Walmart, NVIDIA Discuss How They’re Working Together to Transform Retail


Walmart’s the biggest retailer on Earth. It’s also one of the most competitive technology companies around. Two of its sharpest tech tools: NVIDIA GPUs and RAPIDS data science software.

Speaking at the VentureBeat Transform conference in San Francisco on Thursday, Walmart Chief Data Officer Bill Groves discussed Walmart’s data processing and machine learning ecosystem.

“The paradigm has flipped, now we have things available to us that we never had,” Groves said in a conversation with Josh Patterson, NVIDIA’s general manager for data science, noting that, in the past, data scientists didn’t have access to the computing power to put ideas like neural networks to work.

“GPUs are going to make technologies like that available to us and we’re going to solve a lot of problems,” he added.

One of his key tools: NVIDIA RAPIDS. The open-source suite of GPU accelerated libraries helps Walmart tear through their massive-scale data analytics and machine learning.

“This is data science everyone coming out of college knows, we just want to scale it, put it on GPUs,” Patterson said. “We’re enabling them to do more, more quickly.”

Groves’ team pairs RAPIDS with two other technologies, Dask and XGBoost.

They use Dask, a library for parallel computing in Python to scale out data science workloads.

They also use XGBoost, a popular machine learning algorithm, to train their machine learning models on servers equipped with multiple GPUs.

The result of this combination: Groves’ team can crunch data at speeds 100-times faster than traditional methods.

Technology industry analyst Maribel Lopez, left, moderated a conversation between Wal-Mart’s Bill Groves, center, and NVIDIA’s Josh Patterson.

That’s key for Walmart. From product forecasting to supply chain management and predicting consumer buying trends and last-mile delivery, machine learning algorithms help Walmart serve their more than 265 million weekly customers even better.

Today, Walmart is using the technology for product forecasting of more than 500 million store-item combinations each week, Groves explained.

“More compute power allows us to bring in more data and get better faster,” Groves said. It’s simply retail 101 that events like weather and local sporting events, drive sales, he explained. GPUs give his team the computing power to harness data about such events.

“So when the customer comes in the product they want is sitting on the shelf,” Groves said. “That’s what we’re always working towards.”

Going forward, Walmart plans to expand the technology to other use-cases – including supply chain, logistics, merchandising, and other in-store and eCommerce functions.

Working together, Walmart and NVIDIA can integrate cutting-edge GPU technologies sooner. That gives Walmart a first mover advantage, saves them money, and steers development and optimization to meet specific use-cases.

“We can’t build everything, nor do we want to,” Groves said. “We really want to build this foundation, this platform, to let this next generation of data scientists solve these problems.”

For NVIDIA, collaborating with Walmart lets it test its tools and technology against real-time production data from the world’s largest retailer – at a scale unlike any other.

“Having this team really allows the ecosystem to grow really quickly, and also ensure we have useable software,” Patterson said.

Improved Product Forecast Accuracy

So far, Grove reports they’ve seen a significant improvement in product forecast accuracy – which requires only four hours to run models that would take several weeks on CPUs.

And one GPU server requires only four percent of the time needed to run the same forecasting models vs a 20-node CPU server, Groves explained.

Without that power, data scientists had to “dumb down,” their algorithms so they would run fast enough.

No longer. “GPUs allow us to do things we couldn’t do before,” Groves said.

That translates into less idle time for Walmart’s data scientists. So they can test more new features and reduce development cycles, Groves said.

“Math and science hasn’t changed in 30 to 50 years,” Groves said. “What has changed is the technology that enables it, so that we have opportunities that we didn’t have until recently.”

The result: Walmart gets more computing power at a lower cost and with less environmental impact. All for efforts that result in more choices and lower prices for the millions of consumers who shop at Walmart every day.

error: Content is protected !!