DF Protein Changeover

[bfour]

Looks like the changeover is starting, tried to refresh the DF team folding statistics, and looks like the server is down!

 

[FoBoT]

first person to get the new one installed be sure and post up a quick benchie, so we can know how fast/slow the next one is

i don't recall seeing any hints from howard about the new protein

 

[CrazySaint]

It appears that I *might* have the new client, in which case its about 30% slower than the old one, but that's still about 50% faster than the one before that

 

[m0ti]

Any word on protein size?

 

[FoBoT]

what AA?
is the client you are looking at autoupdate? or did you download it?
how big is the .zip file?
i think i am getting the new file from the ftp site, it is under 7 mb, that small of a file is usually indicative of a smaller protein

hmmm... guess i should just wait/be patient instead of guessing

 

[FoBoT]

got it!!!!!!!!!!!!!


108AA

running a preliminary benchmark now

i can't find my bench from the last protein for comparison

but i found one from the time before last protein (October, the last "slow" one)

new protein

------------------------------------------------------------
Distributed Folding Windows dfGUI v2.0 Benchmark

Sample Size : 1190 structures over 590 seconds.
Protein Size: 108AA

Structures Per Second: 2.02
Structures Per Minute: 121.0
Structures Per Hour : 7261
Structures Per Day : 174264

OS : Windows XP MHz: 1063
CPU: Intel(R) Pentium(R) III Mobile CPU 1066MHz
Client Switches: -df -i f -rt
------------------------------------------------------------

old old protein (October)

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 3610 structures over 3951 seconds.
Protein Size: 129AA

Structures Per Second: 0.91
Structures Per Minute: 54.8
Structures Per Hour : 3289
Structures Per Day : 78943

OS : Windows XP MHz: 1063
CPU: Intel(R) Pentium(R) III Mobile CPU 1066MHz
Client Switches: -df -qt -rt -s 3000
------------------------------------------------------------

so this new one should still be considered "Fast" i believe, lets CRUNCH!!!!

 

[CrazySaint]

I dunno what the deal was earlier, but I check again and its a bit faster than the last protein, so things are looking good

 

[Insidious]

I didn't see this thread until now....

Originally posted by: Insidious
------------------------------------------------------------
Distributed Folding Windows dfGUI v1.9 Benchmark

Sample Size : 15660 structures over 4340 seconds.
Protein Size: 108AA

Structures Per Second: 3.61
Structures Per Minute: 216.5
Structures Per Hour : 12990
Structures Per Day : 311757

OS : Windows XP MHz: 1761
CPU: AMD Athlon(tm) XP 2100+
Client Switches: -rt -g 15 -s 999
------------------------------------------------------------

This machine was a pretty consistant 3.3+ f/s on the last protien.
My other comp is slightly faster (1775MHz), but has been comandeered for BF1942 at the moment!

Not too much difference here, but that probably stands to reason after the algorithm update and similar protien size.

-Sid

 

[Terrahawk]

Just updated about half of my machines so far... it does appear to be a bit faster. I'll be doin' my servers tonight

 

[bfour]

Not sure if I like going back to a smaller protein, but if that is what howards wants....

 

[CrazySaint]

Originally posted by: bfour
Not sure if I like going back to a smaller protein, but if that is what howards wants....

I like it because we can gain on slow teams faster and move up in the standings!

 

[Robor]

Okay, Okay... I don't want to look like a total dumba$$ on the project but... What are ya'll talking about? How do you know about as far as the "size" or "speed" of the new protein? Do you base this on RAM usage (with or without "useram=1+) or do you wait to gauge how the client runs under the new protein? How can you tell if a one is slower or faster.

Next question... What would benefit us more?

PS - Just got home from watching the (home team) Lightning beat the Red Wings! WOOHOO!! If you've never watched hockey in person you're missing out. Oh, free 2nd row seats didn't hurt either.

 

[FoBoT]

the AA is determined by using dfGUI, there is a file somewhere that contains the info also, but i forget which one, dfGUI is the "lazy" way out

also, the fastnest is quickly gauged by running the benchmark feature of dfGUI or you could just start DF manually and watch the clock

dfGUI can get a pretty close estimation after running the benchmark for just a portion of an hour, 10-15 minutes will get a close mark

as to which on will benefit us more, it all comes down to participation

if our team has higher cpu on the faster protein, we will catch teams leading us and widen the gap between those behind us , if our current protein participation is higher than those other teams

 

[m0ti]

Downloading now, actually getting decent speeds (~28k/sec).

