FaD - We need your help!

[furie27]

Have you found a particularly efficient way to run it? More specifically, is there an optimal number of threads to run at once?

 

[Mloot]

Actually, it's best to only run one thread per processor. IIRC, even on HT P4's, there is no advantage to running more than one thread/physical CPU. Your points are based mainly on time and CPU rating. Higher ratings, of course, return more points. From what I've seen, when you run 2 instances on one CPU, your rating drops to half of what it would be if you were just running one thread. At least that's how I understand it.

 

[Ross Whitehead]

The TeAm has taken 16'th place!

 

[Ross Whitehead]

Originally posted by: Mloot
Actually, it's best to only run one thread per processor. IIRC, even on HT P4's, there is no advantage to running more than one thread/physical CPU. Your points are based mainly on time and CPU rating. Higher ratings, of course, return more points. From what I've seen, when you run 2 instances on one CPU, your rating drops to half of what it would be if you were just running one thread. At least that's how I understand it.

Are you sure about only running a single instance of FaD on a HT P4? I agree that each instance will have half the CPU rating but overall I would expect that running the two would benefit you the typical 5 - 10% gain which HT has demonstrated in other tests and scenarios.

 

[Ross Whitehead]

Originally posted by: grendel8
Ross, check out our production for today. I'm trying to figure where it came from, but more to the point I'm glad it's there!

I am not sure I understand the daily and weekly averages which are posted on the TeAm stats page. I recall everytime I have tried to compare them to the ones I generate for our stats, they are not the same. Maybe, I include or exclude an extra day in my average?

 

[Mloot]

Originally posted by: Ross Whitehead

Originally posted by: Mloot
Actually, it's best to only run one thread per processor. IIRC, even on HT P4's, there is no advantage to running more than one thread/physical CPU. Your points are based mainly on time and CPU rating. Higher ratings, of course, return more points. From what I've seen, when you run 2 instances on one CPU, your rating drops to half of what it would be if you were just running one thread. At least that's how I understand it.

Are you sure about only running a single instance of FaD on a HT P4? I agree that each instance will have half the CPU rating but overall I would expect that running the two would benefit you the typical 5 - 10% gain which HT has demonstrated in other tests and scenarios.

To be honest, I'm not sure at all. I just seem to remember reading about it in the FaD forums. I could very well be wrong, though, since I have no experience w/HT P4's.

 

[Ross Whitehead]

Our latest database performance article AMD Opteron vs. Intel Xeon: Database Performance Shootout stated on Page 4:

"Our tests have shown that Hyper Threading improved performance 3 - 5% on average and thus we left it enabled for all of our tests here."

 

[Mloot]

Cool. Thanks for clearing that up.

 

[BCinSC]

Is there ever a time when HT is NOT beneficial?

 

[Mloot]

Originally posted by: BCinSC
Is there ever a time when HT is NOT beneficial?

Couldn't say, really. I've not been fortunate enough to play around with any HT P4's. My P4 rig is of the 400mhz FSB variety, so no HT love there.

 

[Unforgiven]

Originally posted by: BCinSC
Is there ever a time when HT is NOT beneficial?

in DC projects....no....in other applications...yes. there has been shown to show degredation in speed in some applications with HT enabled. all i know is that it kicks major a$$ in DC projects and im glad ive got a few in my fleet

 

[Ross Whitehead]

In the early revisions of HT there were several situations/applications where HT was not beneficial. I believe that Intel has improved the HT logic and that in most cases there is a small gain recognized by enabling HT and very seldom is there a degredation. In support of this, I believe that 99% of all systems ship w/ HT enabled by default.

 

[BCinSC]

Does anyone actually make a 4-way or 8-way Opteron box? All the big names: IBM, HP/CPQ, and Sun seem to max out at Dually.

Oops, wait, Compaq has a Quad.

 

[Ross Whitehead]

The database server which hosts the databases for everything AnandTech related is a quad opteron similar to http://www.appro.com/product/server_4144h.asp

 

[BCinSC]

Whoa, whoa, whoa. What just happened? With last update to stats page, we've dropped to 18th and I personally have lost 200,000 points!

 

[Mloot]

I don't know why, but on the FaD forums they are saying that the stats page is being revised and that alot of weirdness is happening as a result.

 

[BCinSC]

Certainly instills a world of confidence.

 

[Mloot]

Originally posted by: BCinSC
Certainly instills a world of confidence.

Yep, certainly does. If they can't get this problem fixed by the end of the week, I may have to go to a different DC project, much as I like this one. Although this is as good a cause as any other DC program, I'll be honest enough to admit that I'm a stats whore. If FaD can't get their stats page right after all this time, maybe another project can.

 

[grendel8]

Speaking as one who just lost 300K points and dropped back to 10th, we are NOT happy. :frown:

Let's hope this is a temporary aberation.

 

[Ross Whitehead]

I also am a stats whore... who would have guessed. This stuff is irritating as hell and I hope this a growing pain and will sort itself out in the very near future!

If anyone wants a list of the number of jobs etc. you had as of Sunday I can provide that info.

:|

 

[BCinSC]

Oh, I'm a major stats whore.

Now Mloot, I know you're an AMD whore, too. I've got three MSI K7N2 Delta-L nForce2 Ultra 400 with Barton 2500+ and a 256MB stick of Corsair PC3200 ValueRAM. I set the FSB to 200MHz, thus CPU becomes a 3200+. My rating in FAD is 210 (though sometimes I get a brief run at 280). RAM defaults to 8-3-3-3.0. I've tried tweaking all sorts of settings, FSB to max of 210, 7-3-3-2.5, 6:5 CPU:RAM ratio, but either I blue screen or I stay at 210. I ask you and any other AMD tweaker: Is there anything I can do to improve?

 

[Mloot]

I'm afraid I can't help you, BC. The Opteron dually is the first pc I've built, and I no next to nothing about RAM timings and such. My other systems include a Dell (plug into the wall and play) and an HP laptop. Other than the new dually and the laptop, I have no experience with AMD procs. Sorry.

 

[Ross Whitehead]

Here is an update from THINK in one of the threads at their site regarding the state of stats:

"We now know why some recent jobs were not included and this will be addressed tomorrow. We also have a suspicion that we double counted some jobs which might explain why we have found more missing jobs than some members thought they had lost.

Obviously, we would have liked to get this right first time but wait until tomorrow before drawing any conclusions."

This gives me a little bit of hope and I think I will have to give them a day or two to sort things out.

I will not generate TeAm stats tomorrow if the stats look like a dog's breakfast still.

 

[Ross Whitehead]

Update from THINK:

"We have corrected some major omissions in the stats re-totalling yesterday and re-loaded the results table on the stats server. Members who observed problems yesterday should check again today.

The number of missing jobs was largely than we had expected but we have no reason to believe that we are double counting now."

The member stats look realistic at a quick glance, but the TeAm is still in 18'th place... I hope that is not correct.

 

[grendel8]

Looks like they fixed the individual member totals, but the team totals are still hosed (I hope).

I think they should have shut the servers down during the retotal. Not having anything visible would have been better than thinking I'd lost a whole lot of points.