Amanda Randles

We introduce GPU Halo Replay, a solver-aware in situ framework for stencil-based applications that creates a lossless simulation “twin” for advanced visualization and analysis. By decoupling analysis from the primary simulation and delegating it to dedicated twin simulations, GPU Halo Replay overcomes the synchronization and fidelity limitations of traditional post hoc and in situ methods. We demonstrate the functionality and feasibility of GPU Halo Replay on the computational fluid dynamics code HARVEY, a massively parallel solver. Compared to existing in situ approaches, it is co-designed with stencil solvers and exposes new analysis patterns. To meet the diverse needs of real-world applications, we developed GPU Halo Replay variants tailored to common usage scenarios. We illustrate the adaptability of these variants across three representative case studies: concurrent simulation replay, spatially-targeted subdomain analysis, and temporally selective domain evaluation. These results illustrate the potential of simulation twins to enhance analysis capabilities without sacrificing performance or fidelity.

BACKGROUND AND PURPOSE: Intracranial aneurysms treated with endovascular embolization often require surveillance imaging using DSA, an invasive, risky, and expensive procedure. Existing noninvasive imaging modalities (standard or MRA) are often unsatisfactory for evaluating treated aneurysms due to artifacts from embolization devices. The objective of the present study was to determine whether photon-counting CT (PCCT) imaging parameters could be optimized to confer satisfactory imaging resolution in an anthropomorphic phantom of treated intracranial aneurysms. MATERIALS AND METHODS: Phantom studies were performed using a model of the major intracranial arteries with appropriately sized, endovascularly-treated MCA (coil embolization) and basilar artery (Woven EndoBridge embolization) aneurysms. A series of imaging acquisition procedures was performed using a conventional energy-integrating CT scanner and a PCCT scanner. Key imaging acquisition and reconstruction parameters were varied to identify the optimum protocol for treated aneurysm characterization. Artifact reduction was performed on all images using the iterative metal artifact reduction (iMAR) algorithm (Siemens). Contrast-to-noise ratio (CNR) and metal artifact magnitude were quantitatively analyzed and displayed in tabular form to provide objective criteria for determination of optimal processing parameters for treated aneurysm visualization. RESULTS: Imaging was successfully obtained in phantom studies across a range of imaging parameters. Quantitative metal artifact magnitude was greater for 100 keV virtual monoenergetic images (VMIs) and lowest for 55 keV VMIs without iMAR, but this trend was reversed with iMAR applied. The 55 keV VMI was chosen as the optimal reconstruction parameter for visualization of treated intracranial aneurysms because it demonstrated a low magnitude of metal artifacts and the highest CNR in adjacent vasculature. Similarly, the CNR of the largest vessel adjacent to the coil mass was increased for all images after iMAR was applied. CNR was highest in the 55- keV VMR images both before (3.61 [SD, 0.14]) and after (6.82 [SD, 0.34]) application of iMAR. CONCLUSIONS: Virtual monoenergetic images combined with metal artifact reduction algorithms created from PCCT scans conferred excellent visualization of previously-treated intracranial aneurysms and adjacent vasculature. It was feasible to extend these results to preliminary clinical applications in human patients.

The advent of digital twins has increased the demand for longer-duration simulations that span multiple physiological states. Digital twins have emerged as powerful tools in cardiovascular modeling, enabling patient-specific simulations of coronary blood flow for noninvasive diagnosis and treatment planning. Although these simulations achieve high fidelity under steady or periodic heart rates, modeling real-world transitions, such as those arising from physical activity, requires careful evaluation of temporal convergence, the stabilization of hemodynamic parameters through the simulation of preceding cardiac cycles, or pre-flows. In this study, we present a physiologically grounded approach for determining the minimum number of preceding cardiac pre-flows necessary to achieve temporal convergence following abrupt heart rate (HR) changes. Using high-resolution patient-specific three-dimensional (3D) simulations and inflow waveforms scaled from both synthetic and wearable-derived HR data, we quantify convergence behavior across velocity, pressure gradient, and wall shear stress at both cross-sectional and full-domain levels. Results show that simulating just two pre-flows is sufficient to achieve physiologically stable outputs across high-to-low and low-to-high HR transitions (<2% difference). These findings are further verified using continuous HR data obtained from wearable devices, with low- and high-HR segments extracted to represent natural extremes, confirming the robustness of the proposed convergence criterion under real-world dynamic inputs (<1% difference). This work establishes a computationally efficient and physiologically consistent criterion for dynamic-state simulations, facilitating the integration of cardiovascular digital twins with real-time sensing technologies.

