UPDATED 16:00 EDT / AUGUST 16 2018

INFRA

The GPU takeover: AI screams for more, more, more cores in the data center

The niche technology once developed to boost PC gaming is all grown up. Specialized graphics processing units are gaining popularity in enterprise-grade computing for their ability to aid in artificial intelligence, thanks to their parallel processing muscle. Some say that as the trend barrels forward, GPUs will essentially become the new central processing units in data centers of enterprises and cloud providers.

Nvidia Corp. is the standout GPU maker sallying into data centers attracting storage partners such as NetApp Inc. and Dell EMC.

While lots of public cloud providers have built GPUs into their infrastructure as a service offerings, some enterprises opt to run them on-premises for a number of reasons, among which are the usual suspects — privacy, security and the ability to custom configure them for specific workloads. Piecing GPUs together with all the other components necessary for on-prem AI can get complicated and costly.

“We think you’ve got to really look at [total cost of ownership], because customers want to build these great solutions for the business, but they can’t afford it unless vendors give them viable options,” said Santosh Rao, senior technical director of AI and data engineering at NetApp.

Rao  spoke with Peter Burris (@plburris), host of theCUBE, SiliconANGLE Media’s mobile livestreaming studio, during the “NetApp Journey to AI” event held at NetApp’s Data Visionary Center in Sunnyvale, California. Burris spoke with Rao and other thought leaders at the event about how to best deploy and scale AI in your organization. They also discussed NetApp’s partnership with Nvidia and how GPU computing and data management can shorten the trek to AI Shangri-La. (* Disclosure below.)

NetApp is one of a host of companies to strike up partnerships with Nvidia to bring GPUs’ massive parallel compute power to on-prem infrastructure products. The just-announced NetApp ONTAP AI proven architecture leverages Nvidia DGX supercomputers and NetApp AFF A800 cloud-connected all-flash storage. It is a simple, pre-integrated solution built to “simplify, accelerate and scale the data pipeline across edge, core and cloud for deep learning deployments and to help customers achieve real business impact with AI,” according to NetApp’s website.

In recent years, NetApp has zeroed in on the problem of holistic data management, which is key to integrating massive data sets from disparate sources for accurate AI insights, according to Rao.

Watch the complete video interview with Santosh Rao below:

AI to the core

It’s hard to overstate the advantage that GPUs have over CPUs in parallel processing. A CPU typically has just a few cores for sequential serial processing; GPUs might have hundreds or thousands of smaller cores designed to process multiple tasks simultaneously.

SQL is a standard language for storing, manipulating and retrieving data in databases. It performs a specific operation on each row in a data set. GPUs assign each core to a single row in a data set, thus chewing up and spitting out SQL queries much faster than CPUs can. That is how parallel processing works, in a nutshell.

Years ago, some savvy developers started making applications for GPUs with a language Nvidia developed called CUDA. CUDA uses a syntax that can make calls to a GPU instead of a CPU, and can do so thousands of times in parallel.

CUDA is one of the reasons for Nvidia’s GPU market dominance both inside and outside of gaming. The company suffered a dip in the last quarter, but overall its earnings grew impressively over the last year. Its only real competitor in the add-in-board space, Advanced Micro Devices Inc., ended Q1 with roughly 35 percent market share compared to Nvidia’s 65 percent, according to a Jon Peddie Research report. Programming an AMD GPU requires use of a library called OpenCL that does not enjoy the broad support of CUDA.

“I give Nvidia a lot of credit. They set up at universities all over the world, hundreds of them, to teach CUDA,” said Jon Peddie, president of Jon Peddie Research, as quoted by ComputerWorldHK. “So when a student graduated, they were pre-trained CUDA developers and set the foundation for getting CUDA into industries as we know it today.”

If GPUs are indeed becoming the new CPUs, is hardware-chip legacy Intel Corp. going to watch from the sidelines as Nvidia and AMD eat the market? Intel has built graphics into its CPUs for some time, of course, but it’s been missing in action in the bazaar of standalone GPUs for the data center. When Intel hired Raja Koduri, the ex-chief architect of AMD’s Radeon GPU unit, earlier this year, one could already catch a whiff of GPUs cooking.

This past June, the company tweeted that its first discrete GPUs will be hitting the streets come 2020. Details remain scant; no one knows for sure whether Intel will gear them more toward the gaming market or go for the throat of exploding data-center and cloud-service AI.

Data is the new source code

Big data and GPUs together can perform tasks and generate insights that used to require much more time and human labor, according to Jim McHugh, vice president and general manager at Nvidia. His company is focused on novel ways of writing software where data actually informs the finished product.

“We say data is the new source code,” McHugh said. “Instead of having humans going in and do the feature engineering and feature sets that would be required, you’re letting data dictate and guide you on what the features are going to be in software.”

Nvidia gobbles up all the feedback it can get from channel partners using GPUs for data center infrastructure and cloud services, McHugh explained. “They are that last mile. They are those trusted advisers,” he said.

The business rewards of quickly processing data with GPUs and data aggregation software are cropping up across industries. It can help companies fail fast and iterate rapidly, leading to well-honed new products and services, according to Monty Barlow, head of artificial intelligence at Cambridge Consultants. The consultancy helps clients put technology to use toward practical outcomes.

“Key to our business is a fast turnaround on proof of concepts; how would this work, what would happen?” Barlow explained. Customers may have a certain amount of data and desire to run a trial to find out if they should collect more, for instance. “Getting through jobs quickly is what matters most to us, and that’s what the Nvidia and NetApp equipment is all about,” he added.

Watch the complete video interview with Monty Barlow below:

The data integration capabilities that NetApp brings to the partnership can help scale a product or service from pilot to a full roll-put, according to Octavian Tanase, senior vice president at NetApp Inc. “That seamless data management across the edge to the core to the cloud is also important,” he said. 

Big data analytics and AI are making an impact in businesses from startups to blue chips. For example, precision agriculture is using AI to enhance crop yield; planes can fly over a crop, distinguish weeds from produce, and apply pesticide or water accordingly. 

“You can now point a camera at a road and recognize all of the different vehicle types instead of just how many axles they’ve got,” Barlow said. 

Out in the trenches, businesses like Bloomberg are using GPUs to kill off major time consumption in compute-intensive jobs. While GPUs process more data ounce-for-ounce than CPUs, they also draw more power. That trade-off has to be checked continuously, according to David Rosenberg, a data scientist in the Office of the CTO at Bloomberg LP, as quoted by Advanced HPC.

“We’re constantly looking at power,” he said. “If we put 500 GPUs in a data center, that’s a lot of other computers that can’t be there. GPUs are more power efficient for the compute they provide than CPUs. It’s just that they are doing so much more compute than CPUs that they end up using a ton of power.”

Watch the complete video interview with Jim McHugh and Octavian Tanase below:

Be sure to check out more of SiliconANGLE’s and theCUBE’s coverage of the NetApp NVIDIA AI Digital Launch event. (* Disclosure: TheCUBE is a paid media partner for the NetApp/Nvidia AI Digital Launch event. Neither NetApp Inc., the event sponsor, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE

A message from John Furrier, co-founder of SiliconANGLE:

Your vote of support is important to us and it helps us keep the content FREE.

One click below supports our mission to provide free, deep, and relevant content.  

Join our community on YouTube

Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.

“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well” – Andy Jassy

THANK YOU