Can not fully use GPU?

[LauraVP1994]

Dear,

I have been using Beast to make a phylogenetic tree from 250 influenza samples. Recently I had the chance to use GPUs to let this process go faster, however, for some reason the script won't use the GPUs. I also contacted the helpdesk for these GPUs and they also don't understand why he doesn't want to use both GPU's. In attachment, you find the output of "beast -beagle_info" to show that there are two GPUs. I also included the scripts I tried to run and that it is not running on the GPUs...

Can someone help me? I would be very grateful!

Kind regards Laura

[GuyBaele]

Hi Laura, Can you also provide the output of BEAST to the screen? Are the different likelihoods assigned to the 2 GPUs in an alternating way? Thanks, Guy

[LauraVP1994]

Hi Laura, Can you also provide the output of BEAST to the screen? Are the different likelihoods assigned to the 2 GPUs in an alternating way? Thanks, Guy

Hi Guy,

Which kind of output would you need? There are multiple files for the xml-file "allSegments" being generated

• allSegSegments.location.rates.log
• allSegSegments.log
• allSegSEgments.trees
• phyleo.Beastlog
• phyleo.error ( only mentioned: "Warning likelihood component, null, created but not used in the MCMC")
• phyleo.output (=empty while running for the moment))

I'm not (yet) very experienced with Beast (but I'm eager to learn!), so for the second question, I'm not very sure where to find this information? The purpose would be to speed up the analysis because for the moment it takes multiple weeks to run,...

Kind regards and thanks in advance Laura

[GuyBaele]

The .log and .trees file won't provide this information; we would need what's printed to screen, so on your system that probably ends up in phyleo.output (i.e. the standard output). Perhaps do a very short run of a few minutes so you can send that information. It's typically the information between "Creating site rate model." and "Creating the MCMC chain:"

Best regards, Guy

[LauraVP1994]

The .log and .trees file won't provide this information; we would need what's printed to screen, so on your system that probably ends up in phyleo.output (i.e. the standard output). Perhaps do a very short run of a few minutes so you can send that information. It's typically the information between "Creating site rate model." and "Creating the MCMC chain:"

Best regards, Guy

Dear Guy,

If I'm correct this information is being saved in the BeastLog file, I copied it to a document:

20200123 Output.docx

Kind regards Laura

[GuyBaele]

Dear Laura, Best I can tell, this isn't a bug so I'm closing this issue and we can continue discussing via e-mail. Can you please e-mail me your XML file so I can advise you on how to proceed? Best regards, Guy

[GuyBaele]

Need to reopen this. No matter what I try, I can't get the separate discrete location partition to run on the second GPU. There's no problem splitting up sequence data into 2 multi-partition likelihoods and run those on 2 different GPUs, but the discrete location likelihood keeps getting assigned to the first GPU (despite being explicitly assigned to the second GPU).

[oushujun]

@GuyBaele Just want to follow up on this issue, do you have a fix for this bug? We are seeing the same issue with BEASTv1.10.4 and BEAGLE library v3.2.0.

[bpetros95]

having the same issue here-- discrete location gets assigned to same GPU as sequences, no matter what I try