How to speed up BEAST?

[Данияр Ерболатов]

Hi everyone,

I’m running BEAST  on a Linux server and I’d like some advice on speeding up my analysis.

My XML includes 96 taxa and ~73,000 aligned sites(no partitions), and I plan to run an MCMC chain of 900 million steps.

Could anyone suggest the best practices or settings to improve runtime efficiency?

Any suggestions or experience with similar-scale analyses would be very helpful.

Thanks in advance

[Miller, Mark]

It is likely that you should use 1-4 GPUs to get great speedup for your data set.

For BEAST or BEASTX, these are the rules we use at CIPRES, we would use V100 or A100 gpus for your data set (-threads 1 -beagle_GPU -beagle_cuda).

You can try it if you like.

 

Revised 8/25/2023

                                               beagle_

   Data            Data          Memory    Slurm     thread_                    GPU   Slurm

partitions    patterns   needed   partition   count   cores  threads   type  memory 

 

    <8          <3,000            regular    shared          3           3                        6G

    <8    3,000-79,999    regular  gpu-shared               10             1      V100     90G

    <8        >=80,000         regular  gpu-shared              10             1      A100     90G

   >=8         <10,000         regular    shared           3          3                         6G

   >=8        >=10,000        regular    shared          4          4                         8G

   any           any                extra     shared               6         6                         12G (only valid when gpu is not used)

                                                                   

AA data

 

   any           any    regular  gpu-shared            10       1      V100     90G

 

For BEAST2 we would use 1 GPU for this data set as well.

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/beast-users/90774586-4b5e-451f-9b80-1eaf20e93746n%40googlegroups.com.