Folding@Home

blakerwry

Storage? I am Storage!
Joined
Oct 12, 2002
Messages
4,203
Location
Kansas City, USA
Website
justblake.com
hmm.. for some reason the stats page shows me as only having 1 active CPU.. when I have 2 folding full time... i checked my config and everything should be correct... bah!
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
I'm sure you've all noticed but...

Extremeo/c's Folding at home stats pages have been down for a while. I think it's been four days since last time I've benn abl to access it. Statsman down, ExO/C down... We're running out of statisticians.

Statistically speaking, what's our chance to find a reliable F@H statistic system? So far, numbers aren't on our side.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
May I suggest that people with AMD processors that support SSE start using the beta 4.0 clients!

Windows console client 4.0 beta
Windows Graphical Client 4.0 beta
Linux Console Client 4.0 beta

You then need to remove the core files (65, 78, ...) from your Folding@home Directory/folder. Don't worry they will automaticly be re-downloaded to new versions when you start up FAH.

Then you can use the -forceSSE option. Your gromacs WU's will run aprox. 30% faster with SSE than 3d-NOW.
 

blakerwry

Storage? I am Storage!
Joined
Oct 12, 2002
Messages
4,203
Location
Kansas City, USA
Website
justblake.com
does that mean Palmonio and up? I was thinking the thunderbird and duron upto 1.3GHz didn't support SSE.... what about pIII CPU's, I assume the client has SSE turned on for them by default.. correct?
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
blakerwry said:
does that mean Palmonio and up? I was thinking the thunderbird and duron upto 1.3GHz didn't support SSE.... what about pIII CPU's, I assume the client has SSE turned on for them by default.. correct?

If the processor does not support SSE then this is obviously a bad idea. Early AMD processors like the thunderbird and early durons don't have SSE support. Intel processors have SSE support on by default, so you don't need to do any changing for them.
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
P5-133XL said:
May I suggest that people with AMD processors that support SSE start using the beta 4.0 clients!

[...]

Then you can use the -forceSSE option. Your gromacs WU's will run aprox. 30% faster with SSE than 3d-NOW.
Well, so far, on an Athlon XP 2400+ and comparing with the FAH3 core, the boost is far more spectacular than just 30%. On a unit (called p1032_A21nat_337_94) that took 30 minutes per frame (8 had already been processed when I changed the core), FAH4-beta with -forceSSE now brought it down to ...13 minutes. It's too good to be true, yet I can't explain it any other way than just by the FAH core upgrade. No other application has been shut down, nothing else has changed except the core.

Anyone has reports about the latest core's performance under Linux? My not-so-little self wants to know if it's as bad as it was before compared to Windows' core performances. This is a major obstacle for me to fully convert to Linux.
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
And from 28 minutes per frame down to a mere 10 minutes for a p572_L939_K12M protein. Simply incredible.

I've nailed the cause to the -forceSSE additional command, because the FAH4 core alone doesn't apply the SSE optimizations to an SSE-capable Athlon XP processor. You must tell it to apply them or it simply won't use them.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
The 30% gain is from 3d-now to SSE: I'm trying to be conservative and not exadurate. from no optimizations to SSE the gain is closer to 300%
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
I'm using firedaemon and I added the "-forceSSE" into the parameters field. So now it reads "-service -forceSSE".

Is anyone else doing it this way, and/or am I doing it incorrectly?

In the log file it says:

Arguments: -service -forceSSE

Warning:
By using the -forceSSE flag, you are overriding program
safeguards that monitor the stability of SSE
instructions on your system. If you did not intend
to do this, please restart the program without
-forceSSE. If work units are not completing fully,
then please discontinue use of the flag.

Then at the bottom of the log it says:

[23:57:32] Writing local files
[23:57:32] Completed 425000 out of 500000 steps (85)
[23:57:33] Extra SSE boost OK.

BTW, Thanks Mark for the info on the new core.
 

NRG = mc²

Storage is cool
Joined
Jan 15, 2002
Messages
901
My new CPU is here tommorow, XP2400 thoroughbred, to replace my old 1800 palomino I've had for close on 2 years now. Should give a nice boost, I would have gone with something better like a barton but my memory is 2100 and cannot afford to replace it right now (the CPU was actually free for building someone a PC so I can't complain)
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
Handruin,

You are doing it correct.

JoJo,

Sorry that this is so. It is a beta and thereby bugs can and do exist. You might try -verbosity 9 and then look at the log file to see what the problem is. Do note that overclocking is not recomended with the -forceSSE or the -forceasm options: If your CPU is flakey then it usually causes problems though they normally only result in lots of corrupted WU's.
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
Same for me with F'core_65. I wanted to test the relative speed of the new client under Linux but I stumbled on a unit processing with the core_65. It took forever for each frame (ok, just 6m45s, but it was a 400 frames unit), so I restarted the system under Windows.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
The Tinker core (65) does not use any optimizations and thereby will be very slow relative to Gromacs core (78). Remember that there is a 300% improvement between no optimizations and SSE optimizations.

One could learn to hate Tinker givin that info. It is my understanding that FAH will eventually be abandoning Tinker in favor of Gromacs. But untill then ...
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
Ah ben Tabarnack. He clinched me before the end of November. I guess I'm not the only one for whom the new beta client makes miracle.

