Founders make intuitive leaps.
Maybe they are making a bet on a looming demographic trend combined with a technical capability, such as Seamless providing meals on demand to well-heeled urban dwellers who can’t cook. Or Birchbox founder Katia Beauchamp’s realization that today’s tech-savvy women would order cosmetics online if they could sample the products first. For Todd Mostak, the realization was far more esoteric.
Mostak was a Middle Eastern studies major sitting in an elective computer science class when he realized that memory in graphics processors was going to get bigger because of the demands of high-definition gaming and 4K televisions. And because at that time he was also completing a thesis analyzing a huge number of tweets from the Arab Spring uprisings, he decided to use his eureka moment to build a SQL database that relies on graphics processors instead of traditional AMD or Intel chips.
This led Mostak to found MapD, a startup that later garnered $2 million in seed funding from Google Ventures and graphics chip maker Nvidia (NVDA). MapD is currently raising a Series A round of about $8 million to $9 million.
MapD’s breakthrough is a subtle one. Graphics processors are usually better at certain types of computing jobs because they are made up of hundreds of components called compute cores that each can tackle a part of the job. This makes them a popular “accelerator” in supercomputing and other jobs like financial trading simulations. But traditionally, a database relies on an x86-based Intel or AMD chip, that can take a few minutes or even a few hours to run through a large set of data. This is because each piece of the job must wait its turn to go through the limited number of processing cycles the CPUs had to offer.
With GPUs getting more memory, Mostak realized that the Intel and AMD part of the database can now take a backseat to the graphics processors because the GPUs finally have the memory capabilities to store the data as its being processed. MapD machines still use a CPU for some processing, but it’s a secondary processor instead of the primary one, and as the memory in GPUs increases, the CPUs will recede further.
The company makes a box that is really great at taking large quantities of data, performing analytics on it and then turning the results into visualizations. Those visualizations can be as simple as bar or line graphs for financial results, or as complicated as maps or bubble charts. For now, the boxes are designed to work with terabytes of data as opposed to the petabytes of data that really large companies might create in a quarter, or even a day depending on what they are tracking.
“We can handle big data, not massive data,” says Mostak.
As next-generation GPUs come out with more memory, the amount of data MapD can handle will get bigger. Current Nvidia cards that Mostak’s machines use have 192 gigabytes of RAM per box, but Mostak is hoping that next-generation products from Nvidia will have at least a quarter to a half a terabyte of RAM per box within the year. Nvidia confirms that it will deliver cards that would offer 32 GB of RAM within a year, which would enable the MapD box to reach 256 gigabytes of RAM.
Subscribe to Data Sheet, Coins2Day’s daily newsletter on the business of technology.
For more on data, check out the following Coins2Day video:
