arXiv subjects

Rick Wagner

Publications and source records attributed to Rick Wagner.

14 recordsLinked to original sources

arXiv2025

Serving LLMs in HPC Clusters: A Comparative Study of Qualcomm Cloud AI 100 Ultra and NVIDIA Data Center GPUs

This study presents a benchmarking analysis of the Qualcomm Cloud AI 100 Ultra (QAic) accelerator for large language model (LLM) inference, evaluating its energy efficiency (throughput per watt), performance, and hardware scalability against NVIDIA A100 GPUs (in 4x and 8x configurations) within the National Research Platform (NRP) ecosystem. A total of 12 open-source LLMs, ranging from 124 million to 70 billion parameters, are served using the vLLM framework. Our analysis reveals that QAic achieves competitive energy efficiency with advantages on specific models while enabling more granular hardware allocation: some 70B models operate on as few as 1 QAic card versus 8 A100 GPUs required, with 20x lower power consumption (148W vs 2,983W). For smaller models, single QAic devices achieve up to 35x lower power consumption compared to our 4-GPU A100 configuration (36W vs 1,246W). The findings offer insights into the potential of the Qualcomm Cloud AI 100 Ultra for energy-constrained and resource-efficient HPC deployments within the National Research Platform (NRP).

cs.DC↗

arXiv2020

Targeting SARS-CoV-2 with AI- and HPC-enabled Lead Generation: A First Data Release

Researchers across the globe are seeking to rapidly repurpose existing drugs or discover new drugs to counter the the novel coronavirus disease (COVID-19) caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). One promising approach is to train machine learning (ML) and artificial intelligence (AI) tools to screen large numbers of small molecules. As a contribution to that effort, we are aggregating numerous small molecules from a variety of sources, using high-performance computing (HPC) to computer diverse properties of those molecules, using the computed properties to train ML/AI models, and then using the resulting models for screening. In this first data release, we make available 23 datasets collected from community sources representing over 4.2 B molecules enriched with pre-computed: 1) molecular fingerprints to aid similarity searches, 2) 2D images of molecules to enable exploration and application of image-based deep learning methods, and 3) 2D and 3D molecular descriptors to speed development of machine learning models. This data release encompasses structural information on the 4.2 B molecules and 60 TB of pre-computed data. Future releases will expand the data to include more detailed molecular simulations, computed models, and other products.

q-bio.BM↗

arXiv2014

IVOA Recommendation: Simulation Data Model

In this document and the accompanying documents we describe a data model (Simulation Data Model) describing numerical computer simulations of astrophysical systems. The primary goal of this standard is to support discovery of simulations by describing those aspects of them that scientists might wish to query on, i.e. it is a model for meta-data describing simulations. This document does not propose a protocol for using this model. IVOA protocols are being developed and are supposed to use the model, either in its original form or in a form derived from the model proposed here, but more suited to the particular protocol. The SimDM has been developed in the IVOA Theory Interest Group with assistance of representatives of relevant working groups, in particular DM and Semantics.

astro-ph.IM↗

arXiv2013

Energy cascade and scaling in supersonic isothermal turbulence

Supersonic turbulence plays an important role in a number of extreme astrophysical and terrestrial environments, yet its understanding remains rudimentary. We use data from a three-dimensional simulation of supersonic isothermal turbulence to reconstruct an exact fourth-order relation derived analytically from the Navier-Stokes equations (Galtier and Banerjee, Phys. Rev. Lett., vol. 107, 2011, p. 134501). Our analysis supports a Kolmogorov-like inertial energy cascade in supersonic turbulence previously discussed on a phenomenological level. We show that two compressible analogues of the four-fifths law exist describing fifth- and fourth-order correlations, but only the fourth-order relation remains `universal' in a wide range of Mach numbers from incompressible to highly compressible regimes. A new approximate relation valid in the strongly supersonic regime is derived and verified. We also briefly discuss the origin of bottleneck bumps in simulations of compressible turbulence.

