For this special episode of the Futurum Tech Webcast, Principal Analyst and host Daniel Newman welcomes Jordan Plawner, Director, Products & Business, Artificial Intelligence at Intel to discuss how cloud service providers are increasingly adopting AIaaS due to recent hardware and software optimizations.
Their conversation includes a look at:
- Integrating AI into cloud workload strategies including how to deploy AI on Intel in the cloud.
- How Intel worked with CSPs to optimize Xeon for machine and deep learning needs, making Xeon the foundation for AI.
- Why service providers are adopting Xeon for their AIaaS needs due to workload flexibility.
- We also explored a few case studies including Kingsoft, a Chinese cloud service provider that adopted Intel’s latest software optimizations to provide internal and external users better AI performance on CPU .
It was an interesting conversation and we’re glad to have you as part of it. Check it out below:
Disclaimer: The Futurum Tech Podcast is for information and entertainment purposes only. Over the course of this podcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we do not ask that you treat us as such.
Other insights from the Futurum Research team:
Image Credit: Computer Business Review
Daniel Newman: Welcome to Futurum Tech TV. I’m Daniel Newman, Principal Analyst, and Founder at Futurum Research and I’m excited to have a conversation about AI and AI’s migration into the cloud. And I’m joined today by Intel’s Jordan Plawner. Jordan, welcome to this little conversation here.
Jordan Plawner: Hi Daniel. Thanks for having me today. My name is Jordan Plawner, as you said, and I’m the Director of AI at Intel Corporation, focused mostly on the data center and cloud.
Daniel Newman: And that’s a big topic right now. AI has made huge inroads, especially in cloud. We’re hearing and seeing so much going on right now. All the big hyperscalers making investments. Intel, you are certainly working closely, side-by-side, with those companies. But in this short conversation, I want to talk a little bit more about what moving into the cloud, and moving AI workloads into the cloud, looks like.
But you know what? Let’s do the whole introductory thing really quick, because just saying that I think sometimes people might have their own interpretations. Jordan. Talk a little bit about what is going on with AI in the cloud. What do you even mean, for the novice, and what are you seeing for those that are starting to really get into this?
Jordan Plawner: Yeah. So a good way to start by thinking about this is that AI itself is becoming ubiquitous. So when you say, where is AI, you have to say, well, what are the customers, or companies, that we’re talking about even doing? So the cloud itself has become incredibly diverse from the first days when we thought about a company just doing search or just doing social networking. These companies have become quite diverse. And there’s three main aspects of their businesses that we can talk about when we talk about AI in the cloud.
One is these customers of ours run their own workloads and they enhance those workloads with AI. So we don’t think of Gmail as an AI workload, but Google is using AI to enhance how Gmail’s operating. So it’s not an AI service. Gmail’s the service. AI is making it more intelligent.
And we can think of AI as a service. So we all hear about how hard AI might be to do, so the CSPs, the hyperscalers, the cloud service providers are taking the learnings that they have made and adopting AI for their internal workloads. And they’re turning that around and trying to mature them and commoditize them, to some extent, and make that as a service, so that you, as a developer, do not actually have to set up all the AI techniques. You can actually try to use AI through an API.
And, of course, the more traditional way is you’re just renting infrastructure in the cloud and you bring your own software and really just renting compute. And the developer and the data scientists set up the AI solution themselves.
Daniel Newman: And we’re seeing a lot in all three of those areas, Jordan. I think, AI, it was one of my top digital transformation trends for the year and with COVID-19 and what’s going on with the pandemic and companies accelerating their transformation, it’s been a big investment area. Conversational, you’re seeing more recommenders, but you’re also just seeing a lot of enrichment of data, volumes of data.
We hear a lot about IoT and a lot about Edge. So as people are thinking about this then, Jordan, how are you approaching working with these clients, these customers, and these users, as they ask questions like, “Hey, how do we integrate AI into our cloud workload strategies?”
Jordan Plawner: Well, there are three levels that we can interact with a customer, right? A more traditional level, especially for a company like AI, is just the hardware and the platform. So we clearly have some customers that just say, give me compute, give me flops, as we like to say, right? They just want compute cycles. And the cloud service providers, or the hyperscalers, are the ones that are the most sophisticated and they want to own the entire software stack, and often they just say, “Give me the hardware.”
And, of course, we have some sophisticated customers also outside the cloud who might be there. But the majority of the market, the broad part of the market, the thousands of customers that are out there, they want our enabling software and our tools. And then they also want ISV’s, or software companies, or SIs, system integrators, to come in with solutions built on top of our hardware and our software. So many customers just go, “I want a fraud detection system. I want an intelligent AI infused fraud detection system.” And they might just go to an SI, or a software company, and say, “Give me that solution running on, say Intel Xeon.”
So we have companies at three different levels and we deliver great hardware, great software, and a great ecosystem of SIs and software vendors who could provide solutions. And by the way, when we say AI, I should mention, we don’t just mean the sexy part of deep learning, the deep neural nets that people are talking about. A lot of companies are still using machine learning, classical machine learning techniques and extracting a lot of value out of them. Some are even looking at how to combine simulation, which is an HPC technique, machine learning and deep learning together. So we define AI at Intel, very broadly, as meaning really all kinds of advanced data analytics.
Daniel Newman: Yeah. And there’s a lot of success stories with Xeon, DL Boost. The companies gained a lot of traction working with your big data systems, the ERP platforms that companies use and being able to accelerate those workloads. Which is really important right now, as companies are trying to better understand customers, better understand their data, optimize fulfillment centers, things like that, where working in these systems, and just the vast amount of data that is being worked with, there’s no way to do it well anymore without enrichment, without acceleration.