Digital twins offer a promising approach to advancing healthcare by providing precise, noninvasive monitoring and early detection of diseases. In heart failure (HF), a leading cause of mortality worldwide, they can improve patient monitoring and clinical outcomes by simulating hemodynamic changes indicative of worsening HF. Current techniques are limited by their invasiveness and lack of scalability. We present a novel framework for HF digital twins that predicts patient-specific hemodynamic metrics in the pulmonary arteries using 3D computational fluid dynamics to address these limitations. We introduce a strategy to determine the minimal geometric complexity required for accurate pressure prediction and explore the effects of varying boundary conditions. By validating our digital twins against invasively-measured data, we demonstrate their potential to improve HF management by enabling continuous, noninvasive monitoring and early identification of worsening HF. This proof-of-concept study lays the groundwork for integrating digital twin technology into personalized HF care.

Introduction: Aortic coarctation (CoA) comprises 6%-8% of all congenital heart diseases and is the second most common cardiovascular disease requiring neonatal surgical correction. However, patients remain at high risk for long-term complications, notably recoarctation. Methods: Hemodynamic simulations were performed in a group of six patients following CoA repair, as compared to a group of age and sex-matched healthy controls. Progressive narrowing at the CoA repair site was modeled to simulate the recoarctation process. Key measurements included time-averaged wall shear stress (TAWSS) in the aortic arch and CoA repair site. Results: 0.05). A pronounced nonlinear relationship between stenosis severity and TAWSS was observed suggesting that increasing stenosis corresponds to progressively abnormal shear stress. Discussion: The persistent high TAWSS in CoA-repaired aortas may underlie the poor long-term outcomes observed in this population. The identified nonlinear relationship between stenosis severity and TAWSS magnitude suggests a potential positive feedback mechanism, where abnormal shear stress exacerbates pathologic remodeling in the repaired aorta, highlighting the potential role of hemodynamic simulations in the clinical management of CoA patients.

Provisional stenting is the standard treatment for coronary bifurcation lesions, relying on real-time assessment of side-branch (SB) hemodynamics to guide intervention. Functional metrics such as fractional flow reserve (FFR) and instantaneous wave-free ratio (iFR) are used for this purpose. Still, their application in bifurcation lesions is limited by procedural complexity and the lack of preoperative planning tools. We developed a simulation-based machine learning (ML) framework to predict iFR under resting, and FFR under hyperemic conditions. The framework leveraged a synthetic hemodynamic dataset of 252 bifurcation lesions generated from 7 patient-specific geometries using HARVEY, a massively parallel computational fluid dynamics (CFD) solver. Anatomical variability was incorporated using four linear mixed-effects (LME) models to establish robust predictions. Two clinically relevant data-splitting strategies were evaluated: (1) a scenario excluding untreated cases, simulating models blind to new geometries, and (2) a scenario incorporating matched untreated cases, reflecting real-world conditions where preoperative anatomical data is available. Morphological features, including lesion severity, length, and curvature, were systematically varied alongside inherited anatomical parameters like bifurcation angle and side-branch count. Splitting approach 2 demonstrated superior predictive performance, achieving a maximum diagnostic accuracy of 0.847 (AUC: 0.899) for FFR and 0.797 (AUC: 0.874) for iFR. Mixed-effects models effectively account for patient-specific anatomical variability, with Bland-Altman analyzes confirming minimal bias between CFD and ML predictions. Incorporating preoperative anatomical information reduced variability and improved diagnostic accuracy across the studied thresholds. The proposed ML framework offers precise, real-time functional assessments of SB hemodynamics, reducing procedural uncertainty in provisional stenting strategies using only pre-operative lesion-specific features and a precomputed synthetic hemodynamic dataset of 252 bifurcation lesion instances. Using synthetic data sets and patient-specific anatomical insights, this approach paves the way for personalized coronary intervention planning, bridging the gap between computational modeling and clinical applicability.

