This article is more than 1 year old

US Army's spun-off GPU database gets ready for more matrix operations

Also: We could rebuild Trump's social score, we have the technology

Kinetica, the in-memory GPU-accelerated database, is adding user-defined functions (UDFs) to contribute to more sophisticated analytic workloads.

Eric Mizell, Kinetica's global solution engineering veep, told The Register that the company was keen to discourage the idea of “data science as a black box”.

Although Kinetica believes offering customers the opportunity to dodge the tedium of defining schemas would be a good sell, Mizell acknowledged the company was “still in the convincing mode” rather than eating into the competition's market. He was full of praise for the city's fintech, telling us: “Financial groups in London are more advanced than anywhere else I've seen, I have to give you guys props.”

High performance computing workloads are often “homegrown”, said Mizell. While the financial and pharmaceutical sectors are a little mature here, “major enterprises, the big companies, don't know what 'GPU' means,” he told us.

Kinetica began in 2009 as a US Army research project at Fort Belvoir, in Virginia. The military had found itself unable to take advantage of its datasets on potential threats due to the lack of speed and scale available on the market at that time. Kinetica's founders managed to move matters in-memory, utilising GPUs to execute the parallel functions of SQL queries to massively speed up the processing of queries against databases. You can you can read about this in-depth on The Next Platform.

The patented technology was commercialised in 2014, and the business is now committing to those workloads, according to its CTO and co-founder, Nima Negahban: “GPUs are well suited to the types of vector and matrix operations found in machine-learning and data science. UDFs in Kinetica open the way for these advanced analytics workloads to take advantage of the GPU, in-database, and alongside BI workloads.”

Running libraries from TensorFlow, BiDMach, Caffe, and Torch in-database would “democratise” data science workloads, according to Kinetica. “Businesses get more efficient and effective business process outcomes, faster time to market, and net new business value,” the company stated.

Additionally, Kinetica has launched "Reveal" for interactive real-time data discovery. The data exploration framework is intended to allow business analysts to make faster decisions by visualizing and interacting with billions of data elements instantly, without executing SQL commands but just dragging and dropping data tables to play with the data and create on-the-fly data analytics.

At the moment, Reveal has over a dozen analytical widgets to create dashboards, as well as featuring mapping capabilities and integrating with mapping providers including Google, ESRI, MapBox, and Bing, to conduct interactive location-based analytics on massive datasets.

The Register spoke to Mizell the day before Donald Trump's inauguration, and when we raised the topic, he suggested Kinetica could “help with his social score”.

“People are calling for [Donald Trump] to stop tweeting,” Mizell explained, adding that the firm had done "some work" for a nominee candidate in the recent US presidential election. He said Kinetica had created “a social sentiment UI for them” utilising the kind of UDF features which the company is announcing – and disappointingly bringing the discussion back on topic.

“We can do some interesting things with advanced analytics; that traction is driving enormous interest from customers,” said Mizell. ®


Similar topics


Send us news

Other stories you might like