TPU's F@H Team

[Steevo]

If the mods will allow I will make a .zip file that I use on all my systems available. It is a complete service installer package with our team number and preconfigured for hih point-speed units.


The file is 1.6MB if this does not work I can-will create a yahoo mail address and provide the user-password via PM so all can download it at yahoo's bandwidth.

Or if I get my FTP back up I can make it available tonight.

 

[W2hCYK]

I'm interested, and i think that would be a great help to everyone. Preconfigured and everything, would prevent alot of confusion.

I'm not a mod, but i'd love to see this be done. talk to wazzle

 

[Alec§taar]

Steevo, need your input/feedback on a question I had last page

See subject line/title: and, answer this question if you know the answer to it please, tia!

(I also wrote you in "pm" as well, regarding this line of questioning)



I would like an answer to my questions above regarding WHICH of the 2 executables this process runs is MOST directly associated with the actual work-in-process on the unit data!

E.G./I.E.-> FAH504-Console.exe, OR FahCore_65.exe (noted above in my last post).

APK

P.S.=> Others have responded (Jimmy2004) but, I would like a "2nd opinion" to verify this, as it can affect what I am "up to" here in messing around w/ the console mode variety/version of this project suite's apparently "client-server"/"mgt.-workhorse" design (what it appears to be like to me, much how SETI@Home/BOINC work)... thanks! apk

 

[Steevo]

Fahcore is the working unit, and the priority is fine as low. You will notice the numbers and letter behind change to represent the core that is being used a the time.


The core "Fahcore65" etc... if the actual shell that the work is done in, to see your unit ID you will need to open the GUI or in the console version open the worklog.txt, then you can compare the workunit number to what is listed onthe FAQ page for folding to see what you are actualy working on.

 

[Alec§taar]

Jimmy2004 &/or Steevo: Thanks much fellas! Going to put this to work now, or try to

See subject line/title: Thanks guys!



* NOW, time to try to "put this to work" & experiment more + see how it works out!

APK

P.S=> In addition to the console mode app's commandline switchwork I use now:

-local -forceasm -verbosity 1

Next, it's running the actual "workhorse" portion (apparently FahCore_65.exe since both of you guys noted this to me) @ HIGH cpu, & on each of my 2 instances running from their own folders, onto their own CPU core as well!

(All done via taskmgr.exe's affinity & priority settings via the process tab right-click on processname feature)... apk

 

[Alec§taar]

REALTIME, can be done on "DualCore" but NOT 2 instances with "affinity" to CPU

See subject line/title: May save some of you experimenting time...

E.G.-> I set the 1st instance of it to "REALTIME" for FAHCore_65.exe (& now I startup the process for the FAHConsole.exe to LOW, via start /low in a .cmd file (32-bit batch basically)).

(& it's doable IF you only run 1 instance of it (via the -local switch) set with AFFINITY to CPU 0 for instance & set to REALTIME on said 1st possible instance via the -local switch for the consolemode/tty/DOS commandline client).

BUT, when I tried setting the 2nd instance to HIGH cpu priority (and set affinity to CPU 1, the 2nd core here on this CPU) & the system stayed up & running (but laggy)!

HOWEVER, when I kicked up the 2nd instance & set it to REALTIME (w/ affinity to 2nd CPU 1)? The appearance of "lockup" happened...

* Oh, taskmgr.exe eventually responded so I could "kill" the process, but I wouldn't recommend this, even on a DualCore/SMP rig.

(Wouldn't even TRY it on an Intel single core, even with HyperThreading)

APK

P.S.=> It appears that going with HIGH cpu priority works ok though, for BOTH instances... apk

 

[Jimmy 2004]

I would say 'told you so' but as it was an experiment to find out if it did happen I doubt you'd care!

 

[Alec§taar]

As long as I don't post here (or, anything else)? I can do it (sleeptime operation)

See title/subject-line, & check it:

I would say 'told you so' but as it was an experiment to find out if it did happen I doubt you'd care!

I will do it that way, while I sleep (HIGH, or @least AboveNormal (cpu #1 affinity & priority), & REALTIME (cpu #2 & priority)), it'll work!

(I did things like this with SETI@Home, you just have to find the right "mixture" of what you leave running - all else is @ minimum! Even explorer.exe gui killed + trimmed to a minimum of services only (nothing else))



I may not be able to do BOTH @ REALTIME cpu priority assigned, 1 to each cpu, but I can do 1 of them that way, & the other @ AboveNormal, or even HIGH!

(This much I know I can pull off... or @ least think so in theory, going to find out again!)

APK

P.S.=> The minute I close Opera, from posting here?

Back up she goes running dual instances of it as proscribed above priority & cpu affinity-wise, & then I will see how it is later (if I can pull up taskmgr.exe to kill instance #2 in REALTIME) & if I can 'pull out of it'...