Provisional stenting is the standard treatment for coronary bifurcation lesions, relying on real-time assessment of side-branch (SB) hemodynamics to guide intervention. Functional metrics such as fractional flow reserve (FFR) and instantaneous wave-free ratio (iFR) are used for this purpose. Still, their application in bifurcation lesions is limited by procedural complexity and the lack of preoperative planning tools. We developed a simulation-based machine learning (ML) framework to predict iFR under resting, and FFR under hyperemic conditions. The framework leveraged a synthetic hemodynamic dataset of 252 bifurcation lesions generated from 7 patient-specific geometries using HARVEY, a massively parallel computational fluid dynamics (CFD) solver. Anatomical variability was incorporated using four linear mixed-effects (LME) models to establish robust predictions. Two clinically relevant data-splitting strategies were evaluated: (1) a scenario excluding untreated cases, simulating models blind to new geometries, and (2) a scenario incorporating matched untreated cases, reflecting real-world conditions where preoperative anatomical data is available. Morphological features, including lesion severity, length, and curvature, were systematically varied alongside inherited anatomical parameters like bifurcation angle and side-branch count. Splitting approach 2 demonstrated superior predictive performance, achieving a maximum diagnostic accuracy of 0.847 (AUC: 0.899) for FFR and 0.797 (AUC: 0.874) for iFR. Mixed-effects models effectively account for patient-specific anatomical variability, with Bland-Altman analyzes confirming minimal bias between CFD and ML predictions. Incorporating preoperative anatomical information reduced variability and improved diagnostic accuracy across the studied thresholds. The proposed ML framework offers precise, real-time functional assessments of SB hemodynamics, reducing procedural uncertainty in provisional stenting strategies using only pre-operative lesion-specific features and a precomputed synthetic hemodynamic dataset of 252 bifurcation lesion instances. Using synthetic data sets and patient-specific anatomical insights, this approach paves the way for personalized coronary intervention planning, bridging the gap between computational modeling and clinical applicability.

Long-term plans and comprehensive vision are needed.

More than four decades after its release, the 1982 Lax Report remains a landmark blueprint for U.S. high-performance computing (HPC) policy. This retrospective revisits the report’s four principal recommendations, evaluating each on its current fulfillment. While the HPC ecosystem has matured through network advances, broader access, and the integration of AI and distributed computing, gaps in workforce development, scientific software sustainability, and architectural innovation tailored to science needs remain. This article traces HPC’s evolution from centralized supercomputers to today’s cloud-augmented, AI-converged landscape and highlights how market-driven hardware design increasingly diverges from the needs of rigorous scientific computing. The paper also calls for sustained public investment in hardware-software co-design, workforce pipelines, and open, science-centered infrastructure.

As biological research demands simulations with increasingly larger cell counts, optimizing these models for largescale deployment on heterogeneous supercomputing resources becomes crucial. This requires the redesign of fluid-structure interaction tasks written around distributed data structures built for CPU-based systems, where design flexibility and overall memory footprint are key considerations, to instead be performant on CPU-GPU machines. This paper describes the trade-offs of offloading communication tasks to the GPUs and the corresponding changes to the underlying data structures required, along with new algorithms that significantly reduce time-to-solution. At scale performance of our GPU implementation is evaluated on the Polaris and Frontier leadership systems. Real-world workloads involving millions of deformable cells are evaluated. We analyze the competing factors that come into play when designing a communication layer for a fluid-structure interaction code, including code efficiency, complexity, and GPU memory demands, and offer advice to other high performance computing applications facing similar decisions.

