Help running beast on GPU

[Nicoletta Commins]

Hi there, 

I'm trying to run BEAST v2.6.3 on GPU (Tesla V100s) but I'm having some issues. What I'm submitting:

module load gcc/6.2.0 cuda/10.1

~/sw/beast/bin/beast -beagle_GPU -beagle_double my_file.xml

But my output gives me this:

  Using BEAGLE version: 3.2.0 (PRE-RELEASE) resource 0: CPU

    with instance flags:  PRECISION_DOUBLE COMPUTATION_SYNCH EIGEN_REAL SCALING_MANUAL SCALERS_RAW VECTOR_SSE THREADING_NONE PROCESSOR_CPU FRAMEWORK_CPU

Does anyone know why my run is using CPU instead of GPU? Also wondering if there is any advantage to using multiple cores on GPU if I only have one large partition. 

Thanks,

Nicoletta

[Miller, Mark]

Hi Nicoletta,

I cant help with why it isn’t using gpus, but here is how we configure cpu/gpu runs on our machne:

GPUs give the best speedup for 1 partition data sets when there are lots of patterns.

 

Data          Data        Slurm                                Other beagle

partitions     patterns    partition  -threads -instances  GPUs   parameters

 

Nucleotide data

 

  1 to 3           <750     shared        1         1             -beagle_SSE

  1 to 3      750-2,999     shared        3         3             -beagle_SSE

  1 to 3    3,000-9,999     shared        6         6             -beagle_SSE

  1 to 3  10,000-39,999   gpu-shared      1         1        1    -beagle_GPU

  1 to 3       >=40,000       gpu         4         4        4    -beagle_GPU

                                                                    -beagle_order 1,2,3,4

  4 to 19        <1,200     shared        1         1             -beagle_SSE

  4 to 19   1,200-4,999     shared        3         3             -beagle_SSE

  4 to 19  5,000-19,999     shared        6         6             -beagle_SSE

  4 to 19      >=20,000   gpu-shared      1         1        1    -beagle_GPU

 

   >=20             any     shared        4         1             -beagle_SSE

 

Amino acid data

 

     1           <5,000   gpu-shared      1         1        1    -beagle_GPU

     1          >=5,000       gpu         4         4        4    -beagle_GPU

                                                                    -beagle_order 1,2,3,4

  2 to 39           any   gpu-shared      1         1        1    -beagle_GPU

 

   >=40             any     compute      24         1             -beagle_SSE

--
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 on the web visit https://groups.google.com/d/msgid/beast-users/732176bb-bd8b-455c-89ea-08a87f7ba426n%40googlegroups.com.

[Dylan Ohearn]

Thanks for the very helpful guide.  Is the number of patterns total or per-partition?  For example, if I have 1500 patterns total in 3 partitions, would that call for settings threads and instances 1 or 3?

[Miller, Mark]

The patterns reported is the total number of patterns across all partitions in this case.

 

Mark

To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/99646f8c-eded-4af9-b20f-f64285d060abn%40googlegroups.com.