I did to post this - so it will be doable again later! apk

 

[Alec§taar]

Found out 1 instance of REALTIME is possible, & 1 HIGH (cpu priority) on AMD DualCore

I am keeping it @ 1st & 2nd instances running @ HIGH cpu priority!

I had to "toss" doing 1 @ REALTIME, & 1 @ HIGH (or, AboveNormal) cpu priority allotted PER CPU by using AFFINITY settings as well!

(The problem is, that if done that way? It is WAY too hard to get out of it: I.E.-> It takes a lot of time to get taskmgr.exe up on the screen to get the system actually DOING things again, after it has been interactively commanded by me to do so via taskmgr.exe & the mouse/keyboard (even these lag setup that way - device driver driven, or not)).



* Experiment goes on, but I think that is the way I will run it from now on...

APK

P.S.=> The "rumors" I heard while on the SETI@Home team for MS is true as well/also - the units processing times on "FAH" are a great deal longer as well imo... apk

 

[Jimmy 2004]

Well my AMD has now overtaken the crappy Celeron. I've left it on for almost 5 hours so far today and I've broken the 200 frames barrier . How long does it take everyone else to do a frame? On my quick system it takes about 4m 10s.

 

[Steevo]

Frames are dependant on the work unit being processed.
Larger frame count work units process faster. IE 400 frame count unit has higher processing-memory and network load than a 2000 frame count.
Some QMD cores "96" are very large and only get assigned to specificly.
Intel CPU's
More than 512Mb of free RAM
Higher than 2.0Ghz (from what i have seen) processors.
Only with the -advmethods, large WU's tags used. (this is setup as default in the service installer)


I still have 6 machines to add. In order of power.
1) 1Gb RAM, 3.2Ghz P4
1) 512Mb RAM. 2.4Ghz P4
2) 512MB RAM, 2.0Ghz Celery
1) 256MB RAM, 1.3Ghz Celery
1) 512MB RAM, 400Mhz K6-2 (fear this bitch!!!)


Plus I am trying to get another tower as my really old Compaq server at home is out of space, plus the SCSI drives are whining. I have to go wipe some systems for sale and I might bargain one out of them and add a few drives on a RAID card for my FTP.

 

[Steevo]

The zip file is now available on yahoo mail, with a username and passoword that I will give to anyone interested.


PM me.

 

[Alec§taar]

I don't even see the frames output reporting...

-verbosity 1

* That consolemode/tty/Character mode is in effect/place here, in use as I write this...



APK

P.S.=> Gotta fly... apk

 

[Jimmy 2004]

-verbosity 1

* That consolemode/tty/Character mode is in effect/place here, in use as I write this...



APK

P.S.=> Gotta fly... apk

What does that command actually do?

 

[Alec§taar]

What does that command actually do?

Stall charmode/consolemode/tty output to the screen - that too, is overhead, so it improves performance by not doing it & taking those cycles for the actual data-processing.

Arguments: -local -config -forceasm -verbosity 1



* Best performance I can get out of it... software-config. side concentration, whereas I was working on the timeslice & affinity portions earlier OS-side (hardware see below, this runs from the SSD, 2 instances, 1 to each CPU, mgt. portion @ LOW cpu, & unit processing client portion @ HIGH cpu across 2 cpus)!

APK

 

[Jimmy 2004]

Stall charmode/consolemode/tty output to the screen - that too, is overhead, so it improves performance by not doing it & taking those cycles for the actual data-processing.

APK

You know a lot of stuff...

268 frames so far!(on my AMD) Finally folding properly.

Edit: we should make it our goal to break into the top 1000.

 

[Alec§taar]

You know a lot of stuff...

Not really, I just read about it during its setup & went @ it, making it not even report frames... I've written this stuff for years and output to screen (or any device) costs cycles, heavier ones imo.

To get the MOST out of it & toying w/ my OS (stalling services & the GUI shell even while I sleep & this runs) + other running apps (stopping them) as well before that.

APK

P.S.=> I've done 2 units so far, 482 credit, means I can "push" apparently 241 per cpu scoring so far...

http://fah-web.stanford.edu/cgi-bin/main.py?qtype=teampage&teamnum=50711

That's GOTTA be up there for a single-system score, per-cpu output, on our team (you have to look @ it machine by machine & avg. per-unit though to determine this)! apk

 

[Jimmy 2004]

What is the credit score based on exactly? How fast you do it?

 

[Alec§taar]

So, Check it out:

See subject-line/title, & this image:

* There/here we go... out to help cure some bogus disease hopefully!

APK

P.S.=>

What is the credit score based on exactly? How fast you do it?

Simple - I did 2 units so far, & scored 482 - division (do the math):

If these stay @ least SOMEWHAT LINEAR, you can get a performance index out of it of somekind, vs. peer machines/systems on your team @ least, & over time, one of your system vs. itself really!

