Tinkers...

tefleming

I thought they were phasing Tinkers out (like a while back) in favor of Gromacs? I'm not complaining, my past coupla wu's have been ~70 pt Tinkers...

//EDIT:
Have I neglected some flags?? (specs below)

a2jfreak

(whatever you use, console or gui--don't know about screensaver--I will call fah.exe)

client 3.x
fah.exe -advmethods -forceasm

client 4.x
fah.exe -advmethods -forcesse

I used this utility to setup fah as a service.

I issued:
service.exe -cpu=1 -params="-advmethods -forcesse -forceasm" -noconfig
I had previously setup my client.cfg file and had it in the directory that service.exe resided.

-advmethods will help you get gromacs.
-forcesse (v4.0 client) will enable SSE optimizations
-forceasm (v3.x and 4.0) will enable optimized assembly

When you get a gromacs core (core_78) double check your log and make sure it is v1.56 and not a previous version. You should get v1.56 (I did twice today when setting things up).

edcentric

There are some of the projects that require tinkers for the calculations. When they need to follow up on one of those you will get tinkers. Switches don't matter if that is all that they are assigning.

mmonnin

a2j: I edited your post to not confuse people. -forcesse overrides -forceasm so it doesnt need to be there. You cant use SSE and 3dnow! at the same time.

Look in your log file for the performance fraction after each WU is done. If its less than 0.8 then you will get tinkers as they want the faster machines to do gromacs.

profdlp

mmonnin wrote:

...Look in your log file for the performance fraction after each WU is done...

I must be blind, I can't find this anywhere in my log file.

mmonnin

You may need to have the verbosity up. Add "-verbosity 9" to your list of flags. The default is 6 so 9 should show more in your log file.

Should look like this

[00:50:41] Unit 9 finished with 96 percent of time to deadline remaining.
[00:50:41] Updated performance fraction: 0.978482

Straight_Man

mmonnin wrote:

a2j: I edited your post to not confuse people. -forcesse overrides -forceasm so it doesnt need to be there. You cant use SSE and 3dnow! at the same time.

Look in your log file for the performance fraction after each WU is done. If its less than 0.8 then you will get tinkers as they want the faster machines to do gromacs.

On my Barton 2500+ box, I use BOTH, and get WUs out in about .89 times (average) the time they take with just -forceasm (just this switch gets better results than just -forcesse) or just -forcesse. The combined tunings DO work with the Barton. Older AMD CPUs will not get that gain, but the Barton 2500+ DOES get the gain. P4's do not like the -forceasm switch, they have no real 3DNow tuning in them. PIII's do not like the -forceasm switch either.

Two notes:

This only works with Client 4.0 and Core_78 1.54 or 1.55 active. The Barton is minorly OC'd, the RAM is unremarkable DDR333 (Corsair CMX).

Hypothesis-- Did I find an exception??? Yes. The Barton does not use pure 3DNow in the older sense of that term. It uses an enhanced 3DNow. What -forceasm really does is optimize SSE to 3Dnow translation, it optimizes the 3Dnow code sent to the CPU core. But, the enhanced 3DNow can and does handle code that is tuned SSE fairly well when the code is 3DNow tuned also, and does so better than earlier CPUs. Look at the detailed Barton specs, it uses 3DNow+. -forceasm is acting as a translator that the Barton likes, but pretuned SSE code (using the 3DNow optimizations) is easier for the Barton to process.

John D.

mmonnin

-forcesse overrides your -forceasm. -forceasm is slower than -forcesse thus the increase. So using both would be just like using forcesse alone.

Your hypothesis is wrong.

a2jfreak

The log says:

# Windows Console Edition #####################################################
###############################################################################

Folding@home Client Version 4.00

http://folding.stanford.edu

###############################################################################
###############################################################################

Arguments: -service -advmethods -forcesse -forceasm

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.

Warning:
By using the -forceasm flag, you are overriding
safeguards in the program. If you did not intend to
do this, please restart the program without -forceasm.
If work units are not completing fully (and particularly
if your machine is overclocked), then please discontinue
use of the flag.

[02:31:28] - Ask before connecting: No
[02:31:28] - User name: a2jfreak (Team 93)
[02:31:28] - User ID = 6791CEEA1DADC102
[02:31:28] - Machine ID: 1
[02:31:28]

Looked to me like it recognized both, but whatever. I didn't see any documentation that said not to use them together (or actually, any documentation period). I checked the LOG file for the performance fraction and found no match. A gromac has finished with core 1.56 so I should have had a performance fraction printed in the log file, correct?

// Edit: N/M. I guess I'll up the verbosity to 9

csimon

run it for a few minutes both ways and see which way is faster ...using a gromacs wu of course.

profdlp

mmonnin wrote:

You may need to have the verbosity up...

I added it. Guess I'll have to wait for the next WU to finish to see it.

Thanks.