VizWorld.com

Ever since GPU’s began to encroach on the long-held arenas of CPU’s, there has been one realm where GPU’s could not go: Databases.  Inherently complex and problematic, they simply did not lend themselves to the type of atomic massively-parallel algorithms that GPUs excel at.  Many have tried, but most have wound up with hybrid solutions that generally caused more problems that answers.

empulse GmbH, a german company working on database technology, and chosen NVidia’s GTC as the launching pad for their newest product “ParStream” that shatters that stereotype once and for all.  They have the world’s first completely-GPU driven database system, capable of petabytes of storage across massive clusters, all searchable in a fraction of a second.  In a short discussion with Managing Director Micheal Hummel, he told me of a web analytics company that was currently using a process that took approximately 3 minutes to execute a query, that ParStream was able to replicate in a mere 15ms.  Yes, that’s milliseconds.  The results were so astounding that they first thought their original system must be broken, so they compared ParStream against every publicly available columnar database, and found ParStream was still 35x faster than the 2nd-place winner.

Thanks to its use of massively-parallel GPUs, ParStream is unsurpassed in its performance abilities. “Based on the current state of technology, the technological advancement of empulse in the area of GPU-based databases and the complexity of such an endeavour, it is unlikely that competitors will be able to develop a comparable product within the next two years,”

Unlike traditional transactional database systems like Oracle, this is an analytical database.  That means that insert and update operations happen infrequently, and as such it is optimized for queries and reporting.  That makes it perfect for web analytics, stockmarkets, and genome projects which have massive input datasets that change on a regular non-realtime basis (once a minute, etc).

Read the full press release after the break, and if you’re at GTC you can hear more from empulse at their booth (#39) or Wednesday at 4:00pm via their “CEO On Stage” presentation.

Full announcement after the break.

Read more…

Science database, empulse, gpgpu, parstream

Got a presentation coming up and need some data for a graph?  Well, Data360 aims to please with their open data collection.

Data360 is an open-source, non-profit and free website. The site hosts a common and shared database, from which any organization which is committed to neutrality and non-partisanship (meaning “let the data speak”), can use the site for presentation of their reports and visualizations about the data.

They’ve got a great collection of visualizations and datasets, all freely available.  Some of the charts we’ve covered before as they come from the New York Times or USA Today, but there is a lot of new material there.

via Data360 Homepage. via Simple Complexity.

Science data, database, list, Website