Points drop off

Moderators: Site Moderators, FAHC Science Team

Re: Points drop off

Postby 7im » Sat Sep 01, 2012 11:47 pm

XZ, I understand about the change to BigAdv, but that was decided long ago ans has no bearing on a discussion about regular SMP work units. Your comments about low points would also be a lot more helpful if you could include which project numbers were scoring "low" for you.
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 10189
Joined: Thu Nov 29, 2007 5:30 pm
Location: Arizona

Re: Points drop off

Postby Xavier Zepherious » Sat Sep 01, 2012 11:49 pm

just look at my stats
http://folding.extremeoverclocking.com/ ... =&u=541641

3930k produces about 50k PPD(I was doing 55k) before all of this
my other rig is a constant 20k (gtx570+ q6600) gives approx 18-22k

it's now taking both rigs to maintain around 50K avg
that be a 30% drop
Xavier Zepherious
 
Posts: 133
Joined: Fri Jan 21, 2011 9:02 am

Re: Points drop off

Postby Joe_H » Sun Sep 02, 2012 12:13 am

Xavier Zepherious wrote:the point of having forums is to discuss things and maybe improve things - if they all felt there is a problem and you let people have free speech here and not lock out threads it be nice


Actually, from viewtopic.php?f=2&t=15, I will quote:

Our primary goal is to help people who are having trouble getting the FAH program to work for them and, in the process, collect reports of bugs or other problems that potentially need to be fixed by the Pande Group.


To that end, some discussion of points discrepancies does go on, and a number of projects have had adjustments made. Those adjustments have been both up and down. When the discussion goes too off-topic or repetitious it gets moved and/or halted.
Joe_H
Site Admin
 
Posts: 6682
Joined: Tue Apr 21, 2009 5:41 pm
Location: W. MA

Re: Points drop off

Postby Grandpa_01 » Sun Sep 02, 2012 12:18 am

Umm most of us folders do not care about statistical data only what the end results of our folding is, if there is a large variation in PPD for work done then that is what it is. The claim I made is correct for my machine and others have noticed the same thing on other forums. I really do not care about the logistics of stats as most do not, I very seriously doubt that anybody has done any stats collecting that would meet the criteria so there are just claims, I have shown the data I have and that is real to me. There is a very real variation in PPD on WU's day in and day out for me and for others, which points are correct I do not know or care but the reality is there is a wide spread and there really should not be, equal pay for equal work and in some cases it is not close equal.

7im your last two links are the same and point to the same thread the title to the last one, I believe is wrong the points or k factor were never changed for the p7645, 7646, 7647 that I can remember and if it was there still is a 50% variation on my machine, I currentley have one running and it is still 50% below the upper end smp WU's in PPD. I am not saying it needs to be increased to the upper end but it does need to be looked at again.
Image
2 - SM H8QGi-F AMD 6xxx=112 cores @ 3.2 & 3.9Ghz
5 - SM X9QRI-f+ Intel 4650 = 320 cores @ 3.15Ghz
2 - I7 980X 4.4Ghz 2-GTX680
1 - 2700k 4.4Ghz GTX680
Total = 464 cores folding
User avatar
Grandpa_01
 
Posts: 1122
Joined: Wed Mar 04, 2009 8:36 am

Re: Points drop off

Postby bollix47 » Sun Sep 02, 2012 12:35 am

I believe is wrong the points or k factor were never changed for the p7645, 7646, 7647 that I can remember


Hint.
bollix47
 
Posts: 2871
Joined: Sun Dec 02, 2007 6:04 am
Location: Canada

Re: Points drop off

Postby Grandpa_01 » Sun Sep 02, 2012 2:20 am

Yes I know that but the post I made was after the change to 4, the 1 I have folding right now is the same frame time as I reported in that thread and is reporting the same PPD as the one reported in that post, thus never changed to my knowledge after I reported that they were still quite shy of the rest of the smp WU's
User avatar
Grandpa_01
 