It sounds like that’s something that’s really been focused on. And then of course, like you said, there’s the kind of sexier stuff of what’s going to come next, right? The computer that’s talking to you, it’s fully understanding you and cycling conversations. And that’s going to happen.
Jordan Plawner: Yeah.
Daniel Newman: But most enterprises, that’s not today. When they’re thinking about AI in the cloud, they’re thinking about, I’ve got this big deployment of some ERP system and I’ve got all this data and I need to maximize the return, real-time visualize it. Talk a little bit about some of the use cases, or some of the cases out there, that you’re seeing that are customer stories that have been really successful with moving AI to the cloud.
Jordan Plawner: Yeah. So as I mentioned before, there’s just like the use case of instances that people can come in and rent. And so, one of the biggest successes, and the one that gives Intel the most value to customers is that Xeon itself is a ubiquitous compute of the entire data center and enterprise infrastructure. And so you mentioned Deep Learning Boost before. Deep Learning Boost is just a low level instruction accelerator and it’s built into our hardware that can make the deep learning functions operate much faster than they had in previous generations. And if we’re successful, actually, the end user never knows about that capability. Because the way most users engage on deep learning, there’s these popular frameworks like TensorFlow, which I’m sure you’ve spoken about previously. And so most developers and data scientists used high TensorFlow or Pi Charge. And our job is to make sure that those frameworks of developer environments, at the low level, they’re just interfacing with our hardware and taking advantage of the Xeon compute capabilities and the access to large amounts of memory.
And so where we’ve had a lot of success is that there are lots of use cases, especially in the scientific and medical fields, that process extremely large, high definition images. And the more high definition of an image that you can use, the more enriched solutions and answers you can come across. Why do I say that you can use? Because sometimes these images are such high definition that you have to actually downscale them, right? And one of the advantages of a Xeon platform is it has access to all the main memory, and the host memory, that’s sitting there, that’s residing in DRAM.
And so for people who are doing high definition video, and images, they can use a platform. We’re talking about autonomous driving, obviously scientific fields, like astronomy and weather modeling. And, of course, anything in the medical field dealing with extremely high definition images, and the more they can use that raw image, the more likely they’re able to spot, say disease, but actually discover it even earlier than tools can today. So that’s one of the specific areas where we’ve had a lot of success.
Daniel Newman: Yeah, that’s a great example. And, of course, at Intel it really seems that your focus is partnership, it’s ecosystem working side by side with the cloud service providers, the ones here in the United States and those abroad, to provide that compute horsepower. And, of course, the company’s made a lot of investments and has strategies expanding into new areas. There’s been a lot of interesting announcements that I’m eager to see more. Some of the stuff through Habana, the acquisition-
Jordan Plawner: Mm-hmm.
Daniel Newman: … through Helvetia, through oneAPI. The great thing is there’s been a really big success story with Xeon and acceleration of workloads on the CPU. And it’s really exciting to see Intel doing new things, too, and going to have, ideally, a really exciting roadmap to come in the next a year or two.
Jordan Plawner: Yeah. You mentioned a lot of things that Intel is doing there. I think the biggest surprise for us is that so many people who are starting out on the AI journey just want to use the compute they have right in front of them. So that could be Xeon on premise. On premise meaning at the enterprise. But of course, as you mentioned, many people just going into the cloud. And so once we got all of our software up and running and all the frameworks enabled to use Xeon, we’re just seeing droves of developers and data scientists, and the software developers, ISVs, just starting on Xeon and just making that available to people.
Also, as you alluded to deep learning, or AI, is just part of a pipeline. There’s a whole data analytics pipeline of ingesting data, doing data engineering, doing data labeling. And one of the advantages of doing that on Xeon is you can do the entire pipeline on Xeon. Because accelerators only address a portion of that pipeline, and Xeon, because it’s general purpose, can develop the whole end-to-end pipeline of data preparation, data storage, data training and [inaudible] the whole pipeline on Xeon.
That said, Intel has made tremendous investments in a discrete GPU. You mentioned Ponte Vecchio, that’s the public code name of the discrete GPU. And we also made an acquisition of a company at the end of 2019 called Habana. We absolutely see the need for some companies who are doing AI at massive scale, that they need just more dense compute and more performance per watt. And that’s at the very, very high end. You obviously see a lot of success of other kinds of accelerators there, too. So we look forward to bringing those products to market so we can solve, not only the broad market’s needs with the ubiquitous Xeon running AI on that processor, but we can also address the very top of the pyramid of the people who are doing massive AI at scale.
What I mean massive AI at scale, I mean over hundreds of thousands of servers, 365 days a year, 24/7. As you could imagine, there are very few customers that really need to do that. The broad set of customers need to do some analytics as they gather data and go off and bring that into production, and then intermittently go back and do more analytics as they get more data and want to answer new questions. So we see the demand for AI as being very broad for people who are just trying to get answers to a few questions, using a very simple model. To people, like you said before, who are trying to really break the human intelligence barrier by building AI at the scale of an entire data center and investing hundreds of millions of dollars in that. So like everything else in the IT market, there’s a broad set of needs.
Daniel Newman: Yeah, absolutely. Jordan Plawner, thank you so much for the great conversation. A little insight and education. I guess we went beyond just AI in the cloud, but just with where AI is going and for everyone out there, definitely check out the links in the description below. Or if you’re watching this on Twitter, go ahead and click the follow and stick with us to learn more about what’s going on in the industry with AI. And of course what Intel is doing in the AI space, as well. So thanks Intel. Thanks Jordan. We’ll catch you later.
- NVIDIA Overdelivers Again, and Don’t Expect that to End Soon - June 11, 2021
- This Selloff in Tech is Irrational — Even More so than Last Year’s Climb - June 10, 2021
- Why You Need a Customer Data Platform for First-Party Data - June 1, 2021