Programs like these?

They're benchmarks too imo, depends on how you use them & their data output - I viewed SETI@Home as a benchmark of sorts (cpu/caches/RAM/disk I-O etc. used)... apk

 

[FLY3R]

Frames are dependant on the work unit being processed.
Larger frame count work units process faster. IE 400 frame count unit has higher processing-memory and network load than a 2000 frame count.
Some QMD cores "96" are very large and only get assigned to specificly.
Intel CPU's
More than 512Mb of free RAM
Higher than 2.0Ghz (from what i have seen) processors.
Only with the -advmethods, large WU's tags used. (this is setup as default in the service installer)


I still have 6 machines to add. In order of power.
1) 1Gb RAM, 3.2Ghz P4
1) 512Mb RAM. 2.4Ghz P4
2) 512MB RAM, 2.0Ghz Celery
1) 256MB RAM, 1.3Ghz Celery
1) 512MB RAM, 400Mhz K6-2 (fear this bitch!!!)

The zip file is now available on yahoo mail, with a username and passoword that I will give to anyone interested.


PM me.

OH, this is awsome! I am interested in getting this from you. Can you give me a list of the options that you pre-configured? And also if i am running the console version, not as a service, where do i go to enter comand lines or edit it, where are you guys adding -advmethods with the console version. Because I am only knolagable of how to and -advmethods and other lines when running the console version as a service.

 

[Steevo]

PM sent.

 

[Jimmy 2004]

Guys, if I set priority to idle does it have ANY effect on my gaming performance with it in the background? Or does it not alter it in any way and just go to sleep?

 

[Alec§taar]

http://fah-web.stanford.edu/cgi-bin/main.py?qtype=teampage&teamnum=50711

On 06/27/2006 -> 7th place, AlecStaar, 482 score/2 units done = 241 per unit

On 06/28/2006 -> 5th place, AlecStaar, 964 score/4 units done = 281 per unit

4 units done @ score of 964 = 281 per unit avg. score thusfar (that's up 40 points from my previous 2 unit avg. of 241 per unit)

What I was hoping for appears to be correct - That the switchwork + CPU priority allocation & affinity work are working effectively for me as well : As my avg. score (PER UNIT AVG.) shot up 40 points more ~ 20% increase apparently using those techniques.

Console mode client switchwork I use is as follows (for others reference so they too can run as fast/efficient as possible - the -local one might ONLY be for DualCore/SMP setups though imo):

Arguments: -local -config -forceasm -verbosity 1

Why'd I go & do all of this? Because of what I outlined in this thread here:

http://forums.techpowerup.com/showthread.php?p=110181#post110181

Projects like these do well with L2 cache being "clean" as possible, & the ONLY way I know to do that, is to use AFFINITY settings (see that url above about how L2 cache works, & how you can get it to work more efficiently using affinity settings - avoiding "cache pollution")



* It also appears the data is SOMEWHAT "linear" in nature, which is good - I wasn't sure if the units here were of a uniform nature in complexity/size & processing duration periods is why I say that.

(It can be used as a sort of benchmark then imo)

APK

P.S.=> Besides being a great cause, this can be used to measure system performance as well to some extent - too bad there aren't AMD cpu or Intel cpu SPECIFICALLY OPTIMIZED CLIENTS, as there is in SETI@Home - those also make projects like these run even faster/more efficiently!

Guys, if I set priority to idle does it have ANY effect on my gaming performance with it in the background? Or does it not alter it in any way and just go to sleep?

IDLE is as low a priority as you can make it - it shouldn't "seriously" interfere if it is set that way (I play Doom III &/or Quake 4 smp while it runs, they play seamlessly here on this system @ least while "FAH" runs in the background).

When you sleep though & nothing is being done interactively on your system? Heck - stop EVERYTHING you can, & let it run @ HIGH priority, done via right-clicks in taskmgr.exe while in its PROCESS tab (I even kill off the explorer.exe GUI destkop shell instance & all other trayicon + services oriented apps I can (enough to keep system online only is left))!

That makes it run better/faster/more efficiently, by setting the FAHCore_65.exe (actual unit processing client workhorse) to HIGH & FAHConsole.exe (mgt. server portion) to LOW! apk

 

[Jimmy 2004]

Just realised something strange - the time that Folding@Home reports is not actually correct - it's an hour behind the system clock. I'm guessing it could be something related to daylight savings.

 

[Alec§taar]

Just realised something strange - the time that Folding@Home reports is not actually correct - it's an hour behind the system clock. I'm guessing it could be something related to daylight savings.

Maybe they're just in a diff. timezone than yourself, all based off (iirc) "the prime meridian" & "GMT" (greenwich mean time)...



* I run into that madness quite a lot on websites for stats myself & have wondered about it also & figured it was based on WHERE they report from & what time system is used!

APK