Tyler O'Neal, Staff Editor LATEST

Supercomputing reveals hidden galactic architecture around the Milky Way

Leveraging the capabilities of modern high-performance computing (HPC), astronomers have unraveled a cosmic mystery: the Milky Way and its closest neighboring galaxies are embedded in a sprawling sheet of matter that shapes the movement of surrounding galaxies. This breakthrough, featured in Nature Astronomy, was achieved using advanced simulations powered by cutting-edge supercomputers to model the mass distribution and dynamics of our local universe.
 
For decades, cosmologists have grappled with an apparent contradiction in galactic motion. While most galaxies in the universe recede from one another in accord with the expansion described by the Hubble–Lemaître law, our immediate neighborhood, the Local Group comprising the Milky Way, the Andromeda Galaxy, and dozens of dwarf galaxies, exhibits surprisingly coherent motion patterns that ordinary mass distributions failed to explain. The Andromeda Galaxy itself moves toward the Milky Way at about 100 km/s, a phenomenon long understood as gravitational interaction within the Local Group. Yet the behavior of other nearby galaxies did not align with theoretical expectations.
 
Now, an international team led by doctoral researcher Ewoud Wempe and Professor Amina Helmi at the University of Groningen has shown that the key to this puzzle lies not within the confines of the Local Group alone but in an extended, planar mass structure surrounding it. Using sophisticated cosmological simulations constrained by observational data, including the positions, masses, and velocities of 31 galaxies just beyond the Local Group, the researchers demonstrated that the vast majority of dark matter and visible matter in our vicinity is organized in a flat sheet extending tens of millions of light-years. Above and below this planar structure are vast voids with minimal matter.wempe
 
What sets this discovery apart is the critical role of supercomputing in constructing these “virtual twin” universes. The team’s simulations began with initial conditions seeded by early-universe observations and then evolved forward using numerical methods that solve Einstein’s equations of gravity together with fluid dynamics for dark matter and baryonic matter. Such calculations involve millions of interacting elements and demand parallel computation at scale, the exclusive domain of HPC systems. By performing these simulations on powerful supercomputers, astronomers were able to trace the gravitational influence of the large-scale sheet on galaxy motions and verify that this configuration reproduces observed velocities with high fidelity.
 
According to Helmi, this marks the first time that the distribution and velocity field of dark matter in the region surrounding our galactic neighborhood have been quantitatively constrained in a manner consistent with both ΛCDM cosmology and observed local dynamics. “Astronomers have been trying to solve this problem for decades,” Helmi said. “It is extraordinary that, based purely on the motions of galaxies, we can infer a mass distribution that matches the observed positions and motions of galaxies within and just outside the Local Group.”
 
For the supercomputing community, this achievement is profoundly inspirational. It highlights how modern HPC infrastructures, with their massive parallelism, high memory bandwidth, and optimized numerical libraries, are enabling scientists to probe cosmic questions that were once deemed intractable. These simulations not only illuminate the hidden architecture of our cosmic neighborhood but also exemplify how simulation-based science complements observation, allowing researchers to explore scenarios that cannot be directly imaged or measured.
 
Beyond resolving a decades-old enigma in galactic astronomy, this work reinforces the broader scientific view that large-scale structures, from filaments of the cosmic web to planar mass configurations like the one now identified around the Milky Way, are fundamental to understanding the universe’s evolution. Supercomputers are not just tools for speeding up calculations; they are essential engines of discovery that empower scientists to simulate the universe with realism and precision.
 
As supercomputing technology advances, both in terms of hardware and algorithms, scientists are poised to create increasingly detailed “virtual universes.” These sophisticated simulations will not only put our cosmological models to the test but also inform future telescope and space mission observations, leading to a richer understanding of our cosmic context.
 
According to the study’s authors, uncovering the influence of the Local Sheet on galactic motion is more than just resolving a persistent mystery; it demonstrates the remarkable discoveries possible when computational power, observational insights, and scientific curiosity are combined on a cosmic scale.