Introduction: Acute type A aortic dissection (ATAAD) is typically treated by replacement of the ascending aorta (+/- root) and proximal arch. However, 70-85% of patients have residual distal dissection post-repair, and 20-40% require late reoperation for aneurysmal degeneration of the distal aorta (ADDA). Since an individual patient’s risk of ADDA cannot be accurately predicted, current guidelines recommend lifelong aortic surveillance imaging for all patients. Hypothesis: Computational fluid dynamics (CFD) simulations of aortic hemodynamics post-repair can accurately identify patients at late risk of ADDA. Methods: We performed CFD simulations of 50 patients following hemi-arch replacement for ATAAD. Patient-specific 3D models were generated from the aortic root to iliac bifurcation (including arch branches) from postoperative 0.6mm contrast-enhanced CT angiograms taken <1 year after index repair (Figure). Exclusion criteria were known heritable thoracic aortic disease and absence of residual dissection. The primary outcome was ADDA, defined as late growth of the distal arch/descending thoracic aorta (DTA) to a diameter ≥5.5cm. Hemodynamic simulations were run for 6 cardiac cycles on a high-performance computing cluster using HARVEY, a CFD solver implementing the lattice Boltzmann method. The primary hemodynamic metric was time-averaged wall shear stress (TAWSS) ratio between the false and true lumens. Results: ADDA developed in 22 patients (44%) at a mean of 3.2 years postoperatively. There were no significant clinical differences between those with and without ADDA (Table). The development of late aneurysm growth was significantly associated with a higher TAWSS ratio in the proximal DTA ( p <0.05, Figure). Conclusion: ADDA following hemi-arch repair for ATAAD is associated with significantly higher false lumen TAWSS as early as the first surveillance scan. CFD simulations may help clinicians risk-stratify patients years before they meet reoperation criteria.

Understanding the evolving nature of coronary hemodynamics is crucial for early disease detection and monitoring progression. We require digital twins that mimic a patient’s circulatory system by integrating continuous physiological data and computing hemodynamic patterns over months. Current models match clinical flow measurements but are limited to single heartbeats. To this end, we introduced the longitudinal hemodynamic mapping framework (LHMF), designed to tackle critical challenges: (1) computational intractability of explicit methods; (2) boundary conditions reflecting varying activity states; and (3) accessible computing resources for clinical translation. We show negligible error (0.0002–0.004%) between LHMF and explicit data of 750 heartbeats. We deployed LHMF across traditional and cloud-based platforms, demonstrating high-throughput simulations on heterogeneous systems. Additionally, we established LHMF C , where hemodynamically similar heartbeats are clustered to avoid redundant simulations, accurately reconstructing longitudinal hemodynamic maps (LHMs). This study captured 3D hemodynamics over 4.5 million heartbeats, paving the way for cardiovascular digital twins.

Chronic limb-threatening ischemia (CLTI) is one of the leading causes of permanent disability and death in the United States and around the world. Surgical revascularization is the mainstay of treatment for CLTI, but patients still suffer high rates of limb loss and early death, even compared to other cardiovascular diseases. One significant reason for this is the difficulty of accurately predicting whether a planned intervention will, in fact, improve distal blood flow as hoped. Given this need, we have introduced an angiography-based computational tool to enable surgeons to preoperatively assess a given revascularization strategy, before selecting the option that maximizes distal perfusion. We describe this computational model and demonstrate how it can accurately predict distal arterial flow in a pilot study of patients undergoing femoral artery angioplasty for CLTI.

Coronary artery disease (CAD) is the most common form of cardiovascular disease, characterized by gradual narrowing of the artery walls due to plaque buildup. Computational fluid dynamics (CFD) is a non-invasive approach often used to investigate how these anatomical changes perturb local hemodynamics and contribute to the pathological mechanism of progression. Therefore, the accuracy of coronary tree alignment and anatomical feature detection is key to understanding these hemodynamically mediated mechanisms. Despite advances, current methods face challenges, such as the need for manual selection of landmarks, often resulting in a semi-automated experience. This study aims to improve this by developing a fully automated system to detect 3D anatomical characteristics and align coronary tree geometries in large clinical datasets. Our proposed algorithm enables full automatic placement of the corresponding centerline points and alignment evaluation through similarity-based assessment of Jaccard index (intersection over union) in a cohort of 73 coronary geometries.

