NVIDIA RTX PRO 4500 Blackwell でゲノム解析とタンパク質折りたたみワークロードを高速化

Precision medicine depends on two fundamental capabilities: understanding disease at the genomic level and identifying treatments at the molecular level.

NVIDIA’s contributions to precision medicine extend far beyond accelerated computing, delivering a full-stack platform that translates hardware and software advancements directly into healthcare outcomes.

Sequencing the human genome initially took more than a decade, and now it can be completed in a matter of hours. This dramatic shift goes beyond a technical milestone. It has enabled earlier detection, faster diagnoses, and more targeted therapies, fundamentally changing how diseases are understood and treated.

This increase in sequencing speed has shifted the genomic bottleneck from data generation to data analysis. Faster sequencing is only valuable if downstream analysis can keep pace. Clinicians need to make treatment decisions more quickly, particularly in highly time-sensitive settings such as oncology or neonatal intensive care units (NICUs), where every minute matters.

Separately, the traditional process of characterizing a protein’s structure, which is fundamental to drug development, once required years of laborious experimental work, but new AI-based methods like AlphaFold have reduced this to minutes or hours. This shift accelerates drug discovery by dramatically reducing the time and expense of identifying therapeutic candidates and enabling high-throughput screens.

Genomics helps you understand disease, protein structure helps you find treatments for it. They’re two stages of the same journey.

This post explores how the latest advancements in the NVIDIA BioNeMo platform, including NVIDIA Parabricks, and the recently announced RTX PRO 4500 Blackwell Server Edition, are enabling researchers and clinicians in healthcare and life sciences to move faster, with greater accuracy, and at significantly lower compute cost.

NVIDIA Parabricks on NVIDIA RTX PRO 4500 Blackwell

NVIDIA Parabricks, an accelerated genomic analysis solution, plays a pivotal role in addressing this data analysis bottleneck. By providing GPU-accelerated versions of trusted, open-source tools, Parabricks reduces analysis to minutes from hours—enabling researchers to uncover biological insights and clinicians to make decisions faster.

The NVIDIA RTX PRO 4500 Blackwell Server Edition GPU is the newest addition to the RTX PRO data center portfolio. Based on the NVIDIA Blackwell Architecture, this compact, energy-efficient platform delivers powerful compute capabilities for a wide range of workloads deployed across cloud, data center, and edge; including improved performance for NVIDIA Parabricks.

Accelerating alignment and variant calling: Minimap2, fq2bam, and DeepVariant

Typically time-intensive tasks, such as alignment and variant calling, can take hours on traditional CPU-based methods. Minimap2 and fq2bam are widely used for alignment, while DeepVariant is a popular tool for variant calling. Minimap2 is a sequence alignment tool to align DNA or RNA sequencing reads to a reference genome and fq2bam is the Parabricks wrapper to BWA-MEM, including GATK best practices. DeepVariant is Google’s deep-learning based variant caller for germline variants (i.e., inherited diseases).

Parabricks achieves significant speedups across GPU architectures and is continuously optimized for ongoing acceleration. For sequence alignment and variant calling applications, the RTX PRO 4500 Blackwell provides performance speedups over previous GPU editions. Minimap2 and DeepVariant are both roughly 2x faster when compared to the NVIDIA L4 Tensor Core GPU. For fq2bam, the RTX PRO 4500 is 2.4x faster than the NVIDIA L4.

Parabricks v4.7 benchmarks

*Table 1. Time in minutes. Numbers gathered by the NVIDIA Perflab team using Parabricks v4.7.0 with internal nodes. Only use them for reference. Speeds may vary depending on the data set, GPU instance, host CPU, memory availability, and other factors. 30× whole genome sequenced for DeepVariant and fq2bam with Illumina data. 35× whole genome sequenced for Minimap2 with Pacbio data*

PacBio is a genomic sequencing company known for its long-read sequencing technology. By integrating the RTX PRO 4500 Blackwell, PacBio achieved significant speedups in basecalling, the first computational step that converts raw instrument output into usable sequence data for downstream analysis.

“PacBio HiFi sequencing demands accuracy that can’t be compromised, and speed that keeps pace with the biology. The RTX PRO 4500 Blackwell Server Edition GPU delivers both,” says Armin Töpfer, senior director of instrument analysis at Pacific Biosciences.

