Search NASA⌕ Search

Engineering topics

Li, Xinyi

Publications and source records attributed to Li, Xinyi.

FTTN: Feature-Targeted Testing for Numerical Properties of NVIDIA & AMD Matrix Accelerators

While NVIDIA has been the dominant provider of GPUs for HPC and ML, now AMD has several offerings of GPUs. This encourages programmers to try out AMD GPUs for new codes and also port existing codes over. Unfortunately, without understanding the floating-point differences between these GPU types, software development or porting can introduce bugs—and currently such an understanding is lacking. The magnitude of this open question becomes clear if one imagines the the number of floating-point precision choices (FP16, FP32, etc.), floating-point formats (standard floats, brain-float, etc.), and execution units available (elementary units, matrix/tensor cores, etc.) Questions such as rounding modes and subnormal support are also important. Most of these answers are unknown today or are hard to access. We provide the first testing-guided approach that answers a significant number of these questions. We also devise tests to reveal internal information (e.g., extra bits kept) to make sure that our findings are reliable. Many of our tests employ systematically generated random-programs, others apply fast-math flags and some involve fused multiplyadd. Especially for tensor/matrix cores, the tests have nontrivial logic that we present Our testing approach is reusable for the plethora of GPUs yet to be introduced. Our findings include up to 7 ulps of difference between NVIDIA and AMD for sin and cos at FP32 precision and 3 ulp at FP64. In our study of matrix cores (NVIDIA) and tensor cores (AMD), we have extensively characterized rounding modes (truncation versus round-to-nearest), the number of extra internal bits kept (whether 3 bits are kept or not), subnormal support for inputs and outputs across four different floating-point formats and across NVIDIA A100 and AMD MI250X GPUs. We believe that this wealth of data becoming available for the first time may help avoid significant porting bugs when migrating code across these platforms.

Li, Xinyi↗

Understanding Mixed Precision GEMM with MPGemmFI: Insights into Fault Resilience

Emerging deep learning workloads urgently need fast general matrix multiplication (GEMM). Thus, one of the critical features of machine-learning-specific accelerators such as NVIDIA Tensor Cores, AMD Matrix Cores, and Google TPUs is the support of mixed-precision enabled GEMM. For DNN models, lower-precision FP data formats and computation offer acceptable correctness but significant performance, area, and memory footprint improvement. While promising, the mixed-precision computation on error resilience remains unexplored. To this end, we develop a fault injection framework that systematically injects fault into the mixed-precision computation results. We investigate how the faults affect the accuracy of machine learning applications. Based on the characteristics of error resilience, we offer lightweight error detection and correction solutions that significantly improve the overall model accuracy by 75% if the models experience hardware faults. The solutions can be efficiently integrated into the accelerator's pipelines.

Fang, Bo↗

FTTN: Feature-Targeted Testing for Numerical Properties of NVIDIA & AMD Matrix Accelerators

FTTN is a test suite to evaluate the numerical behaviors of matrix accelerators of GPUs (NVIDIA Tensor Cores and AMD Matrix Cores) in a quick and simple setting. Matrix accelerators are heavily used in today's computationally intense applications to speed up matrix multiplications. This test suite provides a comprehensive study on the numerical behaviors of these accelerators, including support for subnormals, rounding modes, extra precision bits and FMA features. Is there

Laguna Peralta, Ignacio↗

A GPU accelerated mixed-precision Smoothed Particle Hydrodynamics framework with cell-based relative coordinates

Smoothed Particle Hydrodynamics (SPH) is essential for modeling complex large-deformation problems across various applications, requiring significant computational power. A major portion of SPH computation time is dedicated to the Nearest Neighboring Particle Search (NNPS) process. While advanced NNPS algorithms have been developed to enhance SPH efficiency, the potential efficiency gains from modern computation hardware remain underexplored. Here, this study investigates the impact of GPU parallel architecture, low-precision computing on GPUs, and GPU memory management on NNPS efficiency. Our approach employs a GPU-accelerated mixed-precision SPH framework, utilizing low precision float-point 16 (FP16) for NNPS while maintaining high precision for other components. To ensure FP16 accuracy in NNPS, we introduce a Relative Coordinated-based Link List (RCLL) algorithm, storing FP16 relative coordinates of particles within background cells. Our testing results show three significant speedup rounds for CPU-based NNPS algorithms. The first comes from parallel GPU computations, with up to a 1000x efficiency gain. The second is achieved through low-precision GPU computing, where the proposed FP16-based RCLL algorithm offers a 1.5x efficiency improvement over the FP64-based approach on GPUs. By optimizing GPU memory bandwidth utilization, the efficiency of the FP16 RCLL algorithm can be further boosted by 2.7x, as demonstrated in an example with 1 million particles. Our code is released at https://github.com/pnnl/lpNNPS4SPH.

97 MATHEMATICS AND COMPUTING↗

pnnl/lpNNPS4SPH

Low precision-based nearest neighboring particle searching algorithm.This work introduces a computationally efficient Nearest Neighbor Particles Searching (NNPS) algorithm tailored for the meshfree Smoothed Particle Hydrodynamics (SPH) method in large-deformation problems simulations. The innovation lies in the strategic use of low-precision float-point 16 (FP16) in NNPS for efficiency purpose.

