GPUs Set to Boost Business Intelligence

Over the last decade, GPU computing has evolved from a fringe technology to a core element that powers some of the world’s top supercomputers.

Plugging in GPUs to boost Top500 performance is not simply an academic endeavor. Outside of the massive academic ad national lab systems, GPUs have also been finding practical applications in the enterprise world, operating against massive data sets on cloistered internal clusters committed to running advanced complex event processing, business intelligence, and other real-time, big data analytics workloads.

GPU-driven companies like NVIDIA have been emphasizing the practical application of GPU computing in enterprise settings—and this year at their annual GTC conference, there are several compelling sessions for the business intelligence and enterprise data analyst attendees.

Below are just a few sessions of note for this year’s GTC lineup that showcase how GPUs are being used to accelerate and enhance business intelligence, data mining and machine learning.

Efficient Top-Down Planning in Business Intelligence

This intermediate-level session, led by Jedox AGs’ Tobias Lauer and Alexander Haberstroh will address the complex topics of BI in the context of corporate planning and what-if analysis. According to the duo, “one main difference is that while the latter only read data, the former require the change of possibly large numbers of existing and creation of new data records in the business model, preferably in real time.”  In this session they will describe an extension of an existing BI tool, Jedox OLAP, by GPU-based parallel algorithms for interactive planning scenarios. They claim that compared to sequential in-memory algorithms, our CUDA approach yields tremendous speedups and can also cope with large amounts of data by using multiple GPUs.

30x Faster Regular Expressions on a GPU

Here is what you might to learn more about—how to make your data work for you faster. According to HP’s David Lehavi, GPUs can be used to accelerate a regex engine—and that all previous attempts to prove this concept have not been reaching their potential. Lehavi wants to show how regex presents imbalanced compute workloads which are very different from common GPU applications (CFD, CG and image processing). Hence, he says, they can teach us general lessons on how to utilize GPUs for more general workloads.Our initial results show 30x improvement in running time relative to single threaded commercial regex engines.

Efficient K-Nearest Neighbor Search Algorithms on GPUs

Nikos Pitsianis from Aristotle University in Greece and Xiaobai Sun from Duke will lead this BI, data mining and machine learning-focused session aimed at beginners. They will present a selection and combination of different algorithms that perform exact k-nearest neighbors search (k-NNS) on GPUs and outperform the competition. The duo will show four different selection algorithms designed to exploit differently the parallelization of the GPU according to the relative size of the corpus data set, the size of the query set and the number of neighbors sought. They also plan to show the application of Logo Retrieval with SIFT vector matching on two different GPUs, the Tesla C1060 and the Fermi GTX480.

Risk Management with Parallelized Algorithms on GPUs

Partha Sen, who works with Fuzzy Logix, will deliver some fodder for the big data, financial services folk who tend to turn out in droves for the GPU conferences (even if they won’t talk to us about the specifics of what they’re working on…ahem). This is probably one of the most advanced sessions that BI/data mining folks could select, but it sounds fascinating. As Sen describes, “The challenge with intra-day risk management is that a very large number of calculations are required to be performed in a very short amount of time. Typically, we may be interested in calculating VaR for 100 to 1000 securities per second based on 100 million potential scenarios. The magnitude of these calculations is not Utopian but it reflects the reality of modern financial institutions and exchanges. In this presentation, we outline how the complex problem of intra-day risk management can be solved using parallelized algorithms on GPUs. The methodology has been proven in a POC at 2 financial institutions.”

Some of the sessions might sound a little low-level for those who are just considering attending to explore what is possible with GPUs for their own business cases. However, it’s a good way to get acquainted with this ecosystem (both CUDA—OpenCL to a lesser extent) and start seeing the way it fits into the larger enterprise datacenter and application puzzle.

For more, check out the Databases, Data Mining and Business Intelligence section of the descriptions of GTC 2012 sessions here. The conference is coming up–just around the corner in May in San Jose.

Related Stories

GPUs Push Envelope on BI Performance

The New Era of Computing: An Interview with “Dr. Data”

Live from GTC Asia: Accelerating Big Science