Equipment: MRI: Track 1 Acquisition of a GPU-Accelerated Computing Cluster for Advanced Optimization and Design in Multidisciplinary Research and Education

该奖项支持收购具有最先进的图形处理单元（GPU）的专用高性能计算（HPC）集群，这些GPU与现有资源相比具有独特的功能。该仪器将使普林斯顿社区内的数百名研究人员，外部合作者和项目参与者受益，加速科学发现，对可持续发展和人类健康产生有益影响。该仪器将催化化学、化学工程、计算机科学、地球物理学、机械工程、神经科学、等离子体物理学和心理学等多个学科的前沿研究，以解开复杂现象并开创应对全球挑战的变革性解决方案。特别是，该系统支持医疗保健和可持续能源解决方案的新材料设计研究，加深我们对复杂化学和物理学的理解，开发用于医学和传感的低成本成像系统，大规模计算和与神经科学相关的分析等等。此外，该工具将通过软件传播和广泛的培训计划促进高性能GPU的教育和利用。主要举措包括与NVIDIA合作举办的年度GPU黑客活动、针对未被充分代表的本科生和以前被监禁的青年的夏季研究计划、参与由学生领导的初创企业开发和启动加速器计划以及GPU相关研讨会。这些活动将为工业，国家实验室和学术界的技术职位培养非常强大和准备充分的候选人。该仪器将配备10个计算节点，每个节点都配备AMD Genoa CPU和NVIDIA H100 GPU。重要的是，H100提供了增强的异步执行和比当前A100大幅度的加速（例如，6倍于Tensor Cores，7倍于DPX指令，2- 8倍于AI模型训练，30倍于AI推理）。有了这些GPU，普林斯顿大学的研究人员和合作者将利用机器学习和高速计算在各自的领域取得显着进展。通过提高分子模拟的吞吐量，该机器将大大加快新型催化材料（如耐用酶和聚合物-MOF复合材料）的计算引导设计的工作流程。GPU加速将产生开创性的基于从头计算的分子和界面现象的描述。这台机器还将使关键的软件开发能够有效地绘制地球内部的地图，推进医疗磁共振成像的廉价替代品，用于聚变反应堆控制的实时缓解策略，以及模拟复杂的燃烧系统。计算能力的增强将支持新型纳米光子成像器的设计，从而开启医学或光学诊断领域的新应用。该机器将进一步支持高通量定量表征，为动物模型中社会沟通和奖励寻求的神经基础提供信息。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award supports the acquisition of a specialized high-performance computing (HPC) cluster with state-of-the-art graphical processing units (GPUs) that offer unique capabilities compared to existing resources. The instrument will benefit hundreds of researchers within the Princeton community, external collaborators, and program participants, accelerating scientific discoveries with beneficial impacts on sustainability and human health. The instrument will catalyze frontier research across diverse disciplines, such as chemistry, chemical engineering, computer science, geophysics, mechanical engineering, neuroscience, plasma physics, and psychology, to unravel complex phenomena and pioneer transformative solutions to global challenges. In particular, the system supports research in the design of new materials for healthcare and sustainable-energy solutions, a deepening of our understanding of complex chemistry and physics, development of low-cost imaging systems for medicine and sensing, large-scale computations and analyses related to neuroscience, and more. Additionally, the instrument will foster education and utilization of high-performance GPUs through software dissemination and broadly accessible training programs. Major initiatives include an annual GPU Hackathon with NVIDIA, summer research programs for underrepresented undergraduate students and formerly incarcerated youth, engagement with an accelerator program for development and launching of startups led by students, and GPU-related workshops. These activities will cultivate exceptionally strong and well-prepared candidates for technical positions in industry, national labs, and academia. The instrument will feature 10 computing nodes, each with AMD Genoa CPUs and NVIDIA H100 GPUs. Importantly, the H100s offer enhanced asynchronous execution and substantial speedups over current A100s (e.g., 6x on Tensor Cores, 7x on DPX instructions, 2-8x on AI model training, and 30x on AI inference). With these GPUs, Princeton researchers and collaborators will leverage machine learning and high-speed calculations to achieve remarkable progress in their respective fields. By increasing throughput of molecular simulations, the machine will dramatically accelerate work flows for computationally guided design of novel catalytic materials, such as durable enzymes and polymer-MOF composites. GPU acceleration will yield pioneering ab initio-based descriptions of molecular and interfacial phenomena. The machine will also enable critical software developments towards efficiently mapping the earth’s interior, advancing inexpensive alternatives to medical magnetic resonance imaging, real-time mitigation strategies for fusion reactor control, and modeling complex combustion systems. Increased computing capabilities will support the design of novel nanophotonic imagers that unlock new applications in medicine or as optical diagnostics. The machine will further bolster high-throughput quantitative characterization to inform the neurological underpinnings of social communication and reward-seeking in animal models.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.