How to use SSE with AMD!!!!!

[Kendan]

Yes, SSE will speed things up a lot I just wish they could nail down what was causing the instability and actually fix it instead of defaulting to 3dnow.

 

[Vertical_Zer0]

Excellent post Kendan. I already started noticing improvements after the 5 minute installation.

 

[ice k16v]

I downloaded the BETA client, replaced it with the 'old' one (after shutting it down), and got SSE working right away - shaved about 1min running a WU from Project343

BTW, I didn't have to delete the existing gromacs core (fahcore78) for it to work, why should it be deleted?

 

[ice k16v]

nice find kendan. got the SSE turned on and folding away. But is it ok with stanford to use the beta client in this way? And dont Beta clients expire after a certain point?

Not sure if anyone mentioned that yet, but it does not expire, see http://folding.stanford.edu/beta/325-changes.txt

Just changed the core on my girlfriend's PC (TbredB XP1800+), and it dropped from 15:48 to 13:02 per frame (p340_gnra_nat) over 20% increase! I almost couldn't believe so I checked over 3 consecutive frames and yes, SSE rocks

I just hope it keeps on folding without crashing!

PS - stupid question, does this apply to Intel CPUs as well?

 

[Kendan]

When the beta client was released sometimes it would not work right unless you deleted the core.


No, you do not need to use the beta client with Intel chips

 

[axhed]

hmmm... i was on a p347_grna_den_mg. frame times were 8:50 before the switch and 9:03 afterwards... weird.

 

[muddocktor]

Also, it seems that not all Tbreds have this problem; only a few will have problems with using SSE on Gromacs. I have 1 such proc, an XP2400 . Even at stock speeds with no overclocking whatsoever, it will lock up when using SSE instruction instead of 3DNow! instructions. However, it will fold forever while using 3DNow!, so I've upped the overclock back on that proc about as high as I want to push it to partially make up for the lost efficiency by using 3DNow!. My other Tbreds(A's and B's) don't have this problem and I run them on SSE with the -forceasm flag.

Kendan is right about them using this flag instead of making up a new one; it sucks. They should have coded a new flag in something like -forcesse when they came up with the 3.25 beta.

 

[RPM_Computing]

I went from 5min18 sec per step to 4 min 8 sec. Thanks for the boost

BTW, do the old slot a athlons support SSE?

 

[muddocktor]

I went from 5min18 sec per step to 4 min 8 sec. Thanks for the boost

BTW, do the old slot a athlons support SSE?

No, slot Athlons, Tbirds and Spitfire Durons don't support SSE, only 3DNow!.

 

[shadowdr]

With p369 my time went up from 6 to 9 minutes per frame.In some cases i guess the 3d now works better.