🚀 AI Software Libraries for Scientific Research from NVIDIA!

[Hasan S. Cemkan]

The world of science is transforming at full speed with artificial intelligence (AI)! NVIDIA is leading this transformation by introducing new software tools to accelerate scientific workflows. Announced at the ISC High Performance Conference in Hamburg, these innovations are poised to revolutionize a wide range of fields, from quantum chemistry to materials science, astrophysics to life sciences.

The new libraries include NVIDIA DAQIRI, the upcoming NVIDIA cuPhoton reference code, and NVIDIA ALCHEMI NIM microservices. These tools are designed to transform CPU-based workloads that take hours into real-time, GPU-accelerated processes. All these technologies integrate seamlessly into the NVIDIA CUDA-X suite, designed for high-performance computing (HPC) and enterprise AI applications.

─────────────────────────

🔭 Observational Astrophysics and Data Flow Accelerates​

In experimental astronomy, the soon-to-be-released NVIDIA cuPhoton reference code offers a specialized software framework for extracting information from high-dimensional, multi-dimensional datasets collected by telescopes, laser experiments, and X-ray instruments. Running on NVIDIA GB200 NVL72 architectures, this reference code accelerates the loading, reading, processing, and visualization of petabytes of Flexible Image Transport System (FITS) data.

In early access evaluations with images from the Rubin Observatory's Legacy Survey of Space and Time (LSST), cuPhoton accelerated FITS file ingestion by 14,900x. When scaled to 32 NVIDIA Grace Blackwell superchips, signal processing and analysis throughput increased by 8,400x. Developed in collaboration with Princeton University and used with Harvard University, this reference code enables rapid processing for large-scale dark energy and astrophysical research.

─────────────────────────

⚡ Real-Time Data Capture: DAQIRI​

Concurrently, NVIDIA introduced DAQIRI (Data Acquisition for Real-Time Instruments with Data Integration), a high-performance networking library developed within the NVIDIA Holoscan Platform. Traditional hardware-centric systems often drop data packets when fast physical sensors or detectors produce information faster than local memory arrays can perform disk recording operations.

DAQIRI overcomes this limitation by managing high-bandwidth sensor data streams in real-time as they arrive. This library is being used in the A-GHOST research project, a collaboration between CERN, the University of Chicago, and University College London within the CERN openlab framework. The system applies real-time AI inference to collision data captured by the ATLAS Experiment, capturing potentially significant signals from over 99% of the dataset that would normally be discarded due to storage constraints.

─────────────────────────

🔬 ALCHEMI Revolutionizing Chemical and Atomistic Simulations​

NVIDIA ALCHEMI consists of domain-specific NIM microservices and a dedicated developer toolkit designed to optimize chemical and materials discovery for use cases such as battery materials, catalysts, organic light-emitting diodes (OLEDs), and consumer beauty products. In March 2026, the company released two specialized ALCHEMI NIM microservices designed to process millions of materials simultaneously using machine learning interatomic potentials (MLIPs):

• 
  []Batched Geometry Relaxation (BGR): An AI-accelerated container that determines the most stable geometry configuration for a group of molecules.
  
  []Batched Molecular Dynamics (BMD): A microservice that hosts models like MACE-MPA-0 to simulate how atomic arrangements and molecular structures change over time.
  

ALCHEMI will also integrate a dedicated microservice for the Vienna Ab initio Simulation Package (VASP), allowing developers to perform multiple density functional theory calculations on a single GPU via NVIDIA Multi-Process Service. This microservice provides a 3x throughput increase for geometry optimization steps. Additionally, researchers can use the ALCHEMI Toolkit and Toolkit-Ops repository to train custom MLIP surrogate models and coordinate atomistic simulation workflows.

─────────────────────────

⚙️ Autonomous Life Sciences and Lila Sciences Collaboration​

Lila Sciences, a developer of scientific superintelligence platforms and autonomous laboratory systems, has integrated the full NVIDIA computing stack into its simulation loop. Collaborating with NVIDIA on a high-fidelity magnet simulation presented at the GTC San Jose conference in March 2026, Lila Sciences used the ALCHEMI BGR microservice to accelerate high-throughput materials screening by 50x to isolate stable chemical compositions.

The company then implemented the early access ALCHEMI VASP microservice to accelerate the calculation of complex magnetic behaviors for shortlisted compounds by 30%. Leveraging ALCHEMI's kernels specifically designed for TensorNet architectures, Lila Sciences achieved a 6x speedup in co-training and inference loops while reducing native memory footprint requirements by three times. The platform supports broad exploration cycles in materials science, catalysis, and electromagnetics by evaluating multiple compositions directly in GPU memory.

For end-to-end model training, inference serving, and digital twin management, the autonomous laboratory platform includes NVIDIA Megatron-LM, Nemotron 3 Nano, Nemotron 3 Super, NeMo RL, BioNeMo, Triton Inference Server, and NVIDIA Omniverse libraries. Andy Beam, Co-founder and CTO of Lila Sciences, stated that the combined computing stack enables scientific discovery at a volumetric scale unattainable with unaccelerated individual research methods.

─────────────────────────

💡 Data Acquisition (DAQ) and Next-Generation Approaches​

Data acquisition (DAQ) in high-energy physics and experimental astrophysics involves the flow of large volumes of raw data from physical detectors directly into digital computing infrastructure. In experiments like the ATLAS detector at CERN, particle collisions occur at sub-nanosecond intervals, producing multi-terabit streams of analog detector outputs. These must be immediately digitized and filtered. Traditional DAQ systems route these packets through layers of field-programmable gate arrays (FPGAs) and central processing units (CPUs).

This standard TCP/IP networking architecture copies data multiple times through the Linux kernel networking stack, creating significant CPU bottlenecks and high latency. This causes the system to discard over 99% of raw events through coarse, hardware-based trigger thresholds. NVIDIA DAQIRI eliminates these software bottlenecks by implementing a software-defined architecture based on Data Plane Development Kit (DPDK) and GPUDirect Remote Direct Memory Access (RDMA) technologies. By using kernel bypass mechanisms, DAQIRI provides user-space applications direct access to the ring buffers of ConnectX network interface cards (NICs), completely removing the operating system kernel from the data path.

The software can split incoming packets via a Header-Data Split mode, where packet headers are directed to the main CPU for initial network validation, while the large scientific data payload is copied directly into the Direct Memory Access (DMA) buffers of GPU memory via the PCIe bus without any intermediate host-side buffering.

Once in GPU memory, raw byte arrays are directly formatted into GPU tensors. This allows real-time deep learning models, such as convolutional autoencoders or temporal convolutional networks, to be executed in-stream.

─────────────────────────

These new NVIDIA software libraries are revolutionizing the speed and efficiency of scientific research, making discovery processes more accessible and real-time. Thanks to these tools, scientists will be able to access previously unreachable data and solve complex problems faster. These developments lay a solid foundation for future scientific breakthroughs.