Reducing the number of CPUs being used

[niradsp]

Hello, I just noticed that for some reason Bonito used up all of the 192 CPUs that my server has. Can this parameter be controlled somehow, or am I wrong about this?

Thanks, Nirad

[iiSeymour]

How are you measuring CPU usage here, what are you running specifically and are you targeting a GPU or CPU?

[niradsp]

Hello Chris,

I am running the bonito basecaller. This is supposed to be GPU. The GPU I have is Tesla V100. Bonito was installed using pip.
The basecaller was essentially run using the following command:

bonito basecaller dna_r9.4.1 /data/reads > basecalls.fasta

What I notice is (when typing nvidia-smi when Bonito is being run) that around 3-4 GB GPU memory is being used.
I am typing "htop" to look at usage. When Bonito is being used, it says all 192 CPUs are running.

Actually, another question, So my samples are human data. I downloaded the Bonito model for Guppy. I am planning on running Guppy using this model. Would you say that this will beat the standard Guppy models in terms of accuracy? I think the following was based on using the Bonito software rather than Guppy with the bonito model: https://nanoporetech.github.io/ont-open-datasets/bonito/

Thanks, Nirad

[iiSeymour]

You should not see high activity on all the cores - what about if you set export OMP_NUM_THREADS=1 before running basecalling?

Yes, the bonito models on rerio should outperform the guppy models.

[niradsp]

Hello Chris, That seems to have worked! So this is something I need to export every time I run Bonito?

The GPU RAM usage is 3.6G. The reads per second metric has also jumped. I am now seeing 200 reads per second as opposed to 80 or 90 previously.

Thanks, Nirad