NVIDIA XANI Cuts X-Ray Data Processing Time to Hours

NVIDIA has unveiled a major breakthrough in nanoscale imaging with its Accelerated X-ray Analysis for Nanoscale Imaging (XANI) workflow. Using its Grace Blackwell Superchips, the company has cut down data processing time for X-ray free-electron laser (XFEL) facilities from nine months to under four hours—an improvement of over 1,000x.

XFEL facilities, such as LCLS-II in the U.S. and European XFEL in Germany, generate massive datasets while probing the atomic and electronic dynamics of advanced materials like semiconductors, batteries, and catalysts. These facilities produce up to 1 million X-ray pulses per second, capturing structural shifts at the atomic level in real time. However, analyzing the resulting terabytes of multidimensional data has traditionally been a computational bottleneck.

NVIDIA’s XANI solution leverages the GB200 Grace Blackwell Superchips to accelerate this process. By combining GPU-based processing with CUDA Python and distributed computing, the team compressed the analysis of 42 terabytes of data to under four hours while maintaining precision. This is a stark contrast to traditional CPU-bound workflows, which often process just 10% of a dataset during experiments.

Key Innovations in XANI

Several technical advancements underpin XANI’s performance:

• GPU Acceleration: XANI achieved a 43x speedup on a single GPU and a 1,000x boost on 64 GPUs compared to earlier CPU-based methods.
• cuPyNumeric Libraries: New libraries, like LMFIT and multithreaded HDF5, improved GPU utilization and enabled 165x faster I/O throughput.
• GPUDirect Storage (GDS): By directly loading data into GPU memory, XANI bypasses CPU bottlenecks, enabling read speeds of up to 700GB/s across 16 Grace Blackwell nodes.

The workflow also introduces a distributed memory architecture that simplifies scientific computing. By swapping NumPy imports for cuPyNumeric, researchers can automatically parallelize operations across clusters without writing complex MPI code. This makes XANI accessible to fields beyond physics, including materials chemistry and quantum computing.

Scaling for Next-Gen Research

The XANI architecture is designed for scalability. With its GPU-centric distributed model, scientists can now analyze data in real time, providing live feedback during experiments. This capability could redefine how XFEL facilities operate, reducing delays between data collection and actionable insights.

Thanks to advances in nonlinear least-squares algorithms and batched GPU computation, XANI can process high-resolution imaging data down to the pixel level. The workflow’s ability to fit damped oscillations to detector data in parallel ensures faster and more precise results than ever before.

Implications for Scientific Discovery

NVIDIA’s XANI workflow represents a paradigm shift for high-performance computing in scientific research. By reducing analysis times from months to hours, it accelerates discoveries in materials science, quantum physics, and beyond. XFEL facilities worldwide now stand to benefit from these efficiencies, unlocking new opportunities for real-time experimentation.

For researchers, the implications are clear: advanced GPU-based systems like Grace Blackwell Superchips are becoming indispensable tools in tackling the data challenges of modern science.

Image source: Shutterstock