Posts: 1122
Joined: Wed Mar 04, 2009 8:36 am

Re: Points drop off

Postby Xavier Zepherious » Sun Sep 02, 2012 2:31 am

Im running a 7646 right now
est 22.8k points
est 34k PPD

609x series produced around 50k PPD on this rig (3930k)


on q6600 - stock 2.4

6097 - 4.7k PPD (13 completed)
6098 - 4.9k PPD (8)
6099 - 4.9k PPD (6)
8052 - 3.8k PPD (9)
8064 - 5.6k PPD (2)
8066 - 5.2k PPD (1)
8067 - 5.5k PPD (3)


What is currently needed on FAH 7 is stats history like FAH GPU tracker
Xavier Zepherious
 
Posts: 133
Joined: Fri Jan 21, 2011 9:02 am

Re: Points drop off

Postby Joe_H » Sun Sep 02, 2012 4:46 am

Grandpa_01 wrote:Umm most of us folders do not care about statistical data only what the end results of our folding is, if there is a large variation in PPD for work done then that is what it is.

This and what follows is better argument than you gave before. When you start using exaggerations like 50% of the average or way less than most, that is a statistical statement and is not supported by what data you have posted. And generalizing from a single sample does not further your case. What you can show is that your one 7646 got half the PPD of your highest, not an average.

As for your last post in response to the hint from bollix47, what you said then was "They are still a little shy of the latest 8xxx", not the "still quite shy" that you use now. Going back to those numbers in May, the 7646 got practically the same PPD as the 8001 and 8004, it was within about 80% of most of the rest.
Joe_H
Site Admin
 
Posts: 6682
Joined: Tue Apr 21, 2009 5:41 pm
Location: W. MA

Re: Points drop off

Postby Grandpa_01 » Sun Sep 02, 2012 7:16 am

Is 80% acceptable that is a real figure of 40% which is 20% either side of the norm.? Will it keep the folding public happy.? There are some pretty simple solutions that would make things like this less likely to occur or at least not such wide spreads, such as verification by the systems that are going to be running them. Maybe some of these should be investigated and tried. What could it hurt, If it failed at least there was an attempt made to correct the problem / problems.
User avatar
Grandpa_01
 
Posts: 1122
Joined: Wed Mar 04, 2009 8:36 am

Re: Points drop off

Postby Xavier Zepherious » Sun Sep 02, 2012 9:27 am

7im wrote:XZ, I understand about the change to BigAdv, but that was decided long ago ans has no bearing on a discussion about regular SMP work units. Your comments about low points would also be a lot more helpful if you could include which project numbers were scoring "low" for you.



As I said I will fold regardless - even with the loss of Bigadv I would have been happy with 50K WU's (609x series - since I was folding them before I did bigadv) and a few of the new ones that score even higher - like 70k
but the drop to 34k with the 764x series is like taking a kick in the teeth after losing the big units(insult to injury)

the only other low ones I have a stats on is the 8051, 8052 (with the 764x series) .these are showing low

there may be more (but V7 keeps no stats records) and the only one that keeps them is FAH GPU tracker for me
(I switch to V7 on the new rig - since other clients will be unsupported eventually)

I will update my list when I find low units - even mid 40's would have been fine - but mid 30's?

on the q6600 the point swing even a 25% difference is only 1-1.5k PPD
on a 3930k it is much more significant
Xavier Zepherious
 
Posts: 133
Joined: Fri Jan 21, 2011 9:02 am

Re: Points drop off

Postby bruce » Sun Sep 02, 2012 9:54 am

Xavier Zepherious wrote:
7im wrote:XZ, I understand about the change to BigAdv, but that was decided long ago ans has no bearing on a discussion about regular SMP work units. Your comments about low points would also be a lot more helpful if you could include which project numbers were scoring "low" for you.



