IBM teams with Nvidia to use GPUs to boost AI research

Faster with need for fewer servers

IBM has teamed with GPU experts Nvidia to give a new generation of artificial intelligence (AI) applications "unmatched capabilities".

Speaking on stage at the tech giant's Think Conference in Las Vegas, IBM's senior vice president of cognitive solutions, John Kelly, said the pair have teamed up to "to turn raw data into intelligence".

At the gathering of more than 40,000 technologists, consultants, developers, and business leaders, Kelly explained: "We've chosen a great partner, Nvidia, to be our partner to accelerate artificial intelligence and machine learning.

"It's a very successful company and a world leader in artificial intelligence with their GPUs."

Using Nvidia's GPUs, IBM can reduce the number of servers it needs to harness data 50 times faster on a key benchmark, the firm said.

Kelly invited Nvidia CEO Jensen Huang on stage to further explain the reason behind the partnership. Huang said companies are racing to train a new generation of AI applications that will give them unmatched new capabilities.

"I believe every company in the world will have high-performance computers or supercomputers inside," the Nvidia CEO explained. "Instead of engineers doing coding, you will have supercomputers learning, and whoever learns the fastest will get to market soonest."

Ahead of Huang's appearance, IBM announced that a team at its Zurich research lab tapped Nvidia Tesla V100 GPUs to beat a key performance benchmark for a machine learning workload by 46 times.

Not only did it accomplish the feat far faster, but it did so using fewer servers than the previous record holder.

The team used a terabyte-scale dataset to predict clicks on advertisements in 91.5 seconds, compared to the previous record of 70 minutes, set in February of last year.

This kind of "logistic regression" analysis is a key tool of the trade for machine learning professionals, IBM claimed.

Kelly explained that more is on the way, arguing that machine learning will transform industries as diverse as retail, oil and gas discovery, automobiles, shipping, and healthcare.

"I cannot think of a single industry that is not only going to be not only impacted, but completely transformed," he added.