Mao, Zirui↗

GPU-FPX

GPU-FPX is a tool based on NVBit that detects and analyzes floating-point exceptions on NVIDIA GPUs through binary instrumentation. Its purpose is to detect and report the occurrences of floating-point exceptions during numerical computations, offering efficient location detections and exception flow. GPUFPX achieves exceptional performance, being 16x faster than comparable prior tools, such as BinFPE.

Laguna Peralta, Ignacio↗

Crystal growth and nucleation at metal-organic framework / solution interfaces

Metal-organic frameworks (MOFs) represent an incredibly diverse class of materials with equally diverse applications in catalysis, separations, and gas storage. Despite the prevalence of MOF structures in the literature, in many cases the synthesis of pure, crystalline MOF materials remains a significant challenge, with failed syntheses often yielding mixed crystalline or non-porous amorphous products. Our recent work aims to enable targeted MOF synthesis by developing a comprehensive and predictive model of the fundamental processes occurring at the MOF-solution interface under synthetically-relevant conditions. Our recent accomplishments include: (i) development and implementation of a graphics processor-accelerated approach to simulation the nucleation of weak electrolytes, and the extension of these methods to large/complex solutes such as MOFs; and (ii) studies of the structure of ZIF interfaces under conditions relevant to crystal growth; and (iii) addressing important open questions in the mechanism of MOF solvent-assisted linker exchange, and important synthetic pathway for MOFs.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

The Circular Velocity Curve of the Milky Way from 5–25 kpc Using Luminous Red Giant Branch Stars

We present a sample of 254,882 luminous red giant branch (LRGB) stars selected from the APOGEE and LAMOST surveys. By combining photometric and astrometric information from the Two Micron All Sky Survey and Gaia survey, the precise distances of the sample stars are determined by a supervised machine-learning algorithm: the gradient-boosted decision trees. To test the accuracy of the derived distances, member stars of globular clusters (GCs) and open clusters are used. The tests by cluster member stars show a precision of about 10% with negligible zero-point offsets, for the derived distances of our sample stars. The final sample covers a large volume of the Galactic disk(s) and halo of 0 < R < 30 kpc and |Z| ≤ 15 kpc. The rotation curve (RC) of the Milky Way across the radius of 5 ≲ R ≲ 25 kpc has been accurately measured with ~54,000 stars of the thin disk population selected from the LRGB sample. The derived RC shows a weak decline along R with a gradient of -1.83 ± 0.02 (stat.) ± 0.07 (sys.) km s -1 kpc -1 , in excellent agreement with the results measured by previous studies. The circular velocity at the solar position, yielded by our RC is 234.04 ± 0.08 (stat.) ± 1.36 (sys.) km s -1 , again in great consistency with other independent determinations. From the newly constructed RC, as well as constraints from other data, we have constructed a mass model for our Galaxy, yielding a mass of the dark matter halo of M 200 = (8.05 ± 1.15) × 10 11 M ⊙ with a corresponding radius of R 200 = 192.37 ± 9.24 kpc and a local dark matter density of 0.39 ± 0.03 GeV cm -3 .

79 ASTRONOMY AND ASTROPHYSICS↗

Towards Precision-Aware Fault Tolerance Approaches for Mixed-Precision Applications

Graphics Processing Units (GPUs), the dominantly adopted accelerators in HPC systems, are susceptible to transient hardware fault. New generation of GPUs feature mixed-precision architectures such as NVIDIA Tensor Cores to accelerate matrix multiplications. While being widely adapted, how would they behave under transient hardware faults remain unclear. In this study, we conduct a large-scale fault injection experiments on GEMM kernels implemented with different floating-point data types on the V100 and A100 Tensor Cores, and show distinct error resilience characteristics for the GEMMS with different formats. In the future, we plan to explore this space by building precision-aware floating-point fault tolerance techniques for applications such as DNNs that exercise low-precision computations.

Fang, Bo↗

Modeling the Nucleation of Weak Electrolytes via Hybrid GCMC/MD Simulation

We have developed a hybrid grand canonical Monte Carlo/molecular dynamics (GCMC/MD) method for simulating the nucleation of weak electrolytes in an explicit solvent. In contrast to brute-force MD simulation, the approach is capable of efficiently simulating the nucleation of dilute solutions while including the atomistic influence of the surrounding solvent and provides access to the full nucleation free energy surface and associated nucleation free energy barrier. After validating the method against a simple model system, we applied the approach to the nucleation of a low-solubility rock-salt structure in liquid water. We find that the calculated nucleation barriers, in conjunction with analytic rate theories, yield predicted nucleation rates that are in excellent agreement with brute-force MD simulations of the supersaturated solution. We anticipate possible applications of this approach to a wide variety of related weak electrolytes, including CaCO 3 , zeolites, and metal–organic frameworks.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Tuning the Core–Shell Structure of Au 144 @Fe 2 O 3 for Optimal Catalytic Activity for CO Oxidation

