Getting Started with NVIDIA Parabricks

NVIDIA Clara Parabricks v4.0.1

Hardware Requirements

  • Any NVIDIA GPU that supports CUDA architecture 70, 75, 80, 86, 89 or 90 and has at least 16GB of GPU RAM. Parabricks has been tested on the following NVIDIA GPUs:

    • V100

    • T4

    • A10, A30, A40, A100, A6000

    • L4, L40

    • H100

  • The fq2bam tool requires at least 24 GB of GPU memory by default; the --low-memory option will reduce this to 16 GB of GPU memory at the cost of slower processing. All other tools require at least 16 GB of GPU memory per GPU.

  • System Requirements:

    • A 2 GPU system should have at least 100GB CPU RAM and at least 24 CPU threads.

    • A 4 GPU system should have at least 196GB CPU RAM and at least 32 CPU threads.

    • A 8 GPU system should have at least 392GB CPU RAM and at least 48 CPU threads.


Parabricks is not supported on virtual (vGPU) or Multi-Instance (MIG) GPUs.

Software Requirements

The following are software requirements for running NVIDIA Parabricks.

  • An NVIDIA driver with version 525.60.13 or greater .

  • Any Linux Operating System that supports nvidia-docker2 Docker version 20.10 (or higher)

Verifying Hardware and Software Requirements

Checking available NVIDIA hardware and driver

To check your NVIDIA hardware and driver version, use the nvidia-smi command:


$ nvidia-smi +-----------------------------------------------------------------------------+ | NVIDIA-SMI 525.60.13 Driver Version: 525.60.13 CUDA Version: 12.0 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 Tesla V100-DGXS... On | 00000000:07:00.0 Off | 0 | | N/A 44C P0 38W / 300W | 74MiB / 16155MiB | 0% Default | | | | N/A | +-----------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=============================================================================| | 0 N/A N/A 3019 G /usr/lib/xorg/Xorg 56MiB | +-----------------------------------------------------------------------------+

This shows the following important information:

  • The NVIDIA driver version is 525.60.13.

  • The supported CUDA driver API is 12.0.

  • The GPU has 16 GB of memory.

Checking available CPU RAM and threads

To see how much RAM and CPU threads in your machine, you can run the following:


# To check available memory $ cat /proc/meminfo | grep MemTotal # To check available number of threads $ cat /proc/cpuinfo | grep processor | wc -l

Checking nvidia-docker2 installation

To make sure you have nvidia-docker2 installed, run this command:


$ docker run --rm --gpus all nvidia/cuda:12.0.0-base-ubuntu20.04 nvidia-smi

When it finishes downloading the container, it will run the nvidia-smi command and show you the same output as above.

Checking python version

To see which version of Python you have, enter the following command:


$ python3 --version

Make sure it's at least version 3 (3.6.9, 3.7, etc).

The Parabricks Docker image can be obtained by running the following command:


$ docker pull

At this point the software is ready to use.

From the Command Line

Parabricks is deployed using a Docker image. There are two parts to customizing a Parabricks run:

  • Customizing Docker container specific options: These are the options that are passed to the docker command before the name of the container. For example, the user should mount their data directories within the Docker container by passing the -v option to Docker. See the Tutorials for more detailed examples.

  • Parabricks specific options: These options are passed to the Parabricks command line to customize the Parabricks run. For example, you can choose which tool to run and pass tool-specific options.

For example, to run the Parabricks fq2bam (FQ2BAM + BWA-MEM) tool using the Docker container, use the following command:


$ docker run \ --gpus all \ --rm \ --volume /host/data:/input_data \ --volume /host/results:/outputdir \ --workdir /input_data \ \ pbrun fq2bam \ --ref /input_data/Ref/Homo_sapiens_assembly38.fasta \ --in-fq /input_data/Data/sample_1.fq.gz /input_data/Data/sample_2.fq.gz \ --out-bam /outputdir/fq2bam_output.bam

Sample data is freely available. See the Getting The Sample Data section in the Tutorials for instructions on obtaining the sample data, and a step-by-step guide to using both fq2bam and Haplotype Caller.

