HP ProLiant DL380p High-performance computing with accelerated HP ProLiant ser - Page 2

Introduction An increasing number of research institutions and businesses are using high-performance computing (HPC) technology for faster innovation and competitive advantage. HP brings HPC within the reach of more organizations by offering complete supercomputer clusters of purpose-built ProLiant servers with x86 processors. These supercomputer clusters use high-performance networking and robust cluster management tools to deliver reliable HPC performance affordably and efficiently. Our approach differs from competitors that require long-term commitments to costly, proprietary supercomputers. We're also raising performance and driving down costs in HPC visualization and computation environments by building industry-standard ProLiant servers that accommodate high-end graphic cards and accelerators. We jointly engineer and test our systems with products from industry-leading accelerator vendors and independent software vendors (ISVs) for a diverse range of HPC environments, including oil and gas, bioscience, financial services, government, automotive, and aerospace. This paper should be of particular interest to organizations with compute-intensive workloads that can run in an HPC environment and organizations that need to expand their use of HPC. It begins with a brief description of accelerators and graphics cards if you want to begin using HPC products for computation and visualization. Then it describes our efforts to research, qualify, and benchmark leading accelerators and graphics cards in ProLiant servers to make it easier for you to determine which is best for specific application workloads. Primarily, this paper describes the key design features we build into ProLiant servers to enable HPC products and our collaboration efforts with a growing ecosystem of partners to help ensure these products work well in your HPC environment. Accelerators and graphics cards Graphics cards render 2D or 3D images and video to direct-attached displays. Some organizations use graphics cards in clustered workstations or servers to send graphical information to remote systems over a network interconnect. In contrast, accelerators execute general-purpose calculations, as well as image images and video. They may be supported by general math libraries, or have applications specifically written to take advantage of them. Accelerators significantly increase application performance and reduce the power and space that servers need to run them. Accelerators A large part of the HPC industry is transitioning from using proprietary supercomputers to using supercomputer clusters of x86 servers with integrated accelerators from vendors such as NVIDIA®, AMD™-ATI, Nallatech™, and others. The most common accelerators are general-purpose graphic processing units (GPGPUs) and field programmable gate arrays (FPGAs). GPGPUs have wide use as powerful coprocessors for floating point computations. FPGAs target more specialized HPC users in areas such as digital signal processing and medical imaging. GPGPU accelerators GPGPU accelerators have many cores that replicate integer and floating-point Arithmetic Logic Units to provide massive parallelism. GPGPUs also have onboard, high-bandwidth memory for local data transfer. GPGPUs are available as expansion cards, integrated modules, or external racked subsystems- typically using PCI Express (PCIe) links to a server. They can boost execution from 2x to 10x (or more) in roughly the same form factor as servers without GPGPUs. 2