[Nvidia/Parabricks] Does pbrun support GPU options?

Healthcare & Life Sciences Parabricks
1

hi, there

When I ran pbrun fq2bam ... to test a small dataset, I met a memory error as shown below.

$ pbrun fq2bam   --ref /home/anonymous/parabricks/dataset/D99999_S99_L001/human_g1000_v37_decoy.fasta   --in-fq /home/anonymous/parabricks/dataset/D99999_S99_L001/D99999_S99_L001_R1.fastq.gz     /home/anonymous/parabricks/dataset/D99999_S99_L001/D99999_S99_L001_R2.fastq.gz   --out-bam /home/anonymous/parabricks/dataset/D99999_S99_L001/out.bam
Please visit https://www.nvidia.com/en-us/docs/parabricks/ for detailed documentation

[Parabricks Options Mesg]: Checking argument compatibility
[Parabricks Options Mesg]: Automatically generating ID prefix
[Parabricks Options Mesg]: Read group created for
/home/anonymous/parabricks/dataset/D99999_S99_L001/D99999_S99_L001_R1.fastq.gz and
/home/anonymous/parabricks/dataset/D99999_S99_L001/D99999_S99_L001_R2.fastq.gz
[Parabricks Options Mesg]: @RG\tID:000000000-C54LP.1\tLB:lib1\tPL:bar\tSM:sample\tPU:000000000-C54LP.1
------------------------------------------------------------------------------
||                 Parabricks accelerated Genomics Pipeline                 ||
||                              Version v3.1.1                              ||
||                       GPU-BWA mem, Sorting Phase-I                       ||
||                  Contact: Parabricks-Support@nvidia.com                  ||
------------------------------------------------------------------------------
Not enough available memory for these GPUs...
Please contact Parabricks-Support@nvidia.com if you are rightfully using the software but cannot run it.
Please contact Parabricks-Support@nvidia.com for any questions. Exiting...

Could not run fq2bam
Exiting pbrun ...

Here is the GPU summary of the Driver and CUDA versions

$ nvidia-smi

image
image726×443 48.7 KB

Is there any methods to restrict the GPU usages ( memory<= 11GB) to fulfill my requirements?

2

Not enough available memory for these GPUs…

Attached the log:

cudaSafeCall() failed at ParaBricks/src/memoryManager.cu:78 : out of memory
cudaSafeCall() failed at ParaBricks/src/samGenerator.cu:781 : out of memory
cudaSafeCall() failed at ParaBricks/src/samGenerator.cu:781 : out of memory
cudaSafeCall() failed at ParaBricks/src/samGenerator.cu:781 : out of memory
Please contact support@parabricks.com for any questions. Exiting...

According to the article “There is no CUDA API to limit memory usage, with or without nvidia-docker.”,

there is a comment:

There is no CUDA API to limit memory usage, with or without nvidia-docker.

so bad…

3

Hi,

we are sorry, but one of the requirements of Clara Parabricks Pipelines is GPU with 12 GB memory RAM or more.

https://docs.nvidia.com/clara/parabricks/v3.0/text/local_installation.html

Myrieme