Base-calling server parameters

The default settings for Guppy GPU should be sufficient for most use cases. If you are really into optimisation, we recommend reading Miles Benton’s Guppy tuning guide - until Dorado comes along this all should work.

The one trick we sometimes use, on a really high performance flow cell is to split the GPUs between readfish and guppy.

This of course assumes that you have a multiple GPU machine, such as PromethION tower. In the new P24 towers, there are 4x NVIDIA A100 GPUs.

  1. Stop guppy sudo systemctl stop guppyd or stop dorado sudo systemctl stop doradod

  2. View the available CUDA devices. nvidia-smi

  3. Restart two guppy/dorado instances, splitting the devices between each.

Guppy:

guppy_basecall_server --port ipc:///tmp/.guppy/5555 --ipc_threads 2 --trim_adapters -c /opt/ont/guppy/data/dna_r10.4.1_e8.2_400bps_fast.cfg -x cuda:0 --log_path /tmp/

guppy_basecall_server --port ipc:///tmp/.guppy/5556 --ipc_threads 2 --trim_adapters -c /opt/ont/guppy/data/dna_r10.4.1_e8.2_400bps_fast.cfg -x cuda:1 --log_path /tmp/

Dorado:

/opt/ont/dorado/bin/dorado_basecall_server --log_path /var/log/dorado --config dna_r10.4.1_e8.2_400bps_fast.cfg --ipc_threads 3 --port /tmp/.guppy/5555 --dorado_download_path /opt/ont/dorado-models --device cuda:0

/opt/ont/dorado/bin/dorado_basecall_server --log_path /var/log/dorado --config dna_r10.4.1_e8.2_400bps_fast.cfg --ipc_threads 3 --port /tmp/.guppy/5556 --dorado_download_path /opt/ont/dorado-models --device cuda:1

  1. Start readfish, base calling on port 5556