AOA Forums AOA Forums AOA Forums Folding For Team 45 AOA Files Home Front Page Become an AOA Subscriber! UserCP Calendar Memberlist FAQ Search Forum Home


Go Back   AOA Forums > General > ThunderRd's AOA FOLDING@HOME Team

ThunderRd's AOA FOLDING@HOME Team Where Protein Acrobats gather to change the world!


Reply
 
LinkBack Thread Tools Rate Thread
  #1 (permalink)  
Old 17th February, 2002, 03:47 PM
Lionfish's Avatar
Member
 
Join Date: September 2001
Location: So Cal, near Robbie
Posts: 2,767

Thoughts on Increasing speed

From: "pandegroup" <pandegroup@y...>
Date: Sat Feb 16, 2002 1:09 pm
Subject: Thoughts on Increasing speed



Michael from F@H... here,

So, a few comments. For compilers, we're using intel fortran for both Windows
and Linux (for our work, I've been timing the intel faster than g77 0-30%
depending on the identity of the work unit). Last I checked (I haven't looked
at it in a while) Windows was about 20-30% faster than Linux - maybe that's
changed since we've changed the work units a bit.

For Macs, we've found that g77 is really poor, and we're using f2c, and then
compiling in C. That makes things work better, but it's still a bit slower - I
haven't been doing the tests, so I don't know the exact numbers - I'll try to
find out, it's about time to do an extensive speed comparison test again.

We've experimented with vectorizing the code using AltiVec and SSE/SSE2 and
OpenMP, but we have not been able to get any appreiciable speedup. First, we
use double precision floats, which parallelize much worse than single precision
floats (SSE, to my understanding, doesn't work with doubles at all, SSE2 can
only parallelize 2 doubles, as opposed to 4 floats - and I don't know much
about AltiVec). Second, the code does not lend itself to parallelization very
well - it's hard to get the conditionals out of the inner loops.

A single simulation is not as easily parallelizable as one might think. True,
you can divide up the atoms into seperate groups - but what counts is the
forces that are BETWEEN all the atoms - so the atoms need to communicate with
each other. There are clever ways of doing this, but it takes very fast
communication, and parallelizes to max 10-20 processors in a supercomputer -
but supercomputers are expensive to use, and not very common.

Since protein folding at any given time depends on what happened at previous
times, it's not trivially parallelizable beyond that. One of the breakthroughs
of F@H... is that our algorithms recognize that in many situations, you can get
folded trajectories by running many parallel simulations. That's the
breakthrough that makes protein folding work for distributed computing, but
it's still more complicated than just a bunch of batch jobs.

Cheers,
Michael Shirts
Folding@H... Team
Stanford University
__________________
Ep0x 8K3A, XP1700+ T-bred JIUHB-(310)XPMW,DLT3C,189x12.5=2.36
1 gig mb PC 3200 Corsair .
Swiftech H20 setup with MCW372 block.
eVGA 6800 GT.
350 w Sparkle PS.

XFIRE= rottys
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #2 (permalink)  
Old 17th February, 2002, 05:16 PM
Allan's Avatar
Member/Contributer
 
Join Date: September 2001
Location: Denmark
Posts: 7,914

Thanks for the headsup!
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
Reply



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Increasing vulerabilities we should be mindful of... Daniel ~ Data Security 0 15th September, 2006 07:33 PM
Increasing the Mulitpliers DKMOC AMD Motherboards & CPUs 3 10th May, 2005 10:34 PM
Can increasing the FSB do damage... rocklobster EPoX MotherBoards 5 15th July, 2003 04:50 AM
Increasing Desk Capacity Spode Random Nonsense! 15 27th November, 2002 10:02 PM
increasing comfort would be.... Max193 Forum Feedback and Suggestion Box 4 21st July, 2002 06:47 PM


All times are GMT +1. The time now is 11:29 PM.


Copyright ©2001 - 2010, AOA Forums
Don't Click Here Don't Click Here Either

Search Engine Friendly URLs by vBSEO 3.3.0