To SSE or to not SSE that's the question

[sforza]

Hello everyone, I'm at Team Short-Media folding again with you :D

Just wanted to share with you my experience with a gromacs WU: I started folding it at an athlon classic 750 (no SSE at the processor and no flags or special tricks but it got a gro)

It was very slow but oh well what could I do...nothing but wait. Then I moved it to a celeron tualatin 1Ghz.

Look at the log:

[08:52:28] - Ask before connecting: No
[08:52:28] - Use IE connection settings: Yes
[08:52:28] - User name: sforza (Team 93)
[08:52:28] - User ID = 254E79530B68554C
[08:52:28] - Machine ID: 1
[08:52:28]
[08:52:28] Loaded queue successfully.
[08:52:28] + Benchmarking ...
[08:52:30]
[08:52:30] + Processing work unit
[08:52:30] Core required: FahCore_78.exe
[08:52:30] Core found.
[08:52:30] Working on Unit 00 [June 13 08:52:30]
[08:52:30] + Working ...
[08:52:31]
[08:52:31] *------------------------------*
[08:52:31] Folding@home Gromacs Core
[08:52:31] Version 1.48 (May 7, 2003)
[08:52:31]
[08:52:31] Preparing to commence simulation
[08:52:31] - Ensuring status. Please wait.
[08:52:48] - Looking at optimizations...
[08:52:48] - Working with standard loops on this execution.
[08:52:48] - Previous termination of core was improper.
[08:52:48] - Files status OK
[08:52:49] - Expanded 363526 -> 1770209 (decompressed 486.9 percent)
[08:52:49]
[08:52:49] Project: 343 (Run 16, Clone 41, Gen 3)
[08:52:49]
[08:52:49] Entering M.D.
[08:53:10] (Starting from checkpoint)
[08:53:10] Protein: p343_gnra_pf2
[08:53:10]
[08:53:10] Writing local files
[08:53:49] Completed 340000 out of 500000 steps (68)
[10:22:19] Writing local files
[10:22:19] Completed 345000 out of 500000 steps (69)

Folding@home Client Shutdown.

1h25min aprox/frame



Then I restarted it again, still at the tualeron:

[10:23:46] - Ask before connecting: No
[10:23:46] - Use IE connection settings: Yes
[10:23:46] - User name: sforza (Team 93)
[10:23:46] - User ID = 254E79530B68554C
[10:23:46] - Machine ID: 1
[10:23:46]
[10:23:46] Loaded queue successfully.
[10:23:46] + Benchmarking ...
[10:23:49]
[10:23:49] + Processing work unit
[10:23:49] Core required: FahCore_78.exe
[10:23:49] Core found.
[10:23:49] Working on Unit 00 [June 13 10:23:49]
[10:23:49] + Working ...
[10:23:49]
[10:23:49] *------------------------------*
[10:23:49] Folding@home Gromacs Core
[10:23:49] Version 1.48 (May 7, 2003)
[10:23:49]
[10:23:49] Preparing to commence simulation
[10:23:49] - Looking at optimizations...
[10:23:49] - Files status OK
[10:23:49] - Expanded 363526 -> 1770209 (decompressed 486.9 percent)
[10:23:50]
[10:23:50] Project: 343 (Run 16, Clone 41, Gen 3)
[10:23:50]
[10:23:50] Assembly optimizations on if available.
[10:23:50] Entering M.D.
[10:24:10] (Starting from checkpoint)
[10:24:10] Protein: p343_gnra_pf2
[10:24:10]
[10:24:10] Writing local files
[10:24:46] Completed 345000 out of 500000 steps (69)
[10:24:46] Extra SSE boost OK.
[10:47:55] Writing local files
[10:47:55] Completed 350000 out of 500000 steps (70)
[11:11:01] Writing local files
[11:11:01] Completed 355000 out of 500000 steps (71)
[11:34:04] Writing local files
[11:34:04] Completed 360000 out of 500000 steps (72)
[12:00:16] Writing local files

23min.

SSE matters!! :banghead: :o


Is there any way to make the athlon classic to always get Tinkers?

Thanks in advance :)

[csimon]

just leave off -advmethods flag and delete the fahcore_78 to increase your chances

[mmonnin]

That wont ensure tinkers cause gromacs are coming out to the general public. I got one on my P2 but I dont think its gonna make it.

[sforza]

No flags are enabled. What mmonnin says it's the explanation, I believe. If I got a gromacs WU even not asking for it (flags) I don't think deleting the core_78 will make a difference. It would just download it again.

I can manually copy all the gromacs I get on the athlon, to the celeron, but it's a pain...

[mmonnin]

You can set it to get genomes if you absolutely dont want gromacs. I have one on the P2 now and its going at 30min per 10% of a frame and there are 30 frames so to speak. Its worth 20 points.

[a2jfreak]

How do you know how many points a particular WU is worth?

mmonnin said
You can set it to get genomes if you absolutely dont want gromacs. I have one on the P2 now and its going at 30min per 10% of a frame and there are 30 frames so to speak. Its worth 20 points.

[mmonnin]

http://folding.stanford.edu/psummary.html

Those are the FAH ones. Here is the forumla if you get a genome and are on a FAH team.

points = (chainlength*chainlength*.0025 +chainlength*.05) *(5/12)

A WU with a chainlength of 129 is worth 20 points then.

[a2jfreak]

I normally see xxx/400 or xxx/500 when I fold, so are those chainlengths of 400 and 500? or is that different? If different, where do I find the chainlength?

Has the formula been the same for the life of FAH, or has it recently changed? I know I used to make considerably fewer points for considerably more WUs.

mmonnin said
http://folding.stanford.edu/psummary.html

Those are the FAH ones. Here is the forumla if you get a genome and are on a FAH team.

points = (chainlength*chainlength*.0025 +chainlength*.05) *(5/12)

A WU with a chainlength of 129 is worth 20 points then.

[mmonnin]

No you have a tinker WU. It uses core_65. It doesnt use 3dnow! or SSE or any of the other optimizations that gromacs do. Gromacs use core_78.

Genomes is a different project. GAH compared to FAH with tinker and gromacs.

The chainlength is printed in the log file if you do get a genome WU.

The point system has changed for genomes. They used to be worth very little even compared to tinker WUs but now they should be pretty even.

[a2jfreak]

Well, I want to use 3DNow!/SSE/SSE2 when I can, so what do I do to make sure I don't get a (s)tinker?

mmonnin said
No you have a tinker WU. It uses core_65. It doesnt use 3dnow! or SSE or any of the other optimizations that gromacs do. Gromacs use core_78.

Genomes is a different project. GAH compared to FAH with tinker and gromacs.

The chainlength is printed in the log file if you do get a genome WU.

The point system has changed for genomes. They used to be worth very little even compared to tinker WUs but now they should be pretty even.

[mmonnin]

Hehe.

Open the client with a shortcut with -advmethods on the end of the target. If its an AMD then you must get the 3.25 client and use -forceasm as well.

C:\FAH directory\fah.exe -advmethods -forceasm or
"C:\FAH directory\fah.exe" -advmethods -forceasm

[a2jfreak]

Thanks!

mmonnin said
Hehe.

Open the client with a shortcut with -advmethods on the end of the target. If its an AMD then you must get the 3.25 client and use -forceasm as well.

C:\FAH directory\fah.exe -advmethods -forceasm or
"C:\FAH directory\fah.exe" -advmethods -forceasm