Supercomputing drives materials breakthrough for green computing: 3D graphene-like electronic behavior unlocks new low-energy electronics

Tyler O'Neal, Staff Editor LATEST
Marking a major advance in sustainable computing, researchers at the University of Liverpool have developed a groundbreaking three-dimensional material that mirrors the remarkable electronic properties of two-dimensional graphene, while offering the durability needed for practical use.
 
Detailed in the journal Matter, this innovation holds the potential to enable greener, more energy-efficient electronics and underscores the essential role of supercomputing in discovering and designing new materials, a development that could reshape the landscape of high-performance and low-power computing.
 
Graphene is a single layer of carbon atoms organized in a honeycomb pattern. This material has fascinated scientists and engineers due to its exceptional electrical, thermal, and mechanical characteristics. Electrons in graphene act like massless Dirac fermions, which allows for extremely fast electron movement with minimal energy loss. Despite these impressive qualities, applying graphene's unique properties to practical, large-scale devices has faced persistent obstacles: its ultra-thin structure is fragile, hard to incorporate into bulk technologies, and expensive to manufacture at scale.
 
The new study addresses this by demonstrating that hafnium tin, HfSn₂, a fully three-dimensional crystal, can mimic graphene’s fast, two-dimensional electron flow. In the HfSn₂ structure, honeycomb layers are arranged in a special chiral stacking pattern that preserves the signature electronic behavior of graphene, specifically, high electron mobility with low energy dissipation, despite the material being fully 3D. This electronic behavior is associated with Weyl points in the material’s band structure, points where conduction and valence bands touch, allowing electrons to move with minimal resistance.
 
These insights emerged from a combination of theoretical modeling, crystallographic simulations, and experimental characterization, and could not have been realized without high-performance computational tools. Supercomputers enable researchers to explore how atomic arrangement, chemical bonding, and quantum mechanical effects interplay across multiple length scales, from electrons to crystals, and to identify Weyl electronic states and transport properties that are inaccessible to simpler computational methods.
 
In particular, density functional theory (DFT) and related ab initio simulation frameworks, inherently computationally intensive, were crucial in predicting how electrons behave within the 3D honeycomb lattice and how different stacking arrangements influence transport. These simulations, typically run on supercomputing clusters equipped with optimized parallel solvers and high memory bandwidth, allow researchers to map out electronic band structures and isolate topological features such as Weyl points with high precision. Without this scale of computation, evaluating the energetic and structural feasibility of such new materials would be prohibitively slow and less reliable.
 
The ability to use supercomputer-driven simulations to screen candidate materials accelerates the discovery process dramatically. Instead of relying solely on costly and time-consuming experimental synthesis of countless samples, researchers can now refine materials candidates through in silico modeling, identifying promising structures that combine desired electronic properties with robustness and environmental resilience.
 
Why does this matter for the green computing agenda? Modern computing systems, from mobile devices to data centers, consume vast amounts of energy. Next-generation logic and spintronic devices (which exploit electron spin as well as charge) require materials that combine low-energy electronic transport with stability under operational conditions. A 3D material that mimics graphene’s electron transport while being easier to integrate into conventional device architectures could lead to significantly lower energy consumption in future information processing and memory technologies, directly addressing sustainability challenges in both artificial intelligence and high-performance computing sectors.
 
Moreover, supercomputing plays a central role beyond discovery; it enables multiscale modeling that connects atomic-scale electronic behavior with device-level performance predictions. By integrating quantum mechanical simulations with larger-scale finite-element and mesoscopic models, researchers can assess how new materials will behave under real operational loads, including temperature variation, stress, and electron-phonon interactions, before ever fabricating a prototype.
 
The discovery of HfSn₂ highlights a compelling convergence of materials science, quantum physics, and high-performance computing. Together, these disciplines are enabling new approaches to energy-efficient electronics. As researchers increasingly rely on supercomputing resources to navigate complex materials landscapes, the pace of breakthroughs aimed at reducing the environmental footprint of computing is expected to accelerate, pointing toward a more sustainable and environmentally responsible digital infrastructure.

