I subscribe to Scientific Computing so that it drops into my email inbox. I read Rob Farber’s article this week entitled “The Future Looks Bright for Teraflop Computing “. His opening question was “Wouldn’t it be great to have a teraflop of computing power sitting in your lab, desktop workstation, or remote instrument server?” What would that mean to your work?

For those of you using ChemSpider you will know that we have about 20 million compounds on the database. With that many compounds population of the database with properties such as InChIStrings, InChIKeys, physchem properties and systematic names can take many days if not weeks. With three computers only in our hands, one of them a web server and one of them the database server, we are limited to one system. Even that dual processor system provides slow throughput. Oh the joys of having access to teraflop processors!!!

In my previous post on focused libraries I commented on ongoing discussions regarding the potential to perform online docking. Evangelists such as Jean-Claude Bradley (on our advisory group) have been talking about this possibility as part of his approach to Open Notebook Science. Docking can be very time consuming and the speed of calculations is very important. I have been working on a project regarding the value of porting docking software to the Cell Broadband Engine processor from IBM. The development of that processor is an interesting story in itself since it was driven specifically by the needs of the gaming industry for better performance in their calculations. Now SimBioSys are porting their docking software to the Cell processor as described in this White Paper. The improvement in performance is quite amazing!!!

While working for a commercial software company we saw productivity gains moving to clusters. Dual processors in our laptops and annual performance gains from the general technology shifts offer faster calculations every year. Teraflops on the desktop (and even laptop) are likely a few years away…but GFlops are here..

Stumble it!

2 Responses to “High Performance Computing – What Accessibility to TFlops Can Offer Scientists: PlayStations and the Cell Broadband Engine”

  1. Rajarshi says:

    Have you considered using something like Hadoop, given that your problem of computing X for millions of compounds (where X is some property) is of the trivially parallelizable class of problems? Of course you’d still need to the 1000′s of nodes, but Yahoo has some very impressive results for Hadoop

  2. A Green Solution for Virtual Screening Using the IBM Cell Broadband Processor at The ChemConnector Blog - Observations and Musings for the Chemistry Community says:

    [...] blogged previously about the possibility to derive processing power from a gaming system (1,2). By the time that eHITS Lightning was unveiled at Bio-IT the Cell Processor had managed to [...]

Leave a Reply