Computational models have emerged as a powerful tool to non-invasively monitor vital biomarkers in coronary artery disease, providing a viable alternative to conventional invasive methods and improving clinical decision-making. The precision of these in silico models, however, is often counter-balanced by their substantial computational cost. Pulsatile flow conditions, which closely mimic physiological conditions, are typically employed in computational fluid dynamics (CFD) simulations to capture dynamic changes in hemodynamic variables, but come at a significant computational cost. This study addresses the critical question of the necessity for complex pulsatile models versus the adequacy of simpler steady-state models to capture hemodynamic metrics such as fractional flow reserve, velocity, vorticity, and wall shear stress within the cardiac cycle. By comparing steady-state and pulsatile flow simulations in 12 patients with stenosed coronary arteries, our research evaluates whether steady-state simulations can accurately reflect patient-specific hemodynamic profiles at key clinical moments. The results indicate a comparability of metric magnitudes and distributions between the two types of simulation, particularly during diastole, with minimal influence from the choice of inlet waveform. This suggests that for metrics such as fractional flow reserve (FFR), steady-state simulations are sufficiently accurate and markedly reduce computational load, whereas pulsatile simulations may be necessary for capturing complex dynamics at systole. This distinction underscores the potential for selective application of steady-state models in clinical practice, facilitating the integration of computational fluid dynamics modeling by identifying when the reduced complexity of steady-state simulations can effectively support patient care in coronary artery disease.

Sickle cell disease encompasses a variety of inherited red blood cell (RBC) disorders characterized by abnormal thrombosis, microvascular occlusion, end-organ ischemia, and early mortality. Understanding how sickle RBCs drive abnormal blood flow and stress on the endothelial wall is essential to predict and prevent blockages in blood circulation. While there are studies comparing blood flow velocity and pressure via computational fluid dynamics simulations, there are still open questions about how sickle cells interact with plasma in a complex capillary network. In order to quantify the hemodynamic differences between normal and sickle cells, we introduced a sickle cell RBC model to massively parallel fluid-structure interaction software HARVEY. Notably, sickle RBCs exhibit increased margination, aggregation at the inner curvature, slower fluid velocity, higher pressure, and greater wall shear stress at standard hematocrit levels. This computational model facilitates detailed cellular modeling for hemodynamic simulations in complex capillary networks, offering predictive insight into blockage and potential vessel ruptures in patients with sickle cell disease.

Continuously measured arterial blood velocity can provide insight into physiological parameters and potential disease states. The efficient and effective description of the temporal profiles of arterial velocity is crucial for both clinical practice and research. We propose a pipeline to identify the minimum number of points of interest to adequately describe a velocity profile of the left coronary artery. This pipeline employs a novel operation that "stretches" a baseline waveform to quantify the utility of a point in fitting other waveforms. Our study introduces a comprehensive pipeline specifically designed to identify the minimal yet crucial number of points needed to accurately represent the velocity profile of the left coronary artery. Additionally, the only location-dependent portion of this pipeline is the first step, choosing all of the possible points of interest. Hence, this work is broadly applicable to other waveforms. This versatility paves the way for a novel non-frequency domain method that can enhance the analysis of physiological waveforms. Such advancements have potential implications in both research and clinical treatment of various diseases, underscoring the broader applicability and impact.