As I said I will fold regardless - even with the loss of Bigadv I would have been happy with 50K WU's (609x series - since I was folding them before I did bigadv) and a few of the new ones that score even higher - like 70k
but the drop to 34k with the 764x series is like taking a kick in the teeth after losing the big units(insult to injury)

the only other low ones I have a stats on is the 8051, 8052 (with the 764x series) .these are showing low

there may be more (but V7 keeps no stats records) and the only one that keeps them is FAH GPU tracker for me
(I switch to V7 on the new rig - since other clients will be unsupported eventually)

I will update my list when I find low units - even mid 40's would have been fine - but mid 30's?

on the q6600 the point swing even a 25% difference is only 1-1.5k PPD
on a 3930k it is much more significant


How about balancing your reports with both low units that you find and high units? Consistency is a fine goal, but that means fixing things that are out-of-line in either direction. It's not reasonable to choose the highest PPD one has ever gotten and call it "normal" and then expect all the lower PPDs to be adjusted upward.
bruce
 
Posts: 20140
Joined: Thu Nov 29, 2007 11:13 pm
Location: So. Cal.

Re: Points drop off

Postby artoar_11 » Sun Sep 02, 2012 12:37 pm

As far as I see from other participants in the discussion, we compare PPD between projects core A3 and core A4. Most of us have worked a long time with projects A3.
A3 WUs are well balanced. With VM/Ubuntu I getting stable over time - 21-23k PPD. Win7 with the same CPU - 17-19k PPD.

With the A4 is not the same. Here with some WUs I getting - 40k+, with others - 14k.
Or WUs points are not scaling well with faster CPU (-/+), or base points is not accurate?

Anyway, folding should continue :)
artoar_11
 
Posts: 656
Joined: Sun Nov 22, 2009 9:42 pm
Location: Bulgaria/Team #224497/artoar11_ALL_....

Re: Points drop off

Postby bruce » Mon Sep 03, 2012 12:32 am

The points and deadlines for FahCore_a3 were originally designed to work with hardware that was, for the most part centered around 4 FPUs. They generally scaled well from smp:2 to about smp:8 (HyperThreaded). Deadlines didn't have to work for the uniprocessor and nobody complained about what happened for 16 ... 24 ... 32 cores because those folks preferred FahCore_a5.

As you say, with FahCore_a4, it's not the same. It was extended to include the uniprocessor so it was more difficult to set deadlines and K-Factors. Considering the recently announced temporary 10% bonus, I suspect that there is still a study going on at Stanford to clarify all the issues.
bruce
 
Posts: 20140
Joined: Thu Nov 29, 2007 11:13 pm
Location: So. Cal.

Re: Points drop off

Postby Xavier Zepherious » Wed Sep 05, 2012 4:54 am

well I let my 3930k fold 8051's for 2 days (all I got was this WU for two days)
back to folding with both now (so my point level stays up)

points down to 37-38k ...from 50k's with 609x's
isn't this exactly like the benchmarks showed. 3.8k versus 5k for others

I keep watch for other units (during the day) to see which seem high and if other seem low

7085 ...30k est ppd
Xavier Zepherious
 
Posts: 133
Joined: Fri Jan 21, 2011 9:02 am

Re: Points drop off

Postby Xavier Zepherious » Thu Sep 13, 2012 2:59 am

7546 est PPD- 34k
7645 est PPD 34k
7085 - 30k PPD est
8064 43k
8065 46k
7083 29k
7032 38k

all one has to do is look at my folding numbers bouncing around from 35k PPD to 62K PPD with 2 rigs (and 2nd rig does around 15-20k)

All I can say it is FIX THE POINT ASSESSMENT
Xavier Zepherious
 
Posts: 133
Joined: Fri Jan 21, 2011 9:02 am

PreviousNext

Return to V7.1.52 Windows/Linux

Who is online

Users browsing this forum: No registered users and 1 guest

cron