I probably won't boost my crunching power until March or April next year. You better put this time to profit, because my next upgrade won't be a half-hearted effort. We'll meat again (or so do I hope).
 

Pradeep

Storage? I am Storage!
Joined
Jan 21, 2002
Messages
3,845
Location
Runny glass
If you get a Tinker unit can you just delete and dl a new unit, till you get one that uses Gromacs? Or have they thought of that?
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
You could delete everything except the client and the FAH_78 core and it would ask you again all your info and download a new unit next time you fire it up. It's not very ethical, but then again, it's part their problem if they still ship units not processed through the newer and optimized core.
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
Since our weekly production hasn't improved much despite the fact that mine doubled and that Mark is now between 3500-4000 points/week, I figure not many among you use the new beta client with the -forceSSE option. I'm sure we could hit 15000 points/week if most of you would use the more optimized client.

About the crashing complains regarding the beta clients, I've found a few things that seem to fix the problem :
  • Disable screen savers. Set the monitor to shut down after 10 or 15 minutes instead.

    System Properties - Advanced - Performance Options - Applications optimized for : Background Services. One of my computers stopped crashing after I did that.

    Don't ever click with the mouse cursor inside the F@h command line window. Two days ago, I lost a 70 points unit (many, many crunching hours) with 3 minutes to go on the 99th frame this way. Fortunately, the resulting enclosure reshape didn't disabled any system component.
The console client shouldn't crash often if you do the above.
 

i

Wannabe Storage Freak
Joined
Feb 10, 2002
Messages
1,080
In an effort to warm up my kitchen a little, I started up the folding@home client on one of my processors. After running all night, it just got to this point:

Code:
...
[15:18:56]  Iterations: 6000 of 6000
[15:18:57]  Finished
[15:18:57]  [SPA] Design complete
[15:18:57]  [SPA] completed successfully
[15:18:58]
[15:18:58]  Finished Work Unit
[15:18:58]  work_hrd.len_arcfile:  48600
[15:18:59]  Leaving Run
[15:19:03]  - Writing 49112 bytes of core data to disk.
[15:19:03]    end (WriteWorkResults)
[15:19:03]  - Shutting down core
[15:19:03]
[15:19:03]  Genome@Home2 Core Shutdown: FINISHED_UNIT
CoreStatus = 64 (100)
Sending work to server

+ Attempting to send results
- Connecting to server (171.64.122.125:8080)
+ Results successfully sent
+ Number of Units Completed: 3
+ Attempting to get work packet
- Connecting to assignment server
Segmentation fault
 

i

Wannabe Storage Freak
Joined
Feb 10, 2002
Messages
1,080
Hmm. Having read through the log file, the work unit I completed this morning was originally downloaded on November 11th.

That's November 11th ... 2002.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
I only installed the latest client on my newest machine because my older Athlon doesn't have SSE. (1.2 GHz t-bird)
 

Fushigi

Storage Is My Life
Joined
Jan 23, 2002
Messages
2,890
Location
Illinois, USA
Since I've only 1 Athlon XP-class system in my 'farm', the SSE boost isn't doing wonders for my point production. Also, it is still pulling down core 65 units and I'm not deleting them in the hope of getting a core 78 unit. It produces what it produces...

In other sucky news, I found one of the machines, for some reason, had 2 instances of F@H running using the same data set. Don't know how it happened nor how long it had been like that; possibly for a week or more. So it's likely I've returned some corrupted data sets recently and may have lost points there as well.
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
I was about to write it. Glad they're back online, I missed their graphs. Dyyryath's tables are ok, but I much prefer ExOC's graphs.
 

Pradeep

Storage? I am Storage!
Joined
Jan 21, 2002
Messages
3,845
Location
Runny glass
I wish they would add a flag so it would flush the already done units at startup. Otherwise if by some freak modem bullshit my upload is interrupted (most of the time on this POS 26K connection), it won't upload again till the next unit is complete.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
I've put on the flag -verbosity 9 previously and found that it does flush it queue of already done units at startup as well as several times in the middle of each WU.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
This past week I've been experimenting with UD_MON. It is a program that caches WU's. That is why I have 23 processors listed when I only have 7 machines with 8 processors. I understand that F@H dissaproves of caching because of the problems with older machines not being able to finish their WU's in time. That is not a problem with my machines: they are all fast enough that even with in the worst case senario they can easily finish all cached WU's.

However, along with caching goes the ability to remove Tinkers before I've used up much CPU cycles on them. The result appears to have gained another 15% increase in production at the cost of spending some time daily cleaning out undesirable WU's (annoying and inconvienant).

On the ethics of the situation: If F@H doesn't like this behavior then they need to fix the scoring of WU's or better yet remove all tinkers, so I don't need to play such games to optimize production. I'm simply optimizing my production according to their scoring system.

If people are interested, I'll give instructions on how to operate UD_MON.
 

CougTek

Hairy Aussie
Joined
Jan 21, 2002
Messages
8,728
Location
Québec, Québec
It looks like a nice little utility, but I already put a tad too much time and efforts for FAH and I know that if I start messing with this, I'll be even more obsessed than I am currently with the FAH stats, which I don't believe would be a good thing for me.
 
Top