MrBayes

Hello,

I suspect that I'm doing something stupid, but I can't get a recent
version (r268) of MrBayes 3.2 to restart from a checkpoint file.  I
run the initial run with something like:

mcmcp ngen=20000000 Nruns=2 nchains=8 samplefreq=4000 printfreq=4000
temp=0.005 Checkfreq=1000 checkpoint=yes;

and I restart with

mcmcp ngen=20000000 Nruns=2 nchains=8 samplefreq=4000 printfreq=4000
temp=0.005 Checkfreq=1000 checkpoint=yes Append=yes;

but the code errors out with:

      ...
      Successfully read tree 'mcmc.tree_16'
   Exiting trees block
   Reading mrbayes block
      Topology of user tree 'mcmc.tree_1' wrong in setting parameter 'V{all}'
      There was an error on at least one processor
      The error occurred when reading char. 15 on line 380
         in the file 'finalV3_2.nex.ckp'

   Returning execution to calling file ...

      Error in command "Mcmc"
      There was an error on at least one processor
      The error occurred when reading char. 15 on line 12957
         in the file 'finalV3_2.nex'

Line 380 in finalV3_2.nex.ckp is:

                V{all}(1,2)=mcmc.tree_1


Any advice is much appreciated!

Thanks,
Chris

http://fylb.free.fr/go.php

Maxim,

I've been playing around with the latest mrbayes 3.2, and I noticed it now has beagle/GPU support.  I have tried to enable the GPU with "set beagledevice=gpu", but when I run the chain it says only the CPU is working.  I do indeed have a working GPU -- I can successfully compile and run my own programs on the GPU.  Is there some hidden trick to get mrbayes3.2 to use the GPU, or is this just not ready yet?

Thanks,

Douglas

On Oct 14, 2010, at 9:56 AM, Maxim Teslenko wrote:

> Unfortunately, I can not give you any advice on the specific card to use since I have not done any comparison of different cards myself. People who develop beagle library would be able to give better advise on this topic http://code.google.com/p/beagle-lib/. General advise would be to get as many GPUs with as many cores on each of them as possible. CPU performance would be not important for most of the analyses where most of the computation is spent on likelihood computation. Memory bandwidth could be an important factor to look at.  By the way OpenCL will be supported by beagle library too, at least that the plan!
> 
> Maxim
> 
> -----Original Message-----
> From: Andreas Wallberg [mailto:andreas.wallberg@...] 
> Sent: Thursday, October 14, 2010 2:55 PM
> To: Maxim Teslenko
> Cc: mrbayes-users@...
> Subject: Re: [Mrbayes-users] <no subject>
> 
> Hi all!
> 
> @Maxim: great! I've been wishing for GPU computations for a long time
> now. Any recommendations of card models and OS platforms for maximum
> compatibility and performance? Any plans for OpenCL?
> 
> @Robert: the last couple of years I have had pretty good performance
> with quad-core machines, running either AMD or Intel CPUs, and in my
> experience, the raw clock frequency of the CPU (GHz) has been more
> important than RAM or cache, as long as you do not hit the "roof" so
> to speak. But with the graphics cards (GPUs) now involved, it looks
> like the best "bang for the buck" may change in the near future. I
> would recommend a 2.5-3.5GHz Quad-core CPU and 4GB RAM for now, but
> with empty slots for further memory expansions (if not for MrB so
> perhaps some other phylogenetic software) and an interchangeable slot
> for a graphics card. When they launch 3.2 you invest in an Nvidia
> card. So in other words, a workstation for a about $1200 or so.
> 
> /Andreas
> 
> On Thu, Oct 14, 2010 at 12:50 PM, Maxim Teslenko <Maxim.Teslenko@...> wrote:
>> Generally MrBayes is not a very memory demanding tool. The actual memory usage depends on specific data set. Especially demanding are codon models due to huge state space. To evaluate how much memory you need you may consider running your analyzes on existing machine and check memory consumption during this run. You do not have to run it for long since memory usage suppose to be pretty much constant through the run.
>> Newer versions of MrBayes 3.2 is using even less memory for the same data set compared to version 3.1. MrBayes 3.2 is due for release in a month or two but you can download prerelease version from http://sourceforge.net/projects/mrbayes/develop. You should have svn client to download from repository. A good one for windows is TortoiseSVN you can get it from http://tortoisesvn.tigris.org/. To compile in windows you need to have MS Visual Studio or you may try to use MrBayes installer which you can find in \projects\vc2008_FR\Debug. The installer has very "buggy" version of MrBayes for now but soon it will be fixed.
>> 
>> Another good thing about MrBayes 3.2 is that it will make use of your graphical card trough use of beagle library. You will need a graphic card from Nvidia with support of CUDA drivers. With decent graphic card you may get an order of magnitude speed up of your analyzes. If you can make a system with multiple GPU you can count on even higher speed up compared to single GPU.
>> 
>> Maxim
>> 
>> -----Original Message-----
>> From: Robert Stephens [mailto:rrs@...]
>> Sent: Wednesday, October 13, 2010 3:09 AM
>> To: mrbayes-users@...
>> Subject: [Mrbayes-users] <no subject>
>> 
>> Hi,
>> 
>> I've been trying to think of what would be the best way to configure a
>> personal desktop pc for it to be the most compatible it can be with how
>> mrbayes is written for windows. I want a windows personal pc that will
>> result in the shortest runtime of mrbayes. For the moment, ignore cost.
>> 
>> I use a personal pc because I have an osteological dataset interleaved with
>> a dna dataset and use a mrbayes block.
>> 
>> For instance how much ram should I be considering- 6 gigs, 32 gigs. Is one
>> not enough or is the other too much.  Just trying to make a good decision
>> and moving on.
>> 
>> Thanks.
>> 
>> Bob
>> 
>> 
>> ------------------------------------------------------------------------------
>> Beautiful is writing same markup. Internet Explorer 9 supports
>> standards for HTML5, CSS3, SVG 1.1,  ECMAScript5, and DOM L2 & L3.
>> Spend less time writing and  rewriting code and more time creating great
>> experiences on the web. Be a part of the beta today.
>> http://p.sf.net/sfu/beautyoftheweb
>> _______________________________________________
>> Mrbayes-users mailing list
>> Mrbayes-users@...
>> https://lists.sourceforge.net/lists/listinfo/mrbayes-users
>> 
>> ------------------------------------------------------------------------------
>> Beautiful is writing same markup. Internet Explorer 9 supports
>> standards for HTML5, CSS3, SVG 1.1,  ECMAScript5, and DOM L2 & L3.
>> Spend less time writing and  rewriting code and more time creating great
>> experiences on the web. Be a part of the beta today.
>> http://p.sf.net/sfu/beautyoftheweb
>> _______________________________________________
>> Mrbayes-users mailing list
>> Mrbayes-users@...
>> https://lists.sourceforge.net/lists/listinfo/mrbayes-users
>> 
> 
> ------------------------------------------------------------------------------
> Beautiful is writing same markup. Internet Explorer 9 supports
> standards for HTML5, CSS3, SVG 1.1,  ECMAScript5, and DOM L2 & L3.
> Spend less time writing and  rewriting code and more time creating great
> experiences on the web. Be a part of the beta today.
> http://p.sf.net/sfu/beautyoftheweb
> _______________________________________________
> Mrbayes-users mailing list
> Mrbayes-users@...
> https://lists.sourceforge.net/lists/listinfo/mrbayes-users