astro-ph.GA↗

arXiv2012

Profiles of Dark Matter Velocity Anisotropy in Simulated Clusters

We report statistical results for dark matter (DM) velocity anisotropy, β, from a sample of some 6000 cluster-size halos (at redshift zero) identified in a ΛCDM hydrodynamical adaptive mesh refinement simulation performed with the Enzo code. These include profiles of β in clusters with different masses, relaxation states, and at several redshifts, modeled both as spherical and triaxial DM configurations. Specifically, although we find a large scatter in the DM velocity anisotropy profiles of different halos (across elliptical shells extending to at least ~$1.5 r_{vir}$), universal patterns are found when these are averaged over halo mass, redshift, and relaxation stage. These are characterized by a very small velocity anisotropy at the halo center, increasing outward to about 0.27 and leveling off at about $0.2 r_{vir}$. Indirect measurements of the DM velocity anisotropy fall on the upper end of the theoretically expected range. Though measured indirectly, the estimations are derived by using two different surrogate measurements - X-ray and galaxy dynamics. Current estimates of the DM velocity anisotropy are based on very small cluster sample. Increasing this sample will allow testing theoretical predictions, including the speculation that the decay of DM particles results in a large velocity boost. We also find, in accord with previous works, that halos are triaxial and likely to be more prolate when unrelaxed, whereas relaxed halos are more likely to be oblate. Our analysis does not indicate that there is significant correlation (found in some previous studies) between the radial density slope, γ, and β at large radii, $0.3 r_{vir} < r < r_{vir}$.

astro-ph.CO↗

arXiv2012

Flux Correlations in Supersonic Isothermal Turbulence

Using data from a large-scale three-dimensional simulation of supersonic isothermal turbulence, we have tested the validity of an exact flux relation derived analytically from the Navier--Stokes equation by Falkovich, Fouxon and Oz [2010 New relations for correlation functions in Navier--Stokes turbulence. J. Fluid Mech. 644, 465]. That relation, for compressible barotropic fluids, was derived assuming turbulence generated by a large-scale force. However, compressible turbulence in simulations is usually initialized and maintained by a large-scale acceleration, as in gravity-driven astrophysical flows. We present a new approximate flux relation for isothermal turbulence driven by a large-scale acceleration, and find it in reasonable agreement with the simulation results.

physics.flu-dyn↗

arXiv2011

On the Density Distribution in Star-forming Interstellar Clouds

We use deep adaptive mesh refinement simulations of isothermal self-gravitating supersonic turbulence to study the imprints of gravity on the mass density distribution in molecular clouds. The simulations show that the density distribution in self-gravitating clouds develops an extended power-law tail at high densities on top of the usual lognormal. We associate the origin of the tail with self-similar collapse solutions and predict the power index values in the range from -7/4 to -3/2 that agree with both simulations and observations of star-forming molecular clouds.

astro-ph.GA↗

arXiv2010

Quantifying the collisionless nature of dark matter and galaxies in A1689

We use extensive measurements of the cluster A1689 to assess the expected similarity in the dynamics of galaxies and dark matter (DM) in their motion as collisionless `particles' in the cluster gravitational potential. To do so we derive the radial profile of the specific kinetic energy of the cluster galaxies from the Jeans equation and observational data. Assuming that the specific kinetic energies of galaxies and DM are roughly equal, we obtain the mean value of the DM velocity anisotropy parameter, and the DM density profile. Since this deduced profile has a scale radius that is higher than inferred from lensing observations, we tested the validity of the assumption by repeating the analysis using results of simulations for the profile of the DM velocity anisotropy. Results of both analyses indicate a significant difference between the kinematics of galaxies and DM within $r \lesssim 0.3r_{\rm vir}$. This finding is reflected also in the shape of the galaxy number density profile, which flattens markedly with respect to the steadily rising DM profile at small radii. Thus, $r \sim 0.3r_{\rm vir}$ seems to be a transition region interior to which collisional effects significantly modify the dynamical properties of the galaxy population with respect to those of DM in A1689

astro-ph.CO↗

arXiv2007

The Santa Fe Light Cone Simulation Project: I. Confusion and the WHIM in Upcoming Sunyaev-Zel'dovich Effect Surveys

We present the first results from a new generation of simulated large sky coverage (~100 square degrees) Sunyaev-Zeldovich effect (SZE) cluster surveys using the cosmological adaptive mesh refinement N-body/hydro code Enzo. We have simulated a very large (512^3h^{-3}Mpc^3) volume with unprecedented dynamic range. We have generated simulated light cones to match the resolution and sensitivity of current and future SZE instruments. Unlike many previous studies of this type, our simulation includes unbound gas, where an appreciable fraction of the baryons in the universe reside. We have found that cluster line-of-sight overlap may be a significant issue in upcoming single-dish SZE surveys. Smaller beam surveys (~1 arcmin) have more than one massive cluster within a beam diameter 5-10% of the time, and a larger beam experiment like Planck has multiple clusters per beam 60% of the time. We explore the contribution of unresolved halos and unbound gas to the SZE signature at the maximum decrement. We find that there is a contribution from gas outside clusters of ~16% per object on average for upcoming surveys. This adds both bias and scatter to the deduced value of the integrated SZE, adding difficulty in accurately calibrating a cluster Y-M relationship. Finally, we find that in images where objects with M > 5x10^{13} M_{\odot} have had their SZE signatures removed, roughly a third of the total SZE flux still remains. This gas exists at least partially in the Warm Hot Intergalactic Medium (WHIM), and will possibly be detectable with the upcoming generation of SZE surveys.

