Weird Points

[7im]

p6040 and p6041 are hardly aces. They have a high point value but are below average performers in terms of ppd on most machines. That's 7im's point. There is no way to find the best performing WUs by looking at psummary.

They worked for me. 6000+ points vs. 3300 points on the average SMP WU.

If you're talking PPD then that's a slightly different discussion; remember that this discussion started as "why did I get crazy points this one update". Running more than one WU on a multi-core processor give me fewer points in a day than running them sequentially. I've checked my stats on KakaoStats over time, and have observed it and acted upon it. That's what I mean by an Ace. 6040 and 6041 haven't been dogs for my point count; even when factoring in a replacement by another SMP WU and determining PPD, it's running WUs in parallel that lower my PPD, so you may be seeing 6040 and 6041 as "dogs" if you are running them that way; they will take forever.

Correlation is not causation.

And again, you have NO WAY to know how a WU will perform by looking at the PSummary. PSummary only shows total points, not bonus, and not PPD. The information on Psummary has little to do with how well your system peforms on a particular work unit. And because everyones' systems are slightly different, they all perform slightly differently. Add to that a slight difference in performance simply because I have Windows and you have Linux, and PSummary becomes completely useless for estimating WU performance.

[World Control]

Correlation is not causation.

And again, you have NO WAY to know how a WU will perform by looking at the PSummary. PSummary only shows total points, not bonus, and not PPD. The information on Psummary has little to do with how well your system peforms on a particular work unit. And because everyones' systems are slightly different, they all perform slightly differently. Add to that a slight difference in performance simply because I have Windows and you have Linux, and PSummary becomes completely useless for estimating WU performance.

I've worked with my data prior to this thread and am confident in my findings. It's my experience and probably won't be yours. Please remember that we weren't talking about differences in hardware or cherry-picking WUs. I referred you to the Project Summary to tell you what WUs I was talking about in a effort to inform the discussion. The thread started as "why am I seeing these points?"

I track my PPD just like everyone else, and for me the Project Summary page is useful for gauging how a WU will stack up points-wise on a single one of my machines once it's determined how WUs will perform after a period of observation.

[7im]

I don't doubt your findings. You are looking at your own data, and only then making like comparisons to data on Psummary. I can see how that works. But then you are basing the WU performance on your own data, not on Psummary. So again, Psummary is not the data on which WU performance is determined.

[World Control]

I don't doubt your findings. You are looking at your own data, and only then making like comparisons to data on Psummary. I can see how that works. But then you are basing the WU performance on your own data, not on Psummary. So again, Psummary is not the data on which WU performance is determined.

Well, it's a fairly rigorous comparison, and the best we've got, I think. Unless we can work with the algorithms and the vagaries of processing then we'll always be guessing. Let's agree to disagree. We've probably burned up too many pixels on this topic already.

[ChasR]

Edit: missed page two posts so this is late to the party.

No one that does the math runs more than one instance on any but the largest multicore systems. The exponential QRB rewards one instance per machine quite well.

Here's a sampling of SMP WUs on a Q6600 @ 3.2 GHz, Ubuntu 10.10:

Code: Select all

 Project ID: 6040
 Core: GRO-A3
 Credit: 1395
 Frames: 100


 Name: COINS SRV4
 Path: \\Coins-srv4\fah\SMP\
 Number of Frames Observed: 100

 Min. Time / Frame : 00:15:07 - 7,893 PPD
 Avg. Time / Frame : 00:15:11 - 7,842 PPD

Project ID: 6050
 Core: GRO-A3
 Credit: 481
 Frames: 100


 Name: COINS SRV4
 Path: \\Coins-srv4\fah\SMP\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:04:45 - 9,012 PPD
 Avg. Time / Frame : 00:04:48 - 8,872 PPD

 Project ID: 7151
 Core: GRO-A3
 Credit: 585
 Frames: 100


 Name: COINS SRV4
 Path: \\Coins-srv4\fah\SMP\
 Number of Frames Observed: 100

 Min. Time / Frame : 00:05:01 - 10,226 PPD
 Avg. Time / Frame : 00:05:02 - 10,175 PPD

Project ID: 7500
 Core: GRO-A3
 Credit: 529
 Frames: 100


 Name: COINS SRV4
 Path: \\Coins-srv4\fah\SMP\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:03:57 - 13,153 PPD
 Avg. Time / Frame : 00:04:04 - 12,591 PPD
 Cur. Time / Frame : 00:04:07 - 12,346 PPD
 R3F. Time / Frame : 00:04:06 - 12,415 PPD
 All  Time / Frame : 00:04:04 - 12,554 PPD
 Eff. Time / Frame : 00:04:09 - 12,210 PPD

Most of us would prefer to fold 3.5 p7500/day at a value of 3525 points each than fold .95 p6040/day for a value of 8200 points.

[7im]

I don't doubt your findings. You are looking at your own data, and only then making like comparisons to data on Psummary. I can see how that works. But then you are basing the WU performance on your own data, not on Psummary. So again, Psummary is not the data on which WU performance is determined.

Well, it's a fairly rigorous comparison, and the best we've got, I think. Unless we can work with the algorithms and the vagaries of processing then we'll always be guessing. Let's agree to disagree. We've probably burned up too many pixels on this topic already.

Since you've developed a good predictive methodology, please share. Even though I've been running this project for 7+ years, and was a forum mod for 2+ years, I'm always open to new ideas.