Some useful Docker options to consider:

  • --gpus all lets the Docker container use all the GPUs on the system. The GPUs available to Parabricks container can be limited using the --gpus "device=<list of GPUs>" option. Use nvidia-smi to see how many GPUs you have, and which one is which.

  • --rm tells Docker to terminate the image once the command has finished.

  • --volume /host/data:/image/data mounts your /host/data (a path on the server) on the Docker container in the /image/data directory (a path inside the Docker container).

  • --workdir tells Docker what working directory to execute the commands from.

  • The rest is the Parabricks tool you want to run, followed by its arguments. For those already familiar with Parabricks and its pbrun command, this Docker invocation takes the place of pbrun.

Running Parabricks Using the Base Command Platform

An example command to launch a BaseCommand container on a single-GPU instance is:


ngc batch run --name "parabricks-germline" \ --instance dgxa100.80g.1.norm \ --commandline "pbrun germline \ --ref /workspace/parabricks_sample/Ref/Homo_sapiens_assembly38.fasta \ --in-fq /Data/HG002-NA24385-pFDA_S2_L002_R1_001-30x.fastq.gz /Data/HG002-NA24385-pFDA_S2_L002_R2_001-30x.fastq.gz \ --knownSites /workspace/parabricks_sample/Ref/Homo_sapiens_assembly38.known_indels.vcf.gz \ --out-bam output.bam \ --out-variants output.vcf \ --out-recal-file report.txt \ --run-partition \ --no-alt-contigs" \ --result /results \ --image "nvidia/clara/clara-parabricks:4.1.0-1"

Note that for other Parabricks commands (i.e. fq2bam, HaplotypeCaller, DeepVariant) the ngc batch run command is similar. Make sure to use the correct paths for your workplace or dataset that contains the data you intend to use.

The goal of Parabricks software is to get the highest performance for bioinformatics and genomic analysis. There are a few key system options that you can tune to achieve maximum performance.

Use a Fast SSD

Parabricks software operates with two kinds of files:

  • Input/output files specified by the user

  • Temporary files created during execution and deleted at the end

The best performance is achieved when both kinds of files are on a fast, local SSD. If this is not possible you can place the input/output files on a fast network storage device and the temporary files on a local SSD using the --tmp-dir option.


Tests have shown that you can use up to 4 GPUs and still get good performance with the Lustre network for Input/Output files. If you plan to use more than 4 GPUs, we highly recommend using local SSDs for all kinds of files.

DGX Users

The DGX comes with a SSD, usually mounted on /raid. Use this disk, and use a directory on this disk as the --tmp-dir. For initial testing, you can even copy the input files to this disk to eliminate variability in performance.

Specifying which GPUs to use

You can choose the number of GPUs to run using the command line option --num-gpus N for those tools that use GPUs. With this command, the GPUs used will be limited to the first N GPUs listed in the output of the nvidia-smi command.

To select specific GPUs, set the environment variable NVIDIA_VISIBLE_DEVICES. GPUs are numbered starting with zero. For example, this command will use only the second (GPU #1) and fourth (GPU #3) GPUs:


$ NVIDIA_VISIBLE_DEVICES="1,3" pbrun fq2bam --num-gpus 2 --ref Ref.fa --in-fq S1_1.fastq.gz --in-fq S1_2.fastq.gz

Uninstalling Parabricks is as simple as removing the Docker image.


$ docker images REPOSITORY TAG IMAGE ID CREATED SIZE 4.1.0-1 516740210042 2 months ago 2.82GB $ docker rmi 516740210042

The exact value of the "IMAGE ID" will vary depending on your installation.


© Copyright 2023, Nvidia. Last updated on Jun 28, 2023.