WOW these SSE2 double gromacs are nice now.

[Fast420A]

1700 PPW from one of the 2 instances on my P4. This make sense because they are very slow on non SSE2 machines and they were benchmarked with SSE2 off and I have it on. Wow is all I can say. 5:36 per frame on a 107 pont protein.

[Miralcos]

Is there another flag we should use for SSE2 or do the regular -forceSSE flags do the trick on p4s?

[BKPBS]

WOOT! i just received one of those p936_fkfe2_all 107 pointers!

one instance is running a p572_L939_K12M 61 pointer, after a couple frames I'll let you guys know how it does on a 2.8Ghz P4 at 3.25Ghz

[BKPBS]

Ok after 4 frames its going at 1720ppw and the second instance of the 61 pointer is at 560ppw.
Together that equals 2280PPW!!

now if i can only find a way to always get these SSE2 WU's:p

[Arkaine23]

Once you turn one in the assignment server will know you have a P4 (with SSE2) and will issue those units to you whenever it feels like it. So long as you are using -advmethods, I think that's the best you can hope for.

The PPW of the unit went up b/c Stanford benched it on a P4 2.8 with SSE2 disabled., instead of on a celeron 500.

[BKPBS]

I actually turned off -advmethods 2 days ago from what everyone was saying about the PPW increase, and I havent recieved a Tinker yet.

[wildfrogman]

wow, those ppw are amazing...looks like sse2 is stretching its legs a bit more now.

[Fast420A]

Yup, time to get more P4's with SSE2 for my farm. :D

[Mustanley]

I'm at over 1700ppw with one instance on a 2.26GHz Dell P4.

[@md0Cer]

I have a question about this, how do you get these double gromacs? Can you only get them if you have SSE2? If not would I get better PPW if I used these double gromacs on my AMD Athlon or am I better off with tinkers and normal gromacs?

[Melhisedek]

Well it doesnt pay off to be away for a long time... Plenty of stuff changed I can see. So just to be sure what flags should I have on with my P4 rigs ???
For now I have:
-local -forceasm -advmethods

only

Should I leave it way it is or change something ?

Thank you for your time!

[new_novice]

you should add -forceSSE in there too.

[Mustanley]

Now my P4-3.06GHz laptop just got one, 4:32 per frame, 2379ppw - incredible

[JetMech]

I thought -forceasm and forcesse were needed for AMD machines only. I just use the -advmethods argument on my P4 machines.

Also I've noticed there is a big hit on production if work other than folding is done while the double gromacs core is folding. Still awesome performance though.

[Arkaine23]

-forceasm or -forcesse are only needed for AMD cpu's

[nut581]

Originally posted by Arkaine23
-forceasm or -forcesse are only needed for AMD cpu's

what about on Athlon 64? I have an A64, and it does support SSE2 .. should I keep the -forceSSE?

and how do you know if a protein is SSE2 gromac?

[Arkaine23]

The core is FahCore_79.exe for the SSE2 units. Keep -advmethods and -forcesse on AMD64 cpu's.

[Melhisedek]

Originally posted by Arkaine23
The core is FahCore_79.exe for the SSE2 units. Keep -advmethods and -forcesse on AMD64 cpu's.

What about P4 CPUs ?

[OCn00b]

I've got a double gromacs on one of my AMD rigs now. I thought Stanford was checking for SSE2 enabled cpu's?

[Arkaine23]

I think P4's will do best with just

-advmethods

[SavageBasher]

The fastest time I've seen on my p936 has been 20 minutes, on my 1.2 celeron...

[Arkaine23]

Originally posted by OCn00b
I've got a double gromacs on one of my AMD rigs now. I thought Stanford was checking for SSE2 enabled cpu's?

I think their checking method can only tell if you've got SSE2 on your CPU after you have been assigned one of those units and tunred it back in. Then the unique ID number for your box (In the registry for windows, in the client.cfg for Linux) will be noted for yes/no on the SSE2 units. Even then, you still might get one once in a great while if there's nothing else available.

[Mustanley]

The SSE2 optimizations definitely make a difference.
I'm looking at 3 x p936_fkfe2_all proteins with times as follows -

8:10 per frame on a Barton @2.42GHz

6:20 per frame on a P4 @2.26GHz

4:24 per frame on mobile P4 @ 3.06 GHz

[walaka7]

it looks like dual opterons for xmas :D

[nut581]

argh .. i wanna try this dgromacs on a64, but my a64 always gets the 572 and 212 ....

[Posidon42]

argh! just as I start adding cpus to my AMD farm, the optimizations come out to make me go P4 :cry:

[Sterculus]

I got one, 1131 ppw for one instance on a 3.2ghz P4.
(p930) 91/0/31 [151]

[L337 M33P]

I just finished one

P938 I think it was - and took ages at ten minutes per frame. Definitely a performance hit with no SSE2. This is on an Athlon XP@2600.

[Sterculus]

Hmmmm, after I finished mine it gave me a tinker :/. I was hoping for another dgromac, oh well. 753ppw for 1 instance on the HT box isn't all that bad.

[Quailane]

Is their a way to keep from getting sse2 proteins? I'm afraid of getting one now!

[Arkaine23]

Originally posted by Posidon42
argh! just as I start adding cpus to my AMD farm, the optimizations come out to make me go P4 :cry:

Socket A AMD now get a almost 2x PPW boost when folding tinker WU's compared to gromacs. Tinker may be thin at the momen but removing -advmethods adn being patient should give us more than 1000 PPW even on slow Athlons and Durons.

[DayUSeX]

yeah standfords new point system is cool for sure, but now i have to relearn the point system to tell how well my system is doing. still those double gromacs have a weird tendancy to not use the full CPU potential at least accord to task maanger.

[BKPBS]

Do you guys think that Stanford will change the point values because of the big difference in PPW for P4's and AMD's?
My P4@3.25Ghz is producing 3,000PPW by itself, and my other 2 AMDs are only producing ~700ppw each. If this difference keeps up I'll seriously consider replacing the AMDs with P4 chips for the performance and PPW increase.

= ~9,000PPW for me soon! (was only producing around 1900 at peak power)

[overdoze]

I'm getting 1500PPW with Dgromacs core on my barton at 2.4GHz.

[Fast420A]

I get about 1100 PPW at the same clock speed on a 256KB Cache Tbred to your 1500 PPW on a 512KB Cache Barton. I think more tests are in order on cache size speed changes. Like the 2.8C vs. 2.8E or a 3.2C vs. 3.2E vs. 3.2EE!

[gustav]

alright, i have two consoles folding on my p4.....i use EMIII and set the flag -advmethods, what does -local do? also, would there be any advantage using WEDO's one click thing versus my current two consoles with -advmethods?