http://economylodgetc.com/super.php

Hi all!

I just noticed that Amazon has started to offer cloud based GPU
computations in addition to their other services and servers over at
Amazon EC2.

While we all scramble to afford the latest and greatest graphics cards
for speedier analyses, it should be possible to at least try out some
analyses on their machines. Each "Cluster GPU" solution has two NVIDIA
Tesla “Fermi” M2050 GPUs (similar to a consumer Geforce GTX 470 but
for servers, not games) and can be rented for $2.10 an hour. So for
the price of two M2050 ($5000, at ~$2500 a piece), we get about 100
days of computation.

It may not sound that much, but keep in mind that the GPUs may
potentially be an order of magnitude or more faster than a powerful
workstation using only CPUs for a given MrB analysis. So whenever you
start doing GPU computations with MrB (in the cloud, or locally),
please post some benchmarks to help people design theirs.

You can read more about their offering here:
http://aws.amazon.com/ec2/hpc-applications/

And a fun use case:
http://stacksmashing.net/2010/11/15/cracking-in-the-cloud-amazons-new-ec2-gpu-instances/

Best regards,
Andreas

Hi all,

I am getting the following warning if I set rates=invgamma.
WARNING: In LIKE_EPSILON - for division 0 char 181 has like = nan

Removing this option i.e. setting rates to be equal (default) removes this error. For my data, I need the rates to be gamma distributed with a proportion of invariant sites. Does anyone have an idea on how this can be fixed?

Thanks
Supriya

Hi all,

I'm trying to diminish convergence time by using a starting tree. As  
per the manual, I've set npert#=2 to "shake" the tree.
When I execute my file, the bayesblock seems to be reading in just  
fine. But in completing the "mcmc" command, I get this error:

Tree to be pruned does not include all taxa
Error in pruning user tree

Of course, the tree was made directly from the dataset and I have not  
pruned it prior to pasting the newick tree into my bayesblock.

Any ideas about why this might be happening? It happened with two  
separate files.

Here's my bayesblock:

partition bymarker = 5: nonCoding, rpl16, matK, ndhF, ITS;
set partition = bymarker;
set autoclose=no;
lset applyto=(2,5) nst=6 rates=invgamma;
lset applyto=(1,3,4) nst=6 rates=gamma;
unlink statefreq=(all) revmat=(all) shape=(all) pinvar=(all);
prset applyto=(all) ratepr=variable;
usertree=(newick tree file, too long to post);
mcmcp ngen=4000000 nruns=2 nchains=4 temp=0.15 nperts=2  
samplefreq=100 mcmcdiagn=yes diagnfreq=10000 relburnin=yes  
burninfrac=0.3 savebrlens=yes ordertaxa=yes Startingtree=user;
end;



Thanks much,
Debra