Ok installed it.

Looking like it's marginally faster here (less than +10% for sure).

 

[bunker]

Originally posted by: Robor
PS - Just got home from watching the (home team) Lightning beat the Red Wings! WOOHOO!! If you've never watched hockey in person you're missing out. Oh, free 2nd row seats didn't hurt either.

Enjoy it while you can! That won't be happening again!

<----Huge Wings fan

 

[Insidious]

I remember when I got to see my first hockey game in person. It was a minor league game in central MI during a skiing trip.

It was A LOT of fun.

 

[Robor]

Originally posted by: bunker

Originally posted by: Robor
PS - Just got home from watching the (home team) Lightning beat the Red Wings! WOOHOO!! If you've never watched hockey in person you're missing out. Oh, free 2nd row seats didn't hurt either.

Enjoy it while you can! That won't be happening again!

<----Huge Wings fan

Hehe... I know. The Bolts needed that game badly though. To be honest neither team deserved to lose the game. It was excellent. Lots of shots, great saves, and action. Oh, and for the most part the refs let them play. I hate games filled with penalties and stopages.

 

[bunker]

Holy Shnikes!

I just checked the benchmark on the new protien on my old PIII600!

OLD PROTIEN:

Sample Size : 21800 structures over 39124 seconds.
Protein Size: 129AA

Structures Per Second: 0.56
Structures Per Minute: 33.5
Structures Per Hour : 2010
Structures Per Day : 48235

NEW PROTIEN:

Sample Size : 54700 structures over 40711 seconds.
Protein Size: 108AA

Structures Per Second: 1.34
Structures Per Minute: 80.6
Structures Per Hour : 4838
Structures Per Day : 116123

I'd say that's a nice hefty boost!

 

[m0ti]

Nice boost for sure.

I may be showing a 10%-15% boost overall, too.

I think it's just due to the smaller protein size (I don't think Howard made any updates to the client).

 

[Insidious]

yeah, the client updates were on the last changeover.

I finally got benchies over a decent period of time with un-disturbed machines:

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.9 Benchmark

Sample Size : 126960 structures over 33719 seconds.
Protein Size: 108AA

Structures Per Second: 3.77
Structures Per Minute: 225.9
Structures Per Hour : 13555
Structures Per Day : 325316

OS : Windows XP MHz: 1775
CPU: AMD Athlon(tm) XP 2100+
------------------------------------------------------------

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.9 Benchmark

Sample Size : 124605 structures over 33718 seconds.
Protein Size: 108AA

Structures Per Second: 3.70
Structures Per Minute: 221.7
Structures Per Hour : 13304
Structures Per Day : 319292

OS : Windows XP MHz: 1761
CPU: AMD Athlon(tm) XP 2100+
------------------------------------------------------------

(a little faster than with the 129AA protien size, but not too much.)

 

[m0ti]

I'm showing a fairly stable 3.5 (up to 3.55), compared with a 3.1-3.2 on the prior protein, so a little over 10%

 

[Terrahawk]

I found an odd thing the other day - DF under WinXP is slower than under Win98. Under 98, I get 3.6 folds/sec for an Athlon XP 1600, and 3.2 folds/sec under XP. Anyone else noticed this?

 

[FoBoT]

Originally posted by: Terrahawk
I found an odd thing the other day - DF under WinXP is slower than under Win98.

yes, this is true

on some PC's it could be related to the amount of RAM used by the OS , but i think XP is slower anyway

i run win95 on some lab PC's just to get the extra speed boost

also, i don't have hard numbers (yet) , but i believe that w2k falls in between win9X and XP

the percentage is about (ok, i am guessing here) , win9x ~10% faster than XP and ~5% faster than W2K , it might be a little less than 10%/5% though

 

[Insidious]

I don't have an 'apples to apples' comparison available, but I think what you are seeing sounds right.

If I remember when XP was first comming out, a lot of people found it to be a little slower than 98se.

There is stuff you can do (kill unnecessary processes etc.) to speed it up, but IMO it's not worth it.

My own personal rationalization was that the gains I made in stability and useability were worth the

difference in speed.

( did compare gaming performance when I was first making the switch and I wasn't able to measure a

change there. )

-Sid

 

[Robor]

I'm with Sid... I'd much rather have a Windows NT/2K/XP system that's a little slower but more stable than a slightly faster Win9x system that is more prone to problems. Plus, on NT/2K/XP you can install it as a service so it's still running even when a user isn't logged in.