“We see more than a 2x improvement in basecalling throughput over theL4 GPU, with a power and size profile that opens new possibilities for how and where sequencing analysis can happen,” he said. “This, coupled with a speed and increase in the Parabricks minimap2 and DeepVariant tools, makes us excited about the future of our platforms.”

Get started with Minimap2

docker run --rm --gpus all --volume $(pwd):/workdir --volume $(pwd):/outputdir \

--workdir /workdir \

nvcr.io/nvidia/clara/clara-parabricks:4.7.0-1 \

pbrun minimap2 \

--ref /workdir/${REFERENCE_FILE} \

--in-fq /workdir/${INPUT_FASTQ} \

--out-bam /outputdir/${OUTPUT_BAM}

Get started with fq2bam (BWA-MEM)

# This command assumes all the inputs are in the current working directory and all the outputs go to the same place.

docker run --rm --gpus all --volume $(pwd):/workdir --volume $(pwd):/outputdir \

--workdir /workdir \

nvcr.io/nvidia/clara/clara-parabricks:4.7.0-1 \

pbrun fq2bam \

--ref /workdir/${REFERENCE_FILE} \

--in-fq /workdir/${INPUT_FASTQ_1} /workdir/${INPUT_FASTQ_2} \

--knownSites /workdir/${KNOWN_SITES_FILE} \

--out-bam /outputdir/${OUTPUT_BAM} \

--out-recal-file /outputdir/${OUTPUT_RECAL_FILE}

Get started with DeepVariant

docker run --rm --gpus all --volume $(pwd):/workdir --volume $(pwd):/outputdir \

--workdir /workdir \

nvcr.io/nvidia/clara/clara-parabricks:4.7.0-1 \

pbrun deepvariant \

--ref /workdir/${REFERENCE_FILE} \

--in-bam /workdir/${INPUT_BAM} \

--out-variants /outputdir/${OUTPUT_VCF}

Advancing structural biology with Openfold3 and cuEquivariance

The integration of Openfold3 with cuEquivariance further enhances the capabilities of RTX PRO platforms for protein structure inference. With the latest generation Blackwell Tensor cores, RTX PRO 4500 Blackwell delivers significant acceleration, offering up to a 2.3x speedup over the L4 baseline allowing it to process proteins up to 1,500 Amino acids.

*Table 2. Time in seconds. Datasets used include: Input MSAs generated using the colabfold database and mmseqs2 from a sampled CASP14 dataset. Inference using BF16 precision*

High performance in Smith-Waterman alignment

Leveraging the new set of DPX instructions for dynamic programming introduced in the Blackwell architecture, the RTX PRO 6000 and RTX PRO 4500 GPUs deliver massive throughput for Smith-Waterman alignment. This hardware-accelerated feature is now broadly accessible to all developers via the latest CUDA 13.2 at both the Math API and PTX 9.2 level, bringing new levels of acceleration for DNA, RNA, and protein alignment methods with 32-, 16-, and 8-bit precision support.

The RTX PRO 4500 Blackwell is now 9.6x faster than the L4 and performs on par with the H100 SXM using previous generation DPX. For even higher throughput, the RTX PRO 6000 BSE outperforms the RTX PRO 4500 BSE by 2.36x.

*Table 3. Performance normalized using giga-cells updated by seconds. CPU baseline measured using SSW library. Affine gap alignment (score calculation) input weights from BWA. Input dataset: HG002 (NA24385) paired-end protocol using Illumina sequencers*

Beyond raw speed, RTX PRO 4500 has up to 4.3x lower power consumption than H100 SXM, while delivering comparable performance for this Smith-Waterman workload.

Learn more

Discover how Parabricks, OpenFold 3, and RTX PRO 4500 Blackwell Server Edition can accelerate your precision medicine journey.

The convergence of accelerated genomics and AI-enabled structural biology is redefining what’s possible in precision medicine, and the pace of progress is only accelerating.

From reducing genome analysis to minutes from hours in an NICU, to generating and experimentally validating protein binders against more than 130 drug targets, the platform enabling this work is no longer a research curiosity; it is made possible by NVIDIA.

• Try Parabricks

• Learn More About RTX PRO 4500 Blackwell Server Edition