Improvements

Improved performance on Hopper and Blackwell GPUs, with more modest gains on older architectures.

Allowed reads up to 512 bp to be processed on GPU, increasing the previous threshold from 500 bp.

• When the --max-read-length filter is set above its default of 480 bp and up to 512 bp, all reads are processed on GPU. If the limit is set above 512 bp, reads longer than 512 bp are processed on CPU through the recovery mechanism.

Added support for manually setting the insert size distribution -I option from baseline BWA with --bwa-options.

• For example: --bwa-options="-I 250.54,20.25,500,20". The user interface is unchanged.

Added the --cigar-on-gpu option to offload CIGAR generation from the CPU (default) to the GPU; enabling this option may improve performance for CPU-bound runs.

Improved performance when using --gpuwrite.

Added a CPU recovery mode when using --gpusort to recover and continue processing even when the GPU may run out of memory.

GPU CRAM writer has reduced CPU memory requirements. Memory usage reductions can go up to 70% based on available host memory.

Enabled read group id comparison in markdup single-ended mode. Can be disabled by passing --ignore-rg-markdups-single-ended.

Added support for the small model with --enable-small-model. This option is supported for shortread (WGS), PacBio, and ONT modes.

Improved host memory usage for long-read (PacBio and ONT) mode.

Updated version of minimap2 to v2.30.

Improved performance in PacBio mode and better GPU utilization.

Improved performance in splice presets with implementation of new alignment batching algorithm.

Added support for --preset options lr:hq and splice:sr.

Added argument --copy-comment for appending FASTQ comments to BAM output with auxiliary tag.

Added argument --md for outputting the MD tag.

Added argument --both-strands that forces minimap2 to consider both strands when finding canonical splicing sites GT-AG.

Added --exclude-intervals (-XL) to specify genomic intervals to exclude from processing.

Auto mode: When --num-streams-per-gpu and/or --num-threads are omitted, their automatic configuration is triggered based on GPU and system memory. These settings can be overridden by providing explicit values.

Performance improvements from code optimizations.

Updates to match the baseline VG Giraffe version v1.70.

Added GPU-accelerated Finalize Mappings (Dozeu) for single-end alignment.

Introduced auto mode for --nstreams that automatically configures the number of streams based on the GPU's device memory specifications. This provides sensible defaults for most systems but may still benefit from manual tuning for specific configurations.

Removed the --minimizers-gpu-sort option, as it is no longer needed.

Now used as the pangenome alignment stage in the new pangenome_germline pipeline.