> This page is for version v4.8.0 (default).
> For other versions, use one of these documentation indexes:
> - v4.8.0 (default): https://docs.nvidia.com/clara/parabricks/v4.8.0/llms.txt
> - v4.7.1: https://docs.nvidia.com/clara/parabricks/v4.7.1/llms.txt

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/clara/parabricks/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/clara/parabricks/_mcp/server.

# 4.8.0-1 Release Notes

**Highlights:**

* Updated base container to CUDA 13.3.1 with Rubin support.
* Updated version of [giraffe](/tool-reference/tools/giraffe) to v1.75.1.
* Performance improvements for single-end alignment with [giraffe](/tool-reference/tools/giraffe).
* Added support for supplementary alignment and CRAM file input/output for [giraffe](/tool-reference/tools/giraffe).
* Updated the compatible BWA-MEM version to 0.7.17 for [fq2bam](/tool-reference/tools/fq2bam) and [fq2bam_meth](/tool-reference/tools/fq2bam_meth).
* Added Hi-C data support to [fq2bam](/tool-reference/tools/fq2bam).
* Better scaling with optimized `--run-partition` for [fq2bam](/tool-reference/tools/fq2bam) and [fq2bam_meth](/tool-reference/tools/fq2bam_meth).
* Performance improvements for [fq2bam](/tool-reference/tools/fq2bam) and [fq2bam_meth](/tool-reference/tools/fq2bam_meth), with the largest gains for paired-end workflows and minor improvements for single-end workflows.
* Better host memory management in Marking Duplicates & BQSR step.
* Performance improvements for long-read technologies with [deepvariant](/tool-reference/tools/deepvariant).
* Performance improvements for short-read RNA alignment with [minimap2](/tool-reference/tools/minimap2).

## Improvements

All tools:

* Added the `--dry-run` flag to all Parabricks tools. This flag enables testing a potential command without requiring GPU access.
* Added the `--help-json` flag to all Parabricks tools. This prints out the help message in JSON format for easier machine consumption.

### Tool Updates

[fq2bam](/tool-reference/tools/fq2bam), [fq2bam_meth](/tool-reference/tools/fq2bam_meth), and associated pipelines:

* The compatible BWA-MEM version has been updated from 0.7.15 to 0.7.17.
* Added support for additional BWA-MEM commandline arguments: `-S`, `-P`, `-5`, and `-q` via the `--bwa-options` commandline parameter.
* Hi-C data is now supported in `fq2bam` using the aforementioned arguments, which can be provided as `--bwa-options="-SP5M <additional args>"`.
* Improved end-to-end BAM generation performance, mainly for paired-end alignment, with smaller improvements for single-end alignment.
* Better scaling with optimized `--run-partition` when using more GPUs.

[fq2bam](/tool-reference/tools/fq2bam):

* Updated the default CPU threading configuration for improved end-to-end runtime. The option `--bwa-primary-cpus` now defaults to `auto`, which selects 16 primary CPU threads per GPU, and `--bwa-cpu-thread-pool` now defaults to `1`. Both options can still be set manually to override the defaults.

[deepvariant](/tool-reference/tools/deepvariant):

* Performance improvements for long-read technologies (PacBio and ONT).

[deepvariant](/tool-reference/tools/deepvariant), [deepsomatic](/tool-reference/tools/deepsomatic), and [pangenome_aware_deepvariant](/tool-reference/tools/pangenome_aware_deepvariant):

* Updated TensorRT from version 10.14.1.48 to 11.3.0.99.
* Added timing summary for each stage when enabling `--verbose`.
* Added validation of multiallelic call outputs before predictions are merged. Parabricks now exits with a locus-specific error if the expected alternate allele combinations are missing or invalid, or if records disagree on the canonical variant.

[haplotypecaller](/tool-reference/tools/haplotypecaller), [mutectcaller](/tool-reference/tools/mutectcaller):

* Introduced `auto` mode for `--num-streams-per-gpu` that automatically selects the number of streams per GPU based on available GPU memory. This value can be overridden by providing an explicit value.

[giraffe](/tool-reference/tools/giraffe):

* Updated the baseline vg version from 1.70.0 to 1.75.1. New index files must be generated with vg v1.75.1; refer to the [index generation](/tool-reference/tools/giraffe#generating-required-index-files) section.
* Only GBZ version 1 graphs are supported. Starting with vg 1.72.0, `vg autoindex` and `vg gbwt` write GBZ version 2 by default, which stores node sequences zstd-compressed. Loading such a graph now fails with an error naming the conversion command; refer to [Supported GBZ version](/tool-reference/tools/giraffe#supported-gbz-version).
* Added CRAM output, selected by passing `--out-bam` a path ending in `.cram`. This requires `--ref`, and `--run-ref-verification` is recommended to check the reference against the graph.
* Added support for aligning single-end CRAM input with `--in-se-cram`, together with `--in-cram-ref` for aligned input and `--cram-read-threads` to control decode threads.
* Added `--ultima-tags` to carry the Ultima per-base `tp` and `t0` auxiliary tags from CRAM input through to the output, and `--tp-output-order` to select whether those values follow the output `SEQ` or the sequencing direction.
* Performance improvements for single-end alignment.

