Folding@Home

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
Certain WUs use more resources on your computer and so are worth more points.

They might be larger download and upload, and/or use more memory, and/or use more of the CPU. (You might have to lower an overclock to run these)

Currently BigWUs come from servers with addresses ending in these numbers:
.156 - (600 points / p114*)
.154 - (254-343 points / p13**)
.134 - (364 points / p147*)
.113 - (125 & 450 points / p19** QMD)

To get any of these, you must configure the client to "Allow receipt of work assignments and return of work results greater than 5MB in size" - answer "yes".

To reconfigure the console version: make a shortcut to FAH502-Console.exe by right clicking on it and then click Create Shortcut. Rename the new shortcut to "Configure". Now right click it and select "Properties" so we can add the flags. In the Target textbox, at the end of what is in there add a space, a dash, and configonly. It should look something like this:
C:\DC\fah2\FAH502-Console.exe -local -configonly
This will not run if F@H is already running. It will enter config and quit when done.

There are more flags, -advmethods directs the client to get WUs that are advanced beta WUs that might not be as stable as regular WUs; -forceasm will force the client to always use optimized code on a restart (even if the WU crashed the computer); -local tells the client to not use info in the registry (if you ever installed the GUI); -verbosity 9 fills up the FAHlog.txt file faster with info but that info helps to troubleshoot if something goes wrong. The Target textbox of my start up shortcut looks like this:
C:\DC\fah2\FAH502-Console.exe -local -forceasm -verbosity 9 -advmethods

To add flags to a service install look here.

You need the -advmethods flag to get some BigWUs.

The PROGRAM column on the server status web page shows if a flag is needed. Cross check which server a WU is from on the Currently Running Projects page.

You need almost 500 MB ram and P4/Xeon CPUs to get QMD BigWUs - [.113 - (125 & 450 points / p19**)]

On my PIII 1Ghz Dell laptop:
regular WU =.......... ~50-60 points per day (ppd)
600 point BigWU =.. ~105-112 ppd
p1478 BigWU =...... ~140 ppd

On a P4 3.0@3.3
QMD BigWU alone = ............ 456 ppd
600 point BigWU alone = ..... 253 ppd
IIRC 600 BigWU x 2 on HT = 180 x 2 = 360 ppd

The QMD WU uses up all or almost all of the memory bandwidth and so running two on HT can lose ppd!! It does with the 600 point BigWU anyway :confused:

You can see that the QMD really helps the ppd!

PS: a screensaver running really slows down the WU and will get less ppd. I set the monitor to power off to save electricity.
 

JeffCos

Golden Member
Mar 10, 2003
1,615
5
81
Ok, so you're saying that being as the console version on my HT3.2 only gets QMD's, I should shut off the Graphic client that I also have running because it'll boost my PPD?
 

trevinom

Golden Member
Sep 19, 2003
1,061
0
0
That is EXACTLY what he is saying. Given the choice, the "no frills" version is the fastest.
 

JeffCos

Golden Member
Mar 10, 2003
1,615
5
81
it's weird that the console version on that computer only gets 1910's 1911's or 1912's rarely does it get anything else. ok, i'll shut off my graphic version on that and just let it run the CLI and let you know what happens. You guys are the ones who told me Hyperthreading will boost my PPD so that's why i had 2 versions going, but we'll see.
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
Originally posted by: JeffCos
Ok, so you're saying that being as the console version on my HT3.2 only gets QMD's, I should shut off the Graphic client that I also have running because it'll boost my PPD?

If you don't have the GUI displaying the folding protein (which uses CPU cycles) it doesn't matter which client you use. It is the core that does the work and both clients use the same cores.

We are still learning this stuff with the BigWUs, especially the QMD ones. It surprised me how much better the QMD did alone than with the 600 pointer on HT.

JeffCos, please check what WU the GUI was doing. Maybe with a Tinker or regular Gromac you can still gain ppd on HT with a QMD. It might be that a QMD does not work well with another BigWU. I know that two QMD at the same time on HT does not give advantage (I don't remember that it hurt ppd though :confused: like it did with the 600 pointer)
 

JeffCos

Golden Member
Mar 10, 2003
1,615
5
81
After about an hour of only the CLI running i've gained 6mins/frame from about 20min/f to 14min/f. The GUI's project was an 1139 Tinker.
 

episodic

Lifer
Feb 7, 2004
11,088
2
81
Can someone post how to add flags to a service install? It seems your link above is broke, and I am really interested.
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
Originally posted by: JeffCos
After about an hour of only the CLI running i've gained 6mins/frame from about 20min/f to 14min/f. The GUI's project was an 1139 Tinker.

The difference in ppd:

1910 @ 20 min. = 324 ppd

1910 @ 14 min. = 463 ppd

For a difference or 139 ppd

How long was the 1139 Tinker taking per frame?