Core–shell heterostructures have been utilized as a catalyst that is thermally stable and exhibits a synergistic effect between core and shell, resulting in increased catalytic activity. We report on the synthetic procedure involving a Au 144 core with an iron oxide shell which can be varied in thickness. The Au 144 @Fe 2 O 3 particles with Au:Fe mass ratios of 1:2, 1:4, and 1:6 were synthesized and then deposited onto silica via colloidal deposition. Using CO oxidation, each Au 144 @Fe 2 O 3 /SiO 2 catalyst gave varying degrees of full CO conversion depending on the thickness of the iron oxide layer. The 1:4 Au 144 @Fe 2 O 3 /SiO 2 catalyst produced the best catalytic activity and was further investigated via thermal treatments, where calcination at 300 °C presented the best results, and the 1:4 ratio was still active at 100 °C after thermal treatments.

2-propanol conversion↗

Characterization of Microstructure and Wear Resistance of PEO Coatings Containing Various Microparticles on Ti6Al4V Alloy

Titania-based composite coatings were prepared by plasma electrolytic oxidation (PEO) treatment of Ti6Al4V alloy in electrolyte with α-Al{sub 2}O{sub 3}, Cr{sub 2}O{sub 3} or h-BN microparticles in suspension. The microstructure, composition of PEO composite coatings were analyzed by SEM, EDS and XRD. The wear resistance of composite ceramic coatings was studied by ball-on-disk wear test at ambient temperature and 300 °C. The results showed that the addition of microparticles accelerated the growth rate of PEO coating and changed the microstructure and composition of PEO coating. PEO coating was porous and mainly composed of rutile-TiO{sub 2}, anatase-TiO{sub 2} and Al{sub 2}TiO{sub 5}. PEO/α-Al{sub 2}O{sub 3} (Cr{sub 2}O{sub 3} or h-BN) composite coating only had small micropores and appeared some α-Al{sub 2}O{sub 3} (Cr{sub 2}O{sub 3} or h-BN) phase. Besides, the addition of α-Al{sub 2}O{sub 3} (Cr{sub 2}O{sub 3} or h-BN) microparticles greatly improved the wear resistance of PEO coating. At ambient temperature, abrasive wear dominated the wear behavior of PEO coating, but abrasive wear and adhesive peel simultaneously happened at 300 °C. Whether at ambient temperature or 300 °C, PEO composite coating had better wear resistance than PEO coating. Besides, PEO/h-BN composite coating outperformed other composite coatings regardless of the temperature.

36 MATERIALS SCIENCE↗

In Situ, Time-Resolved, and Mechanistic Studies of Metal–Organic Framework Nucleation and Growth

The vast chemical and structural diversity of metal–organic frameworks (MOFs) opens up the exciting possibility of “crystal engineering” MOFs tailored for particular catalytic or separation applications. Yet the process of reaction discovery, optimization, and scale-up of MOF synthesis remains extremely challenging, presenting significant obstacles to the synthetic realization of many otherwise promising MOF structures. Recently, significant new insights into the fundamental processes governing MOF nucleation and growth, as well as the relationship between reaction parameters and synthetic outcome, have been derived using powerful in situ, time-resolved and/or mechanistic studies of MOF crystallization. This Review provides a summary and associated critical analysis of the results of these and other related “direct” studies of MOF nucleation and growth, with a particular emphasis on the recent advances in instrument technologies that have enabled such studies and on the major hypotheses, theories, and models that have been used to explain MOF formation. We conclude with a summary of the major insights that have been gained from the work summarized in this Review, outlining our own perspective on potential fruitful new directions for investigation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

FGFR antagonist induces protective autophagy in FGFR1-amplified breast cancer cell

Breast cancer, representing approximately 30% of all gynecological cancer cases diagnosed yearly, is a leading cause of cancer-related mortality for women. Amplification of FGFR1 is frequently observed in breast cancers and is associated with poor prognosis. Though FGFRs have long been considered as anti-cancer drug targets, and a cluster of FGFR antagonists are currently under clinical trials, the precise cellular responses under the treatment of FGFR antagonists remains unclear. Here, we show that PD166866, an FGFR1-selective inhibitor, inhibits proliferation and triggers anoikis in FGFR1-amplified breast cancer cell lines. Notably, we demonstrate that PD166866 induces autophagy in FGFR1-amplified breast cancer cell lines, while blockage of autophagy by Atg5 knockdown further enhances the anti-proliferative activities of PD166866. Moreover, mechanistic study reveals that PD166866 induces autophagy through repressing Akt/mTOR signaling pathway. Together, the present study provides new insights into the molecular mechanisms underlying the anti-tumor activities of FGFR antagonists, and may further assist the FGFRs-based drug discovery. -- Highlights: •FGFR1 antagonist inhibits cell viability in FGFR1-amplified breast cancer cells. •FGFR1 antagonist induces autophagy in FGFR1-amplified breast cancer cells. •FGFR1 antagonist-induced autophagy is protective. •FGFR1 antagonist induces autophagy by inhibiting Akt/mTOR pathway.

60 APPLIED LIFE SCIENCES↗