astro-ph↗

arXiv2007

The Statistics of Supersonic Isothermal Turbulence

We present results of large-scale three-dimensional simulations of supersonic Euler turbulence with the piecewise parabolic method and multiple grid resolutions up to 2048^3 points. Our numerical experiments describe non-magnetized driven turbulent flows with an isothermal equation of state and an rms Mach number of 6. We discuss numerical resolution issues and demonstrate convergence, in a statistical sense, of the inertial range dynamics in simulations on grids larger than 512^3 points. The simulations allowed us to measure the absolute velocity scaling exponents for the first time. The inertial range velocity scaling in this strongly compressible regime deviates substantially from the incompressible Kolmogorov laws. The slope of the velocity power spectrum, for instance, is -1.95 compared to -5/3 in the incompressible case. The exponent of the third-order velocity structure function is 1.28, while in incompressible turbulence it is known to be unity. We propose a natural extension of Kolmogorov's phenomenology that takes into account compressibility by mixing the velocity and density statistics and preserves the Kolmogorov scaling of the power spectrum and structure functions of the density-weighted velocity v=ρ^{1/3}u. The low-order statistics of v appear to be invariant with respect to changes in the Mach number. For instance, at Mach 6 the slope of the power spectrum of v is -1.69, and the exponent of the third-order structure function of v is unity. We also directly measure the mass dimension of the "fractal" density distribution in the inertial subrange, D_m = 2.4, which is similar to the observed fractal dimension of molecular clouds and agrees well with the cascade phenomenology.

astro-ph↗

arXiv2007

Scaling Laws and Intermittency in Highly Compressible Turbulence

We use large-scale three-dimensional simulations of supersonic Euler turbulence to study the physics of a highly compressible cascade. Our numerical experiments describe non-magnetized driven turbulent flows with an isothermal equation of state and an rms Mach number of 6. We find that the inertial range velocity scaling deviates strongly from the incompressible Kolmogorov laws. We propose an extension of Kolmogorov's K41 phenomenology that takes into account compressibility by mixing the velocity and density statistics and preserves the K41 scaling of the density-weighted velocity v=rho^{1/3}u. We show that low-order statistics of 'v' are invariant with respect to changes in the Mach number. For instance, at Mach 6 the slope of the power spectrum of 'v' is -1.69 and the third-order structure function of 'v' scales linearly with separation. We directly measure the mass dimension of the "fractal" density distribution in the inertial subrange, D_m=2.4, which is similar to the observed fractal dimension of molecular clouds and agrees well with the cascade phenomenology.

astro-ph↗

arXiv2007

Simulating Cosmological Evolution with Enzo

In this paper we describe our massively parallel version of Enzo, a multiphysics, parallel, AMR application for simulating cosmological structure formation developed at UCSD and Columbia. We describe its physics, numerical algorithms, implementation, and performance on current terascale platforms. We also discuss our future plans and some of the challenges we face as we move to the petascale.

astro-ph↗

arXiv2006

AMR Simulations of the Cosmological Light Cone: SZE Surveys of the Synthetic Universe

We present preliminary results from simulated large sky coverage (~100 square degrees) Sunyaev-Zeldovich effect (SZE) cluster surveys using the cosmological adaptive mesh refinement N-body/hydro code Enzo. We have generated simulated light cones to match the resolution and sensitivity of current and future SZE instruments. These simulations are the most advanced calculations of their kind. The simulated sky surveys allow a direct comparison of large N-body/hydro cosmological simulations to current and pending sky surveys. Our synthetic surveys provide an indispensable guide for observers in the interpretation of large area sky surveys, and will develop the tools necessary to discriminate between models for cluster baryonic physics, and to accurately determine cosmological parameters.

astro-ph↗

arXiv2006

High Resolution Simulations of Supersonic Turbulence in Molecular Clouds

We present the results of three-dimensional simulations of supersonic Euler turbulence with grid resolutions up to 1024^3 points. Our numerical experiments describe nonmagnetized driven turbulent flows with an isothermal equation of state and an rms Mach number of 6. We demonstrate that the inertial range scaling properties of turbulence in this strongly compressible regime deviate substantially from a Kolmogorov-like behavior previously recovered for mildly compressible transonic flows.

astro-ph↗