Supercomputing’s next frontier: NVIDIA, CoreWeave unite to build the AI factories of tomorrow

Tyler O'Neal, Staff Editor LATEST
In a defining moment for the high-performance computing (HPC) and artificial intelligence (AI) landscape, NVIDIA and CoreWeave have announced an expanded collaboration to accelerate the construction of massive AI factories, purpose-built data centers optimized for large-scale AI workloads. This partnership marks a significant leap forward for the supercomputing community, combining cutting-edge hardware, software innovation, and strategic infrastructure expansion to meet the growing demand for AI compute resources.
 
At the heart of the announcement is a $2 billion investment by NVIDIA in CoreWeave’s Class A common stock, underscoring NVIDIA’s confidence in CoreWeave’s strategy and setting the stage for an ambitious build-out of more than 5 gigawatts of AI-optimized compute capacity by 2030. These facilities, often referred to as AI factories, are expected to become the backbone of next-generation AI research, training, and deployment, offering unprecedented access to accelerated computing for enterprises, startups, and scientific institutions alike.
 
This deepening partnership goes beyond financial backing. Under the expanded agreement, CoreWeave will adopt NVIDIA CPU and storage platforms and deploy multiple generations of NVIDIA accelerated computing architectures across its cloud infrastructure, including future innovations such as the Rubin AI platform, Vera CPUs, and advanced Bluefield storage systems. CoreWeave’s purpose-built software stacks, such as CoreWeave Mission Control and its reference architectures, will be jointly tested and validated to ensure seamless performance at scale.
 
For the supercomputing community, this represents more than a business transaction; it heralds the maturation of an ecosystem where dense GPU clusters, optimized interconnects, and advanced orchestration software come together to deliver supercomputing-class performance for AI workloads. These AI factories will support ultra-large neural network training, complex simulations, and inference tasks that push the limits of parallel processing and memory bandwidth, work that would be inconceivable without HPC-grade infrastructure underpinning the operations.
 
CoreWeave’s CEO, Michael Intrator, encapsulated this vision poignantly in his “The Year AI Gets to Work” blog post: the era of AI is no longer about possibility, but about making it operational at a global scale, powering real-world impact across industries and scientific fields. In his reflection, Intrator emphasized that AI has crossed a crucial threshold where the challenge has shifted from what’s possible to how do we deliver it everywhere it’s needed? This “working” phase requires infrastructure that can meet the relentless pace of innovation, and that is exactly what the expanded collaboration with NVIDIA seeks to enable.
 
What makes this partnership especially noteworthy for HPC practitioners is the tight integration of evolving hardware platforms with cloud-native supercomputing architectures. CoreWeave has been among the first cloud providers to deploy NVIDIA’s advanced GPU platforms, such as the GB200 NVL72 systems, at scale, demonstrating that purpose-built AI infrastructure can rival traditional supercomputer installations in both performance and flexibility. These deployments exemplify how the modern supercomputing stack is increasingly GPU-centric, designed to support massive parallel workloads with efficiency and resilience.
 
Moreover, the collaboration underscores a broader industry trend: the convergence of HPC and AI infrastructure, where the traditional boundaries between scientific computing, enterprise AI, and cloud-native services continue to blur. The AI factories envisioned by NVIDIA and CoreWeave will serve not only core AI model training and inference but also data-intensive simulation tasks, real-time reasoning engines, and agentic AI, workloads that demand HPC-level compute, networking, and orchestration.
 
For the supercomputing community, this development is inspirational on multiple fronts. It validates the central role of accelerated computing architectures in driving the next wave of AI and scientific discovery. It illustrates how deep collaboration between hardware innovators and infrastructure builders can unlock new levels of performance and accessibility. And it signals that the age of supercomputers is expanding from traditional national-lab behemoths into a distributed ecosystem of cloud-native AI super-infrastructure that anyone with visionary applications can tap into.
 
As we enter the AI era, our ability to construct, expand, and make these AI factories widely accessible will shape the years to come. The NVIDIA and CoreWeave partnership stands as a model for realizing these remarkable opportunities.