I asked the audience a pressing question at the Quandl Alternative Data Conference last week: “How many of you know you can do more with GPUs than just play Doom or Quake?”
The crowd laughed as they raised their hands. The word is getting out. GPUs aren’t just for video games anymore. They’ve already revolutionized AI. Now they’re transforming analytics.
We’ve all heard by now that the world is drowning in data. It’s not just traditional data warehouses anymore, but data streaming in from point-of-sale (POS) terminals, credit card systems and social media, not to mention all the sensors and telemetry from cars, phones and IoT devices.
There’s enormous value in all this data. But the more of it there is, the harder it is to get at the insights that matter when those insights are still relevant.
That’s where GPUs come in. Until the early 2000s, CPU power was increasing 30 to 40 percent per year. Then engineers hit a speed wall. Now growth looks more like 15 to 20 percent annually.
Meanwhile, GPU performance is leaping ahead by more than 50 percent per year. The reason has to do with their architecture. A CPU-based server might have 10 or 20 fast cores that process in series, where a GPU-based server might have 40,000 slower ones that can process in parallel. On a core by core basis, the CPU wins. But collectively, the GPU cores leave the CPU in the dust.
As I put it to the Quandl audience: “Would you rather have 10 or 20 knights on your side, or 40,000 angry peasants?”
MapD’s SQL engine harnesses the power of the computational mob. Not only does it run on GPUs, but on multiple GPUs per server and multiple servers per cluster, giving it the ability to query tens of billions of records in milliseconds.
That kind of performance changes the way analysts approach data. With CPU-based systems, queries might take hours to complete. Analysts only asked “safe” questions they knew they could answer quickly. Now they’re free to ask anything they want. They can get insights faster and deploy them in the market faster.
At the Quandl conference, I showed the audience how to discover behavioral insights buried in a mashed together data set with hundreds of millions of rows and 260 different features. I demonstrated how to identify the strongest variables in this data set with an eye toward training machine learning segmentation models.
The financial services industry — particularly investment banks and hedge funds — use MapD to find alpha (active return on investment) in the markets, explore large data sets and test hypotheses in real time.
I’m convinced that we will cross the next frontier using GPUs for general purpose data analytics.
One day, I’ll ask attendees at an analytics conference: “How many of you know that GPUs are not just for AI and analytics, but also for video games?”