{"id":540,"date":"2011-03-10T16:04:41","date_gmt":"2011-03-10T15:04:41","guid":{"rendered":"http:\/\/cerezo.name\/blog\/?p=540"},"modified":"2024-10-14T14:22:58","modified_gmt":"2024-10-14T12:22:58","slug":"amazon-will-crown-the-top500","status":"publish","type":"post","link":"http:\/\/cerezo.name\/blog\/2011\/03\/10\/amazon-will-crown-the-top500\/","title":{"rendered":"Amazon Will Crown the <span class=\"caps\">TOP500<\/span>"},"content":{"rendered":"<p style=\"text-align: justify;\">The computer system architecture used for <span class=\"wikinvest-suggestion wikinvest-concept\"><span class=\"wikinvest-suggestion wikinvest-concept\">cloud computing<\/span><\/span> provides a powerful computing environment for web services, but not for the tightly coupled parallel processes required for High Performance Computing (<span class=\"caps\">HPC<\/span>), rendering the vast data centers built for the cloud useless for supercomputing purposes. This fact is very well presented in the paper \u201c<a href=\"http:\/\/www.cs.utexas.edu\/users\/pauldj\/pubs\/uchpc09.pdf\" target=\"_blank\" rel=\"noopener\" class=\"broken_link\">Can Cloud Computing reach the <span class=\"caps\">TOP500<\/span><\/a>\u201d:&nbsp;it is shown that the cost for solving a linear system increases exponentially with the problem size in Amazon <span class=\"caps\">EC2<\/span>, the opposite of what happens in a genuine supercomputer.<\/p>\n<p style=\"text-align: justify;\">A real market demand for the ability to run <span class=\"caps\">HPC<\/span> workloads in the cloud lead Amazon to offer a new <a href=\"http:\/\/aws.typepad.com\/aws\/2010\/07\/the-new-amazon-ec2-instance-type-the-cluster-compute-instance.html\" target=\"_blank\" rel=\"noopener\" class=\"broken_link\">Cluster Computing Instance<\/a>, increasing performance of typical <span class=\"caps\">HPC<\/span> applications by an order of magnitude more than the one offered by the previous <span class=\"caps\">EC2<\/span> instance types. This supercomputer processes information at the rate of 41.8 TeraFLOPS and ranks at <a href=\"http:\/\/www.top500.org\/system\/177002\" target=\"_blank\" rel=\"noopener\">#231<\/a> of the <span class=\"caps\">TOP500<\/span> list as of November 2010. What it\u2019s really noteworthy is that Amazon released a new <a href=\"http:\/\/aws.typepad.com\/aws\/2010\/11\/new-ec2-instance-type-the-cluster-gpu-instance.html\" target=\"_blank\" rel=\"noopener\" class=\"broken_link\">Cluster <span class=\"caps\">GPU<\/span> Instance<\/a> on Amazon <span class=\"caps\">EC2<\/span> some months later, but the <span class=\"caps\">LINPACK<\/span> benchmark was calculated without using the available GPUs, so the Amazon Cluster must really rank much higher in the <span class=\"caps\">TOP500<\/span> list: since almost every supercomputer at the top of the list now uses GPUs, it has become an almost mandatory prerequisite (except for Cray). The <a href=\"http:\/\/blog.cyclecomputing.com\/2010\/11\/a-couple-more-nails-in-the-coffin-of-the-private-compute-cluster-gpu-on-cloud.html\" target=\"_blank\" rel=\"noopener\" class=\"broken_link\">available benchmarks<\/a> show excellent performance, but the real <span class=\"caps\">GPU<\/span> speed-up on a standard benchmark remains unknown.<\/p>\n<p style=\"text-align: justify;\">To get at the top of the list, Amazon should provide more <span class=\"caps\">RAM<\/span> per node and much better interconnects than <span class=\"caps\">10GB<\/span> Ethernet (vg. Infiniband <span class=\"caps\">FDR<\/span> or Gigabit Ethernets). One thing is for sure: meanwhile, I\u2019m having fun renting the <span class=\"caps\">GPGPU<\/span> clusters from $2.10\/server\/hour, each node delivering a TeraFLOP using the dual <span class=\"caps\">NVIDIA<\/span> Tesla <span class=\"caps\">M2050<\/span> GPUs and immediately accessing the performance of an <span class=\"caps\">HPC<\/span> Cluster <span class=\"caps\">GPU<\/span>, with no upfront investment or long-term commitment.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>The computer system architecture used for cloud computing provides a powerful computing environment for web services, but not for the tightly coupled parallel processes required for High Performance Computing (<span class=\"caps\">HPC<\/span>), rendering the vast data centers built for the cloud useless for supercomputing purposes. This fact is very well presented in the paper \u201cCan Cloud Computing&nbsp;[\u2026]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wp_typography_post_enhancements_disabled":false,"ngg_post_thumbnail":0},"categories":[15],"tags":[],"_links":{"self":[{"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/posts\/540"}],"collection":[{"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/comments?post=540"}],"version-history":[{"count":6,"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/posts\/540\/revisions"}],"predecessor-version":[{"id":1656,"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/posts\/540\/revisions\/1656"}],"wp:attachment":[{"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/media?parent=540"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/categories?post=540"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/cerezo.name\/blog\/wp-json\/wp\/v2\/tags?post=540"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}