Although it is well established that hemodynamics can significantly influence the location and progression of cardiovascular disease (CVD), and 3D blood flow metrics are recognized as potential diagnostic indicators of these diseases, the dynamics of these metrics over time and their variation at different activity levels are not well understood. A relevant example is the impact of exercise on the vascular system over extended periods. Although exercise is widely recognized as a preventive measure of heart disease, the specifics of how activity levels and subsequent alterations in blood flow contribute to the mechanisms that drive CVD remain unclear. In this study, we used a digital coronary twin to establish a longitudinal hemodynamic map (LHM) of the rest and exercise states. An hour-long dataset for both the rest and exercise states, acquired from a wearable device for a single patient, was used to drive a complex 3D fluid dynamics simulation. Hemodynamic metrics such as maximum velocity, average velocity, maximum wall shear stress, average wall shear stress, time-averaged wall shear stress, and pressure gradient were compared between the two states. This analysis represents an initial step toward understanding how long-term exercise regimens can influence hemodynamic changes and potentially reduce the risk of cardiovascular disease. Our findings revealed that the maximum wall shear stress exhibited the highest sensitivity to changes in activity level, while the pressure gradient showed the least variability. This study contributes significantly to quantifying how 3D blood flow metrics differ between rest and active states, providing valuable insight regarding exercise-induced hemodynamic alterations and their potential role in mitigating CVD risk.

Background Computational fluid dynamics can compute fractional flow reserve (FFR) accurately. However, existing models are limited by either the intravascular hemodynamic phenomarkers that can be captured or the fidelity of geometries that can be modeled. Methods and Results This study aimed to validate a new coronary angiography‐based FFR framework, FFR HARVEY , and examine intravascular hemodynamics to identify new biomarkers that could augment FFR in discerning unrevascularized patients requiring intervention. A 2‐center cohort was used to examine diagnostic performance of FFR HARVEY compared with reference wire‐based FFR (FFR INVASIVE ). Additional biomarkers, longitudinal vorticity, velocity, and wall shear stress, were evaluated for their ability to augment FFR and indicate major adverse cardiac events. A total of 160 patients with 166 lesions were investigated. FFR HARVEY was compared with FFR INVASIVE by investigators blinded to the invasive FFR results with a per‐stenosis area under the curve of 0.91, positive predictive value of 90.2%, negative predictive value of 89.6%, sensitivity of 79.3%, and specificity of 95.4%. The percentage ofdiscrepancy for continuous values of FFR was 6.63%. We identified a hemodynamic phenomarker, longitudinal vorticity, as a metric indicative of major adverse cardiac events in unrevascularized gray‐zone cases. Conclusions FFR HARVEY had high performance (area under the curve: 0.91, positive predictive value: 90.2%, negative predictive value: 89.6%) compared with FFR INVASIVE . The proposed framework provides a robust and accurate way to compute a complete set of intravascular phenomarkers, in which longitudinal vorticity was specifically shown to differentiate vessels predisposed to major adverse cardiac events.

We were given a charge to assess the necessity for new or upgraded facilities to ensure the Office of Science (SC) remains at the forefront of scientific discovery.This effort included evaluating five specific ASCR facilities for their potential to contribute to this goal and to rate the readiness for construction of each. Our analysis led us to three overarching recommendations:Recommendation 1: Ensure the continued support and development of all five ASCR computational facilities reviewed-ALCF, OLCF, NERSC, HPDF, and ESnet-as they are central and essential to all SC science programs and broader national science and engineering research programs.Each facility provides distinct and critical functionality that are essential to achieve SC science goals.Significant R&D investment is necessary to sustain their crucial roles.A summary of our findings can be found in Table 1.Recommendations on individual facilities are found in Section 3.Recommendation 2: Science demands integration.We advocate viewing ASCR facilities not as isolated entities, but as integral components of a single, larger integrated computational ecosystem (henceforth referred to as Ecosystem), with a single governance model.This effort will require new ways of governing and potentially funding the overall Ecosystem, which should not be developed via individual site procurements.Rather it should be designed, developed, built, and operated as an integrated facility ecosystem for DOE science.It is critical for supporting SC science programs, along with additional software, algorithm, workforce, and science application components, to serve science and engineering research.Further, this integrated ecosystem is required for programs of other agencies, and industry.Its critical role in bolstering national scientific and technological capabilities, as well as its status as a model internationally, cannot be overstated.Recommendation 3: A comprehensive, coordinated R&D program delivering multiple prototype computing systems over a five-year timescale must be mounted to inform

