Near-future Supercomputer Directions
Some idea of what might be happening in the near future in supercomputer design can be gleaned from a press release issued by the US Department of Energy (DoE). It came out of the SUPERCOMPUTING 2002 Conference held last November in Baltimore, MD. The press release announced that the DoE had awarded IBM a $290 (USD) million contract to build the two fastest supercomputers in the world with a combined peak speed of 460 TFlops. To get an idea of the speed computing throughput 460 teraflops represents, the press release states that, “These two systems will have more than one-and-a-half times the combined processing power of all 500 machines on the recently announced TOP 500 List of Supercomputers.”
The first system, “ASCI Purple,” [apparently the DoE likes colorful names] will be the world’s first supercomputer capable of 100 Tflops. ASCI Purple will have a massive cluster of POWER-based IBM eServer systems and IBM storage systems. This supercomputer represents a fifth-generation system under the Advanced Simulation and Computing Initiative (ASCI) Program. It will serve as the primary supercomputer for DoE.
According to the press release, the second system will be a research machine called Blue Gene/L. It will employ advanced IBM semiconductor and system technologies based on new architectures being developed by DoE and IBM. Blue Gene/L is expected to achieve a peak performance of 360 TFlops with 130,000 processors running under the Linux operating system. It will have the capability to process data at a rate of one terabit per second, equivalent to the data transmitted by ten thousand weather satellites. Applications are expected to include the simulation of very complex physical phenomena in areas such as turbulence, biology and high explosives.
The ASCI Purple system will use IBM’s next generation microprocessor, the POWER5, employing a total of 12,544 of them. These 12,544 processors will be spread among 196 individual computers. The total memory bandwidth will be 156,000 GBs, the equivalent of simultaneously playing 31,200 DVD movies. A super-fast data highway with a total interconnect bandwidth of 12,500 GB will interconnect the 196 computers. The IBM AIXL operating system will be used to run this configuration. The operating system will contain 50 terabytes of memory, an amount that is 400,000 times the capacity of the average desktop PC. There will also be two petabytes of disk storage or holding the content of approximately one billion books.
Finally, since the UNIVAC-1’s introduction, raw computer speed has increased by about 11 to 12 orders of magnitude in about 50 years, or a factor of 10 every five years. This is a truly remarkable achievement. It’s also interesting to contemplate that, if this growth continues over the next 50 years, then by the 100th anniversary of the UNIVAC-1, computers will be operating at speeds on the order of 1023 Flops!
參考答案:不久的将来超级计算机方向
一些可能在不久的将来内在超级计算机设计中发生的东西概念能从一个被美国能源部 (母鹿) 发行的新闻稿被拾落穗。 资讯科技在巴尔的摩， MD 从去年十一月被拿着的高速计算 2002 会议出来。 新闻稿宣布了母鹿已经裁定一个 $290(USD) 百万缩短用 460 TFlops 的组合山顶速度建立全球的二个最快速的超级计算机的 IBM 。为了得到 460 teraflops 表现的速度计算生产的一个主意，新闻稿说那,"这二个系统将会有超过 -和-一-一半乘在最近宣布的超级计算机的最高 500本目录上的所有 500部机器的组合处理力量。”
第一个系统， " ASCI 紫色 ",[显然地母鹿喜欢五彩缤纷的名字] 将会是世界的第一个超级计算机能够 100 Tflops。 ASCI 紫色将会有权力基地 IBM eServer 系统和 IBM 储藏系统的庞大群。 这一个超级计算机在先进的模拟和计算机之下表现一个第五的世代系统率先 (ASCI) 计画。 资讯科技将会为母鹿视为主要的超级计算机。
依照新闻稿，第二个系统将会是一部被称为蓝色的基因/ L. 资讯科技的研究机器将会雇用先进的 IBM 半导体和以被母鹿和 IBM 发展的新建筑学为基础的系统技术。 蓝色的基因/ L 被期望用在 Linux 操作系统人之下跑的 130,000个处理器达成 360 TFlops 的山顶表现。 资讯科技将会有能力以 terabit 的比率每一秒处理数据, 被传输一万气象宇宙站的数据同等物。 申请被期望包括在区域 , 像是喧嚣，生物学的非常复杂实际的现象模拟和高的炸药。
ASCI 紫色系统将会使用 IBM 的往后代微处理器， POWER5,雇用他们中的总数为 12,544 个。 这 12,544个处理器将会在 196部个别的计算机之中被传布。 总记忆带宽将会是 156,000个亿位元组,同时地玩 31,200部数字化视频光电影的同等物。 和总数的一条超级快速的数据公路互相连接 12,500个亿位元组的带宽将会互相连接 196部计算机。 IBM AIXL 操作系统人将会用来跑这一个结构。 操作系统将会包含记忆的 50个兆位元组， 400,000 次平均的桌面个人计算机的能力数量。 也将会有磁盘片储藏的二 petabytes 或握住大约十亿本书的内容。
最后, 因为 UNIVAC-1's 介绍,生的计算机速度每五年已经被大约 50 年 , 或一个 10 的因素大约大小的 11 到 12个次序增加。 这是一项真实显着的成就。 资讯科技也是有趣注视那,如果这生长在未来的 50 年期间继续,在第 100个周年之前然后 UNIVAC-1 ，计算机将会以在 1023 砰然落下的次序上的速度操作!