[giraffe](/tool-reference/tools/giraffe) and [pangenome_germline](/tool-reference/tools/pangenome_germline):

* Added `--supplementary` to identify and report supplementary alignments, for both single-end and paired-end reads. It is disabled by default.

[minimap2](/tool-reference/tools/minimap2):

* Added proper fragment handling of pair-ended FASTQ inputs.
* Moved the seeding stage to GPU to reduce CPU load. The `--seeding-chaining-on-gpu [on|off|auto]` flag controls where both the seeding and chaining stage are computed.
* Introduced an "auto" mode for the best performance, which tunes parameters based on the device memory and preset. The parameters can be overridden by providing an explicit value.
* Added a read group tag to each output read.

[bqsr](/tool-reference/tools/bqsr), [fq2bam](/tool-reference/tools/fq2bam), [fq2bam_meth](/tool-reference/tools/fq2bam_meth), [giraffe](/tool-reference/tools/giraffe), [minimap2](/tool-reference/tools/minimap2), [deepvariant_germline](/tool-reference/tools/deepvariant_germline), [germline](/tool-reference/tools/germline), [ont_germline](/tool-reference/tools/ont_germline), [pacbio_germline](/tool-reference/tools/pacbio_germline), [somatic](/tool-reference/tools/somatic):

* Updated the help message of `--knowSites` when enabling Base Quality Score Recalibration (BQSR).

[bamsort](/tool-reference/tools/bamsort), [fq2bam](/tool-reference/tools/fq2bam), [fq2bam_meth](/tool-reference/tools/fq2bam_meth), [giraffe](/tool-reference/tools/giraffe), [minimap2](/tool-reference/tools/minimap2), [deepvariant_germline](/tool-reference/tools/deepvariant_germline), [germline](/tool-reference/tools/germline), [ont_germline](/tool-reference/tools/ont_germline), [pacbio_germline](/tool-reference/tools/pacbio_germline), [somatic](/tool-reference/tools/somatic):

* Skipped checking `LB` (library preparation identifier) when marking duplicates is disabled.

[fq2bam](/tool-reference/tools/fq2bam), [fq2bam_meth](/tool-reference/tools/fq2bam_meth), [giraffe](/tool-reference/tools/giraffe), [minimap2](/tool-reference/tools/minimap2), [rna_fq2bam](/tool-reference/tools/rna_fq2bam), [bamsort](/tool-reference/tools/bamsort), and associated pipelines:

* Optimized host memory usage during BAM and CRAM file writing.

[applybqsr](/tool-reference/tools/applybqsr):

* Clarified that CRAM input and output files are accepted for the options named `--in-bam` and `--out-bam`, respectively.

## Bug Fixes

All tools that take a reference FASTA:

* Fixed handling of Windows (CRLF) line endings in the reference FASTA. A trailing carriage return no longer leaks into contig names or sequence data.

[deepvariant](/tool-reference/tools/deepvariant), [deepsomatic](/tool-reference/tools/deepsomatic), and [pangenome_aware_deepvariant](/tool-reference/tools/pangenome_aware_deepvariant):

* Fixed bug affecting how read support is computed for duplicate reads.

[deepvariant](/tool-reference/tools/deepvariant):

* The `--disable-small-model` argument is no longer required with WES mode. Parabricks now matches Google DeepVariant's default behavior by disabling the small model in WES mode.
* Fixed bug causing improper operation of the `--keep-duplicates` flag.
* Fixed a rare edge-case bug that occasionally caused a deadlock when processing PacBio or ONT data.

[deepsomatic](/tool-reference/tools/deepsomatic):

* Fixed bug causing incorrect handling of the `--no-channel-insert-size` flag.

[haplotypecaller](/tool-reference/tools/haplotypecaller):

* Fixed an edge-case bug that occurred during gVCF writing.

[germline](/tool-reference/tools/germline):

* Added additional argument checks to prevent runs from failing in [haplotypecaller](/tool-reference/tools/haplotypecaller).

[deepvariant_germline](/tool-reference/tools/deepvariant_germline), [germline](/tool-reference/tools/germline), [ont_germline](/tool-reference/tools/ont_germline), [pacbio_germline](/tool-reference/tools/pacbio_germline), [pangenome_germline](/tool-reference/tools/pangenome_germline), and [somatic](/tool-reference/tools/somatic):

* The option `--preserve-file-symlinks` now properly works for pipeline commands.

[fq2bam](/tool-reference/tools/fq2bam), [fq2bam_meth](/tool-reference/tools/fq2bam_meth), [giraffe](/tool-reference/tools/giraffe), [minimap2](/tool-reference/tools/minimap2), [rna_fq2bam](/tool-reference/tools/rna_fq2bam), [bamsort](/tool-reference/tools/bamsort), and associated pipelines:

* Fixed a bug that could cause runs using `--gpuwrite` to fail with an error during BAM writing. An internal check was overly strict, so valid runs could have been flagged as errors.

For further information refer to the
[Parabricks datasheet](https://resources.nvidia.com/en-us-hc-genomics/healthcare-genomics).