www.91蝌蚪I在线看h网站I亚洲黄色特级片I欧美做爰猛烈床戏大尺度I五月婷婷六月丁香I国产免费一区二区三区四在线播放I日本精品视频在线播放I日韩免费片I午夜av成人I黄色a级大片I国产尻逼I妺妺窝人体色777777I伊人日韩I在线观看91视频I尤物在线精品I欧美洲69xxxxxI欧美午夜精品理论片a级按摩I日韩污视频I精品中文字幕在线观看I台湾少妇伦欲电影I秋霞影院avI日本特级淫片I久久久视I美女撒尿无遮挡网站I日本在线wwwI久久这里只有精品6

The high-performance computing applications are diversified. The computing programs require different computing resources in physics, chemistry, biology, CAE/CFE engineering computing and weather and ocean numerical forecast, but most applications require high-density computing and high-perofrmance CPU float computing of the computing system. Some applications such as gene sequence serach and match mainly require integer computing. Some applications can be accelerated by using co-processor such as GPU. The high-performace computing is divided by execution mode into serial, parallel and concurrent. The mainstream parallel model includes the message transmission parallel model such as MPI and sharing memory multi-thread parallel model such as OpenMP. The former realizes the parallelism cross computing nodes. The later realizes parallelism inside a computing node. in addition, different programs require different memories, disk IO and network communication, such as electromagneitc simulation and gene sequence combination requiring mass memory and high disk IO.

So the node type selection and hardware configuration of the computing sub-system should be matched with the application features. The hybrid architecture based on blade cluster plus multi-channel fat node and GPU co-processing unit is recommended for the integrated high-performance computing center to completely cover the application requirements. The Sugon TC3600/TC4600 blade server can meet requiremetns of most applications and features high computing density, low whole power consumption, high reliability and easy managmeent. The Sugon A840, I840 four-channel server and I950 8-channel server is used as the fat node of the computing system. It features high standalone performance, big memory capacity and high local IO and is suitable for some standallone parallel sharing memory multi-thread parallel program and applications requiring high-magnitude memory or high local IO speed. In addition, GPU and MIC co-processor can acceleate some applications remarkably as a supplementary means.