MRI: Acquisition of a GPU-Accelerated Cluster, High Performance Rio Grande Valley Cluster (HiRGV)

收购德克萨斯大学里奥格兰德山谷(UTRGV)的高性能里奥格兰德山谷(HiRGV)计算集群将支持新的和正在进行的多学科研究，并使用不同的计算技术提供教师和学生培训。HiRGV将混合使用多种处理器，如多核处理器和图形处理器(GPU)。虽然GPU最初是为了满足视频游戏中动态可视化的极端计算需求而开发的，但现在它们是许多应用程序的驱动力，例如人工智能，这些应用程序需要大规模并行化来实现高性能。HiRGV支持的项目将在多个与国家利益高度相关的领域推进科学知识：超大规模集成芯片的设计和开发将变得更快；引力波搜索将发现更弱的信号，美国目前在引力波搜索中处于领先地位；燃料电池效率将被利用新材料提高；癌症治疗将被推进；将为社区提供新型加密工具和服务；将开发深度强化学习算法来解决具有挑战性的决策问题；将允许养蜂人通过人工智能系统监测蜂箱健康；将发现2型糖尿病的分子机制。所有上述工作将在该国第二大拉美裔服务机构进行。来自主要代表不足的群体的本科生和研究生将接受高水平计算的实践培训，并获得21世纪劳动力所需的技能。1)大规模稀疏矩阵解的并行计算；最小化填充和左上角非零项将提高稀疏矩阵解的求解阶段的执行时间；2)快速全相干全天(FCAS)搜索引力波(GW)信号；从当前的事件向永远在线的FCAS搜索引力波(GW)将导致探测灵敏度和参数估计精度的显著提高；3)与CO吸附唯象模型关联的催化层原子与分子量子理论(QTAIM)将解决目前小分子吸附和燃料电池技术的不足；4)通过分子动力学模拟研究GRP119受体配体识别和激活的分子机制；阐明GPR119和其他脂类结合受体的配体识别和激活将用于治疗2型糖尿病和其他疾病；5)利用高性能GPU集群进行分子靶标识别和癌症药物发现；发现的生物标志物和靶蛋白可用于针对靶标癌症的药物设计；6)将计算机视觉技术(CVT)应用于蜜蜂健康和监测；CVT记录将提供如何分析蜜蜂的活动、害虫的存在或蜂箱中花粉的数量和类型；7)开发一种新的密码工具，使用GPU以较低的计算成本提供各种理想的安全功能；8)开发深度强化学习算法，提供反应性的类似人类的决策、高级审议和解释能力以及任务之间的高效知识转移。除了上述项目，HiRGV还将在代码开发、测试和性能评估方面支持六个项目，使它们能够在德克萨斯高级计算中心(TACC)的资源上运行。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The acquisition of the High-Performance Rio Grande Valley (HiRGV) computing cluster at the University of Texas Rio Grande Valley (UTRGV) will support novel and ongoing multidisciplinary research and provide faculty and student training using heterogeneous computing technologies. HiRGV will have a mix of processors, such as multi-core and Graphics Processing Units (GPUs). While GPUs were initially developed to satisfy the extreme computing needs of dynamic visualization in video games, they are now the driving force in many applications, such as artificial intelligence, that need massive parallelization to achieve high performance. The projects supported by HiRGV will advance scientific knowledge in multiple areas that are all highly relevant to national interests: Design and development of Very Large Scale Integrated chips will become faster; Weaker signals will be found in gravitational wave searches, where the U.S. is the leading nation at present; Fuel cell efficiency will be improved using novel materials; Cancer therapy will be advanced; A novel cryptography tool and service will be provided to the community; Deep reinforcement learning algorithms will be developed to resolve challenging decision making problems; Beekeepers will be allowed to monitor beehive health by AI systems; Molecular mechanism will be discovered for type 2 diabetes. All of the above work will be carried out at the second largest Hispanic-Serving Institution in the country. Undergraduate and graduate students from a predominantly under-represented group will receive hands-on training in high level computing and acquire skills that are in high demand for the 21st century workforce.The specific goals of the projects supported by HiRGV are as follows. 1) Parallel computation for large-scale sparse matrix solution; Minimized fill-ins and non-zero entries to the top-left directions will improve execution time for SOLVE phase of sparse matrix solution; 2) Fast fully-coherent all-sky (FCAS) search for gravitational wave (GW) signals; Transitioning from the current episodic to an always-on FCAS search for GWs from binary inspirals will result in a major improvement in detection sensitivity and parameter estimation accuracy; 3) Quantum Theory of Atoms and Molecules (QTAIM) on catalytic layers correlated with CO adsorption phenomenological models will solve the shortcoming of current approaches in small molecule adsorption and fuel cell technology; 4) Studying the Molecular Mechanism of GRP119 Receptor Ligand Recognition and Activation Through Molecular Dynamics Simulations; Elucidated ligand recognition and activation of the GPR119 and other lipid binding receptors of the same family will be used for treating type 2 diabetes and other diseases; 5) Molecular target identification and drug discovery for cancer using high performance GPU cluster; Discovered biomarkers and target proteins can be applied to a drug design given a target cancer service; and 6) Applying computer vision technologies (CVT) to honey bee health and surveillance; CVT recording will provide how to analyze honeybee movements, presence of pests, or the amount and type of pollen brought into a hive; 7) Developing a novel cryptography tool that offers various desirable security features with low computational cost using GPUs; 8) Developing deep reinforcement learning algorithms that provide reactive human-like decision-making, high-level deliberative and explanatory capabilities, and efficient transfer of knowledge between tasks. In addition to the above projects, the HiRGV will support six projects in terms of code development, testing, and performance assessment, enabling them to run on Texas Advanced Computing Center (TACC) resources.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.