Silicon Graphics Prism Extreme 128P/1TB

描述（由申请人提供）：本提案要求资源为一组使用成像研究各种疾病的研究人员创建高性能计算资源。磁共振成像（MR!）已经成为一种广泛使用的工具，用于在越来越高的细节水平上研究大脑功能、生理学和解剖学。成像数据的分析现在提出了一套新的极具挑战性的问题，如果适当地处理所需的资源，可以大大改善研究人员的资助研究。问题的具体情况因研究者而异，但有一个共同的特点：需要在内存中有一个1tb大小的单一、大型、扁平的地址空间，可由多个cpu寻址。许多研究人员将特别受益于一种设备的能力，该设备可以计算由新的非常高通道计数的MRI系统产生的数据，如32和96通道的MRI扫描仪，在1.5特斯拉和3特斯拉甚至更高。因此，我们提出了一种基于高速缓存相干非均匀存储器访问（ccNUMA）的高性能计算机，该计算机具有128个Itanium 2处理器和1tb RAM。这些研究人员目前可用的最接近的设备只有可用RAM的1/64，因此该设备代表了一个独特的机会，可以克服特定的计算限制，并显着加快许多研究人员资助的生物医学研究。

DESCRIPTION (provided by applicant): This proposal requests resources to create a high performance computing resource for a group of investigators using imaging to study a variety of diseases. Magnetic resonance imaging (MR!) has become a heavily used tool for studying brain function, physiology, and anatomy at increasingly high levels of detail. Analyses of imaging data now present a new set of extremely challenging problems that, if appropriately addressed with the requested resource, could substantially improve the funded research of the investigators. The specifics of the problems vary by investigator but share a common feature: the need for a single, large, flat address space in memory of 1 terabyte in size, addressable by multiple CPUs. Many investigators would benefit specifically from the ability of a device to compute the data being generated by new very-high- channel-count MRI systems such as 32 and 96 channel MRI scanners at 1.5 Tesla and 3 Tesla and beyond. Therefore, we propose acquiring a cache coherent non uniform memory access (ccNUMA) based high performance computer with 128 Itanium 2 processors and 1 TB of RAM. The closest device currently available to these investigators has 1/64th the amount of RAM available, and therefore this device represents a unique opportunity to overcome the specific computational constraints and significantly speed many investigators funded biomedical research.