As medical technologies advance with increasing speed, virtual imaging trials (VITs) are emerging as a crucial tool in the evaluation and optimization of new imaging techniques. Widely used in many VITs is the four-dimensional extended cardiac-torso (XCAT) phantom, a comprehensive computational model that accurately represents human anatomy and physiology. While the XCAT phantom offers a powerful tool for imaging research, it offers only a limited model of blood flow to compartmentalized organs, potentially limiting the realism and clinical applicability of contrast-enhanced scan simulations. This study bridges that gap by combining realistic CT simulation with an accurate model of blood flow dynamics to enable more realistic simulations of contrast-enhanced imaging. To achieve this, a validated one-dimensional blood flow simulator, HARVEY1D, was used to model flow throughout the vessels of the XCAT phantom. DukeSim, a validated CT simulation platform, was then modified to incorporate the resulting flow into its simulations, thus enabling the genera􀆟on of simulated CT scans reflective of real-world blood-based contrast-enhanced imaging scenarios. To demonstrate the utility of this pipeline in an initial application to cardiac imaging, three heart models were studied: a non-diseased model, a 50% stenosis model, and an 80% stenosis model. Three seconds of contrast propagation were tracked in each heart model, and CT scans corresponding to two timepoints were simulated. Results demonstrated that the presence of stenosis significantly impacted blood flow, with greater resistance to blood flow leading to altered flow patterns visible in the simulated CT images. This work showcases a pipeline that leverages both computational fluid dynamics and medical imaging simulations to enhance the realism of virtual imaging trials and facilitate the evaluation, optimization, and development of diagnostic tools for contrast-enhanced imaging.

The lattice Boltzmann method is a highly scalable Navier-Stokes solver that has been applied to flow problems in a wide array of domains. However, the method is bandwidth-bound on modern GPU accelerators and has a large memory footprint. In this paper, we present new 2D and 3D GPU implementations of two different regularized lattice Boltzmann methods, which are not only able to achieve an acceleration of ∼ 1.4 × w.r.t. reference lattice Boltzmann implementations but also reduce the memory requirements by up to 35% and 47% in 2D and 3D simulations respectively. These new approaches are evaluated on NVIDIA and AMD GPU architectures.

Preparing for the deployment of large scientific and engineering codes on upcoming exascale systems with GPU-dense nodes is made challenging by the unprecedented diversity of device architectures and heterogeneous programming models. In this work, we evaluate the process of porting a massively parallel, fluid dynamics code written in CUDA to SYCL, HIP, and Kokkos with a range of backends, using a combination of automated tools and manual tuning. We use a proxy application along with a custom performance model to inform the results and identify additional optimization strategies. At scale performance of the programming model implementations are evaluated on pre-production GPU node architectures for Frontier and Aurora, as well as on current NVIDIA device-based systems Summit and Polaris. Real-world workloads representing 3D blood flow calculations in complex vasculature are assessed. Our analysis highlights critical trade-offs between code performance, portability, and development time.

Tracking hemodynamic responses to treatment and stimuli over long periods remains a grand challenge. Moving from established single-heartbeat technology to longitudinal profiles would require continuous data describing how the patient's state evolves, new methods to extend the temporal domain over which flow is sampled, and high-throughput computing resources. While personalized digital twins can accurately measure 3D hemodynamics over several heartbeats, state-of-the-art methods would require hundreds of years of wallclock time on leadership scale systems to simulate one day of activity. To address these challenges, we propose a cloud-based, parallel-in-time framework leveraging continuous data from wearable devices to capture the first 3D patient-specific, longitudinal hemodynamic maps. We demonstrate the validity of our method by establishing ground truth data for 750 beats and comparing the results. Our cloud-based framework is based on an initial fixed set of simulations to enable the wearable-informed creation of personalized longitudinal hemodynamic maps.