Folding Forum

Hello,

I'm new to Folding@Home and I have a question about the best client configuration.

I have an Intel Core2 Duo E6550 CPU. I run TWO clients (classical client, not SMP) under LINUX.
I see my two cores being used. It's ok. It take 8hours to complete a work unit (project 68xx).

Today I wanted to try the "BIG" parameter (instead of "normal") in the config file (for each client).
I got a p3865 WorkUnit, and it now takes 1 day and 8hours to complete ! and my PPD dropped considerably (from 440 to 370).

What is best ? two instances with normal WU or two instances of BIG WU ?

Thanks for any help...

If you're qualified for QRB (Quick Return Bonus), a 3865 WU alone should give you over 600PPD at your completion rate (19:12 TPF), according to bonus calculator. The interesting question is, do A4/A6 WUs also count towards qualification?

PS: I assume you used the -configonly switch to set the "big" option? Editing client v6 config file manually may lead to problems.

The stats for 3865-3867 are indeed set up to give a bonus. Note that because of some quirks of hard coding, A6 can only be used on SMP using the v7 client (now in beta).

@Napolean, top of the forum is this link described as 'project descriptions' - http://fah-web.stanford.edu/psummary.html

That page will tell you what you wanted to know, now and in future situations. ( bonus is kfactor ).

mrshirts wrote:The stats for 3865-3867 are indeed set up to give a bonus. Note that because of some quirks of hard coding, A6 can only be used on SMP using the v7 client (now in beta).

Weird, OP mentioned two classic clients on Linux. I can only assume he's using v6 clients, as I don't see the point of installing two v7 clients. Two uniprocessor slots in v7 would make sense, of course.

@MtM: I'm aware of the K-factor, and like mrshirts pointed out, one will certainly get a bonus if one is already qualified. I just wasn't 100% sure if they count in the "must complete 10 SMP/bigadv WUs first" -rule. I got the impression that the OP isn't eligible for QRB yet, since he hasn't been running SMP.

moika555 wrote:I run TWO clients (classical client, not SMP) under LINUX.

That's a very good question, I would like to assume that the rule hasn't been updated since conception when only smp had QRB, and that it's therefore outdated. Logically any unit with a kfactor should count towards the 10 unit requirement.

Edit: I send a pm to Bruce linking to your post, I think your confusion about the subject will be shared by many others, the requirement should be listed differently ( that is if it's true that any qrb wu count's towards the 10 completed number ).

Also, we haven't asked OP if he's using a passkey yet, which is also quite important when considering bonus factors.

And lastly @OP, an observed drop like you're describing could be nothing but a statistical anomaly which will level out over time.

MtM wrote:That's a very good question, I would like to assume that the rule hasn't been updated since conception when only smp had QRB, and that it's therefore outdated. Logically any unit with a kfactor should count towards the 10 unit requirement.

Edit: I send a pm to Bruce linking to your post, I think your confusion about the subject will be shared by many others, the requirement should be listed differently ( that is if it's true that any qrb wu count's towards the 10 completed number ).

Also, we haven't asked OP if he's using a passkey yet, which is also quite important when considering bonus factors.

Right: A passkey is qualified when you've completed 10 bonus-qualified WUs using the specific Passkey+UserName pair.

I didn't check the FAQ, but we'll get it updated if it hasn't already been fixed.

BTW, PM's are not required. I read EVERYTHING that's posted. Sometimes when I'm travelling it may take me a few days, but in cases like that, the PM doesn't help anyway when I'm not on-line.

moika555 wrote:What is best ? two instances with normal WU or two instances of BIG WU ?

There isn't an absolute answer to this question for the what that question does is select your WU from a different group of projects. Projects come and go, so the answer that works today may not work at some later time when the project list is different. In theory big projects use more RAM and network bandwidth compared to normal so the fact that the WU's take more time does not surprise me in the slightest. That being said, all the WU's are bench-marked to match PPD regardless of size. The flaw is that they are bench-marked on a standard machine but your machine is unlikely to be identical to the standard so your mileage will vary.

Without seeing logs my guess is that the two projects also used different cores. Newer uni-processor projects use the A4 core and that uses a different formula to generate points that the old uni-processor cores. They are designed to give similar PPD's but they vary significantly in the shape of their curves. Lower performing CPU's do much better with the old tradional cores and higher performing CPU's do better on the A4 core. This is because the A4 gives bonus points for quick completion but if you are close to the deadline you may not be getting any bonus and in that situation the old WU's will give more PPD.

Another possible issue is how you measure PPD. If you are not calculating the bonus points then obviously the old cores will look quite valuable. Make sure that you are using HFM.net or a recent version of FAHMon that automatically calculate the bonus to monitor your folding. The new V7 client does not currently calculate bonus so may provide inaccurate PPD's. I do not recommend manually calculating PPD with any of the bonus giving A4 projects for it is just too easy to make a mistake.

All the above being said, the 20% difference between a 440 and 370 PPD is not that great. It could just be the natural variation between projects. I've seen far worse discrepancies between two different projects.

@Bruce & MtM: Sorry, I must have missed this somehow. At least Bonus point formula FAQ subtopic already has appropriate wording - "10 bonus-eligible WU's".

Thank you for all your answers.

I'm running client V6, one for each core.

I reverted to NORMAL Wu size (I think it's better for the machine it's running on, even if it's a dedicated one) but It's still working on project 3xxx WU that take almost 2 days to complete.... I restarted the client and in client.cfg I see bigpacket=normal so I'm wondering if it's a bug ?

I'll see if it switched after processing those units.

The normal vs. big switch only matters when getting a new WU. The old/current WU needs to complete (or fail) before a new one is fetched using the small/normal/big switch.