GPU Computing Enabled Linux Computer Cluster

描述（由申请人提供）：电子冷冻显微镜（cryoEM），特别是单粒子冷冻显微镜，在可实现的分辨率方面取得了巨大的成功。现在有可能确定具有高对称性的大型蛋白质组合的近原子结构，例如二十面体病毒。然而，对于对称性较低的大型蛋白质复合体，实现类似的分辨率仍然非常困难。这部分是因为收集非常大的图像数据集的过程非常繁琐，以及处理如此大的数据集对计算能力的巨大需求。我们专门设置了一个独特的易于操作的电子显微镜系统来促进这一过程（自动采集软件和一个非常大的格式，8K x 8K像素的数码相机）。在这个应用程序中，我们寻求资金来建立一个支持GPU计算的计算机集群，用于快速数据处理。与电子显微镜系统一起，我们的目标是建立一个集成的冷冻电镜系统，以克服目前冷冻电镜在研究高对称性的大蛋白质复合物方面的分辨率限制。我们的目标是常规收集和处理105 - 106个粒子图像。提议的GPU计算机集群将是一个中等规模的GPU- cpu单元集群，每个服务器类头节点连接到包含四个特斯拉GPU处理器的Nvidia S1070 GPU盒。它将专门用于冷冻电镜数据集的快速处理，特别是大约100万个粒子图像的非常大的数据集。建立这样的GPU计算机集群是实现单粒子低温电镜常规测定大分子复合物近原子分辨率结构的关键一步。许多由NIH资助的生物医学研究项目，特别是那些需要确定大型复合物结构的项目；将受益于这样的技术进步。所有开发的软件将免费提供。

DESCRIPTION (provided by applicant): Electron cryomicroscopy (cryoEM), particularly single particle cryoEM, has experienced tremendous successes in term of achievable resolution. It is now possible to determine near atomic structures of large protein assemblies with high symmetry, such as icosahedral virus. However, for large protein complex with low symmetry, achieving the similar resolution is still very difficult. This is partly because of the very tedious process of collecting very large datasets of images, and the tremendous demands of computational power to process such large dataset. We have set up a unique easy-to-operate electron microscope system specifically to facilitate this process (automated acquisition software and a very large format, 8K x 8K pixel digital camera). In this application, we seek funds to set up a GPU computing enabled computer cluster for fast data processing. Together with the electron microscope system, we aim to establish an integrated cryoEM system that will overcome the current resolution limitation of cryoEM in studying large protein complexes without high symmetry. Our goal is to routinely collect and process 105 - 106 particle images. The proposed GPU computer cluster will be a medium sized cluster of GPU-CPU units with each server class headnode connected with the just available Nvidia S1070 GPU box containing four Tesla GPU processors. It will be dedicated for fast processing of cryoEM datasets, particularly very large datasets of ~ 1 million particle images. Establish such a GPU computer cluster is a critical step towards routinely determine near atomic resolution structure of macromolecular complex by single particle cryoEM. Many biomedical research projects funded by NIH, particularly those require structural determination of large complexes; will be benefited from such technological advancement. All developed software will be freely provided.