GPU-Accelerated Parallel Computer for Life Sciences Research

项目总结/摘要 我们要求资金购买一台高性能的GPU加速并行计算机， 为范德比尔特的计算密集型NIH资助的生物医学研究提供财政上可持续的支持 大学（VU）和范德比尔特大学医学中心（VUMC）。拟议的系统由20个Exxact组成 “TensorEX”计算节点，每个节点配备四个nVidia A6000“Ampere”GPU，512 GB RAM，双32- 核心AMD EPYC CPU、用于临时I/O的8 TB nVME SSD以及双端口Mellanox 25 Gbps网络接口。 GPU在高性能技术计算中得到了广泛的认可和应用 （HPTC）和数据科学领域。其简化但高度并行的架构产生的处理器可以 同时进行数量惊人的基本算术运算， 因此与传统的基于CPU的架构中的类似能力相比成本更低。内在的~100- 倍的价格/性能优势促使许多重要的HPTC代码和库的作者修改 他们的软件在GPU上运行。现在有成熟的GPU实现许多关键的计算- 密集的软件工具，这是关键，以加快发现的步伐，在我们的研究计划。 拟议的计算机系统将支持研究软件，这是众所周知的最佳执行时， 运行在GPU协处理器上。分子动力学（MD）代码，如AMBER和GROMACS， 我们的目标是满足需求，以支持我们基于结构的药物设计（SBDD）和个性化结构生物学（PSB）的工作。 低温电子显微镜（Cryo-EM）和低温电子断层扫描（Cryo-ET）图像分析工具 像RELION这样的公司，利用我们机构最近进行的并将继续进行的大量投资， 这些地区机器学习应用于基于配体的药物发现和高维统计 用于系统生物学的建模工具是GPU加速工具的示例，这些工具既被开发又被应用 在范德比尔特。这些领域（以及更多领域）的研究将得到显著加强、加速， 成本效益与获得拟议的文书。这将转化为直接的积极影响， 小分子和抗体为基础的疗法的发展，个性化医疗的进展， 我们在原子到细胞尺度上对疾病相关机制的理解取得了重大进展。 拟议的系统将在高级计算研究中心维护， Education（ACCRE），拥有大量24/7随叫随到的员工，他们在管理CPU集群方面拥有丰富的经验 以及支持生物医学研究的基于GPU的计算机，包括以前S10奖项的设备。 这种教师驱动的计算方法非常成功，使ACCRE成为一个关键组成部分 VU/VUMC研究企业的领导人。ACCRE的技术支持由机构基金提供补贴， 减少费用和障碍，使主要和次要用户能够利用这一基础设施。

Project Summary/Abstract We request funds to purchase a high-performance, GPU-accelerated parallel computer to provide financially sustainable support for computationally intensive NIH-funded biomedical research at Vanderbilt University (VU) and Vanderbilt University Medical Center (VUMC). The proposed system consists of 20 Exxact “TensorEX” compute nodes, each equipped with four nVidia A6000 “Ampere” GPUs, 512GB of RAM, dual 32- core AMD EPYC CPUs, an 8TB nVME SSD for scratch I/O, and a dual-port Mellanox 25Gbps network interface. GPUs have achieved wide acceptance and application in the high-performance technical computing (HPTC) and data science fields. Their simplified but highly parallel architectures yield processors that can do a staggering number of fundamental arithmetic operations simultaneously, at a tiny fraction of the complexity and therefore cost compared to a similar capability in a traditional CPU-based architecture. The intrinsic ~100- fold price/performance advantage has driven the authors of many important HPTC codes and libraries to modify their software to run on GPUs. There are now mature GPU implementations of many critical computationally- intensive software tools which are key to accelerating the pace of discovery in our research programs. The proposed computer system will support research software that is well-known to perform best when running on GPU coprocessors. Molecular dynamics (MD) codes such as AMBER and GROMACS are in high demand, to support our structure-based drug design (SBDD) and personalized structural biology (PSB) efforts. Cryogenic electron microscopy (Cryo-EM) and cryogenic electron tomography (Cryo-ET) image analysis tools such as RELION leverage the massive investments our institution has recently made and continues to make in these areas. Machine learning applications for ligand-based drug discovery, and high-dimensional statistical modeling tools for systems biology are examples of GPU-accelerated tools that are both developed and applied at Vanderbilt. Research in these areas (and more) would be significantly enhanced, accelerated, and more cost-effective with access to the proposed instrument. That will translate into direct positive impacts on the development of small molecule and antibody-based therapeutics, advances in personalized medicine, and critical advancements in our understanding of disease-related mechanisms at atomic through cellular scales. The proposed system will be maintained at the Advanced Computing Center for Research and Education (ACCRE), which has a large 24/7 on-call staff with extensive experience managing clusters of CPU and GPU-based computers in support of biomedical research, including equipment from previous S10 awards. This faculty-driven approach to computing has been extremely successful, making ACCRE a critical component of the VU/VUMC research enterprise. Technical support at ACCRE is subsidized by institutional funds, further reducing costs and barriers for access to this infrastructure by the major and minor users on this proposal.