, WOW! Two HT XEON CPUS with 4 GB of DDR RAM, this should crunch nicely. So far it hasn't cost me much money so it was a good experiment in setting up a server/installing an OS and configuring the RAID 5.

 

[JonB]

If your results.dat file is small, then you will run a LOT of 5x verification runs. Keep in mind that the number of results turned in is NOT the same as the amount of Mpts you get credit for.

Two things I'd try:

1. You can let it build up its own results.dat file and it will quickly reduce the number of verification runs. Mpts will still be the same but result numbers will improve.

2. You can copy a large results.dat file from somewhere else.


I've been known to just delete my results.dat file occasionally because then that machine will start trying more new and different seed values and, though it produces fewer rresults in a given time period, it will produce some new results that get into the Top 10 best list.

 

[compudog]

I'll give it a try JonB. As I mentioned, other than electricity, it hasn't cost me anything.

 

[Assimilator1]

If it's a P4 based Xeon you'd probably be better off turning HT off.
Benchmark DPAD (using muo1benchmark.exe) with HT on & off to see for certain.
Afraid the old P4 based architecure was never particularly fast, so DPAD scores aren't great.

Re benchmark graphs, even if the latest graph doesn't include the old scores you can see the older graphs in the earlier pages still .