Performance @ a Penalty?

[Sunin]

Now keep in mind this is not a bitch fest. Regardless of what we discuss I don't care what WU my folding farm gets, but these were observations I noticed.

Q6600 @ 3.1ghz, 2gig of ram with VM's Linux dedicated with 512mb = WU 2653 each and every time and that works out to 2050ish PPD per instance x 2 = 4100 PPD

Q6600 @ 3.6ghz, 3.25gig of ram with Windows SMP's = First WU's of 2653 producing 2250 per instance and together hitting in excess of 4500 PPD. Upon Finishing I was given, from reading the text, a much tighter deadlined WU 3065, worth 2144 Points. This is a beast of a WU hitting just under 18m per 1% which calcs out to about 1750PPD per instance or 3,500PPD (nothing to sneeze at)

But I'm curious what triggered that. My C2D is getting 2653, my VM'd Quad is getting 2653... but since the Q6600 at 3.6ghz finished the 2653 in under 19hrs all of a sudden I got a very tight deadlined WU... Preferred deadline was 1d 9hrs. I will complete it in just under 1d 4hrs.

Now people have commented that some WU speed up as they finish, but I was curious has anyone noticed that their Higher Performance Machines are getting less PPD? I'm not going to change mine and I really don't care in the long run, but was really odd to see...

In the end as I started the thread its all for a good cuase I won't adjust or do anythign and whatever WU I get is what I get, but what has everyone's experience been? I guess its just more of a curiousity. When you think about it my 3.1Ghz will never out produce my 3.6Ghz but on occassion it may out PPD it.

Ok hopefully this won't get flamed cuz really I'm not bitching more curious.

 

[Xilikon]

The problem here is that all WU is benched on a Clovertown (Xeon X53xx) machine. The actual value on other computers is sometime different unfortunately. However, there isn't too much we can do because this is more dependent on your own setup.

I noticed that on 3064 WU, others complained it's performing terribly vs others. On my work computer with a lowly E6300, it keep the same PPD as the 2653 ????

I think the 3065 is exclusive to quads and that's why you got it. I never ever gotten them because even my Q6600 doesn't show as quad to Stanford (all they see is 2 dual-core machines). I'm happy to not receive them...

Your only way to avoid them is to put 2x VM so Stanford doesn't see you have a quad.

 

[Sunin]

Your only way to avoid them is to put 2x VM so Stanford doesn't see you have a quad.

I'm not going to avoid them, someone has to do them, I'll just consider it taking one for the team

 

[Xilikon]

If you don't mind this, no problem at all (unlike a certain crybaby who complained recently) I was just merely stating a easy way to avoid them. There are some who may see them, flush the WU and get a different one, which is not a good behavior.

Myself, on console clients, some WU is better (like DGromacs) than others with easily 2x -3x the PPD vs others but I don't complaint because it's what Stanford assign us. Another crappy WU is the 2652, which is 200-400 ppd less than the 2653 and I crunched about 100 of them.

 

[Crazy Larry]

My work units lately have been big ones - they are usually 500,000 steps but now I'm getting these 10,000,000 step ones. They have been taking longer to complete (20 minutes per 1% as opposed to 18 minutes per 1% for the earlier work units)

 

[Kendrak]

My farm gets between 10k and 12k ppd depending on what mix it has. I just feel it's the luck of the draw and I gets what I gets.