PS: just a few seconds can make a difference in ppd with QMD, are the times you gave pretty close?
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
Originally posted by: episodic
Can someone post how to add flags to a service install? It seems your link above is broke, and I am really interested.

The link still worked for me!

But here is the stuff from that link:

.... if you installed as a service follow these simple steps:

1) Shut down that instance through the Services manager snap-in

2) Run regedit and go to key HKLM\SYSTEM\CurrentControlSet\Services\ and edit the entry ImagePath, adding whatever flags you wish to the end. Exit regedit.

3) Re-open the Services manager snap-in, and verify that your changes appear there,
and then restart the service.
 

JeffCos

Golden Member
Mar 10, 2003
1,615
5
81
Originally posted by: GLeeM
Originally posted by: JeffCos
After about an hour of only the CLI running i've gained 6mins/frame from about 20min/f to 14min/f. The GUI's project was an 1139 Tinker.

The difference in ppd:

1910 @ 20 min. = 324 ppd

1910 @ 14 min. = 463 ppd

For a difference or 139 ppd

How long was the 1139 Tinker taking per frame?

PS: just a few seconds can make a difference in ppd with QMD, are the times you gave pretty close?

Yeah, they're pretty close, it's 14mins 50secs a frame for the QMD which will be done in about 5 mins and we'll see if I get another with the GUI version off. The Tinker was running 8min 24sec a frame. I'm gonna download F@H log stats to track this stuff. I've never had a reason to before so i'll put it on there now.

EDIT: Got a 1912 QMD now, i'm guessing it'll pull about the same 14m50s a frame so i might turn that tinker back on
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
The Tinker at 8 min 24 sec was giving 103 ppd

The QMD at 14 min 50 sec was giving 437 ppd

So if with both running, the QMD was at exactly 20 min (324 ppd) then:
103 + 324 = 427 ppd a loss of 10 ppd.
 

JeffCos

Golden Member
Mar 10, 2003
1,615
5
81
Hmmm so maybe Hyperthreading only works best if you don't have advmethods on? ok, but now i have a different QMD it's now a 1912 which was running at just over 10min/f with 1 instance running and 16min/f with both instances running the other instance is running @ 8m40s per frame.
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
1912 @ 10 min = 648 ppd
1912 @ 10m12s = 635 ppd
(didn't know how much "just over" is)

1912 @ 16 min = 405 ppd
1139 @ 8m40s = 100 ppd

Some QMDs can vary quite a bit from frame to frame. It would be best to average ten frames.

Looks like if you can continue to get QMDs it would be best ppd. If you have a AMD CPU, you could move the Tinker to it. AMD does great on everything. Or just add the -oneunit flag and finish it.

Jeff, thanks for your help in giving proof about QMDs.

Truth be told, I was running two until I wanted to gather data for the first post :confused:

EDIT: I just got one of those small QMDs worth 125 points. I will test it to see if it is any different.
 

JeffCos

Golden Member
Mar 10, 2003
1,615
5
81
I wish I had an AMD, I built my first rig just as a hobby because I thought it'd be cool to say I built this. Unfortunately I knew nothing about Distributed Computing or Overclocking back then, but in about a year when I upgrade I will get a board that can be OC'd and I'm definitely going AMD. Most likely one of those new Dual Cores, i'm sure the prices will drop alittle by then. I paused the Tinker and i'll just let the QMD's crunch by themself till I get a WU that's not a QMD, then I'll turn that tinker back on.
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
On my P4 3.0@3.3

1901 QMD with 600 pointer = 226 ppd
600 pointer with 1901 QMD = 145 ppd
for a total of 371 ppd

1901 QMD alone = 360 ppd

So for the smaller QMDs it doesn't matter.
 

GLeeM

Elite Member
Apr 2, 2004
7,199
128
106
Originally posted by: kmmatney
I just went over to the F@H web site and in the forums I read something to the effect that only P4s get the "best" work units. Something about these work units getting a point bonus. Is this still true?

These would be the QMD BigWUs. HyperThreading another WU does NOT increase ppd, but with the bonus and the SSE2/SSE3 boost, just running one process gives good ppd.

Originally posted by: Insidious
I think that is correct. It has been bothering me a little as well. I don't mind being assigned WUs talored to the resources I am donating, but the points inequity doesn't seem right. I remember UD weighting their points towards those with the "best" machines...... I quit that project.

-Sid

The QMD BigWUs are kind of a test of a new core that uses an Intel licensed library (or something like that).

You have to have 512 MB ram. One QMD uses available memory bandwidth, leaving little for another WU. Faster ram does better with these.

Running two on my DUAL HT Xeon only increases ppd by 20% :(
That computer easily handles four regular WUs, with an increase in ppd with each added process/WU (up to four).