Accelerator Technologies (AT) for High Performance Computing (HPC) (Does Asia Lead the Way?)

随着计算进入PetaFlop时代，并超越ExaFlop级别，设备成本、功耗、可靠性、可编程性等问题，变得越来越重要。在过去的几年里，越来越多的系统依赖于加速器硬件技术（AT）。亚洲技术信息计划（ATIP）计划举办一次研讨会，探讨亚洲科学家将AT技术纳入其研究的方式。演讲者将介绍他们的成功水平和他们遇到的障碍等。他们的进展将与美国进行比较，ATIP将生成一个简短的信息报告，总结在全球HPC社区中分发的调查结果。AT可以非常具有成本效益。最流行的AT是基于游戏社区中使用的技术，因此年轻的研究人员，甚至一些高中生都很熟悉。因此，它们有可能帮助将高性能计算广泛迁移到较小的研究环境中，包括小型大学和其他地方。绝对成本以及成本性能是研究人员决定其研究的实际方向的关键。在亚洲的大部分地区尤其如此，那里的基本硬件必须从国外（美国）购买，因此亚洲人很快就考虑了AT。该研讨会汇集了来自美国和亚洲的顶尖科学家和研究生，希望他们能够建立联系，从而可能导致未来的研究合作或实际HPC应用的合作商业化努力。

As computing moves into the PetaFlop era, and beyond to ExaFlop levels, issues of equipment cost, power consumption, reliability, programmability, etc., are becoming increasingly important. Over the past few years more and more systems are depending upon accelerator hardware technologies (AT). The Asian Technology Information Program (ATIP) plans to hold a workshop that will explore the ways Asian scientists are incorporating AT technologies into their research. Speakers will present on their level of success and the impediments they have encountered, etc. Their progress will be compared with that in the US, and ATIP will generate a short informative report summarizing the findings for distribution throughout the global HPC community.ATs can be very cost effective. The most popular ATs are based on technology used in the game community, and hence are familiar to younger researchers, and even to some students in high school. Therefore, they have the potential to help migrate high-performance computing broadly, into smaller research settings, including small universities and beyond. Absolute cost as well as cost performance is crucial in researchers decisions about the practical directions of their research. This is especially true in most parts of Asia, where essential hardware must be purchased from abroad (US), and Asians have been quick to consider ATs as a result. This workshop brings together top scientists and graduate students from both the US and Asia hoping that they will form connections that may lead to future research collaborations or cooperative commercialization efforts for practical HPC applications.