Hello, and welcome to the NVIDIA AI Podcast.
I'm your host, Noah Kravitz. No deep learning, machine learning, neural nets, or anything else we call artificial intelligence without computer hardware to make it run.
Today we're talking hardware, workstations specifically.
Our guest is Mike Leach. Mike is Solution Portfolio Manager with Lenovo USA, and he's here to talk about how workstations are being used to train deep neural networks and Lenovo's work with deep learning more generally.
Mike, welcome, and thanks for joining the NVIDIA AI podcast.
Yeah, it's great to be here. So Mike, originally, like all the other podcasts, we'd planned to record in person, but also you were anticipating being on the floor of GTC in person. showing people some of what we're going to talk about.
So I know you're an old hand at talking about the work that you're doing with Lenovo, but we're going to ask you to paint a picture here for folks since they can't be seeing, hearing and touching the workstations in person.
Let's start off kind of broadly. Tell us about how Lenovo's workstations are being used to train deep neural networks And just kind of take that as an on-ramp to talk generally about the role of workstations in AI right now.
Oh, yes. It's disappointing that we can't be here in person.
Obviously, GTC was a big event planned for us.
It's a great stage for us to showcase to the wider audience the full breadth of portfolio that Lenovo have in the AI world.
There isn't a single AI project today that doesn't require industry-leading hardware to kickstart the data path.
And for us, we were showcasing some of our high performance desktop workstations for data science. and some of our rack-mounted and data center class products for larger AI deployments in the Cloud.
And most importantly for us was a new addition to the portfolio where we were bringing AI data science class mobile workstations to market.
So users can really start and kickstart their projects from anywhere, you know, whenever they need to.
So to hop in on that for a second, how big of a deal is that maybe for the less initiated listening to the podcast?
When you talk about a mobile workstation that can do the high performance work and high performance data crunching necessary for today's AI, How long ago would this not have been possible even in a mobile form factor?
Well, hopefully, I think everyone on the podcast should be familiar with ThinkPad, the industry-leading enterprise-class laptop.
But for us, the ThinkPad P series, the mobile workstation class products, they've always been the fastest, the most powerful mobile workstations in the market.
However, the laws of physics have ultimately prevented us from putting huge desktop workstation class performance inside a thin and light sort of mobile market right but over the last few years you know we've changed sort of tact and some of you working with our research and development team um We've created some of the industry's thinnest and lightest mobile workstations.
And the key for us this year and was being the world's first at bringing to market a very high performance 16 gigabyte GPU
It was the NVIDIA Quadro RTX 5000 and actually placing that inside a 15 inch mobile workstation.
We've always had a wider portfolio. thin and light 14-inch, the very big powerhouse kind of 17-inch almost desktop replacement class.
Right, right. they're not the most portable things in the world.
People don't want to lug around a huge big backpack with a huge big power pack.
We did work very hard with our engineering team to shrink that platform down.
We were first to bring to market essentially a 15-inch mobile workstation that has the performance of other 17 inches.
And that, for us, enabled to address a much wider audience that wants to be able to be very agile, still obtain very long battery life, and ultimately deliver the performance and compute that you need So Mike, maybe you can talk a little bit about how these workstations, the P-Series in particular, are being used right now in the field to train deep neural networks and otherwise in specific workflows.
Yeah, and this is where we've seen a big adoption.
Machine learning, we've seen a big increase in the demand, deep learning as well.
What the workstations enable data scientists to do today, they kind of sat in the midst of this problem All companies now are essentially AI-enabled companies.
They employ a data scientist. give him what he needs, they dump this huge amounts, this unstructured data on his desk almost, and then almost demands the intelligence out the other end.
And I think what C-suite level execs don't appreciate is that ultimately it's the speed that you can obtain those predictions, which is important. but also the accuracy of those.
So what we see, you know, the workstation, be it a desktop data science workstation, you know, or a mobile ThinkPad P series, And they've got a huge play in ultimately delivering the ability for that data scientist to be fast and agile.
So moving from sort of on the far left where they've got these huge, these unstructured data lakes, all the sales or marketing or analytical data that they may have.
They have to work through this data preparation and training phase reloop through that numerous times in order to get the predictions and ultimately the deployed AI at the other end.
And it's having that ability to sort of prepare that data and train and learn from it very, very quickly with very, very large datasets. that ultimately the workstation becomes kind of this sandbox type environment to give them those accurate predictions.
There's no point racing to the end the data and all the output you have is not accurate enough.
So working on a laptop, people work off generic, thin and light products.
They'll subscribe to cloud-based instances to get sort of compute power on tap. but it's a very singular-based process.
By using the power of a ThinkPad mobile workstation or a ThinkStation desktop workstation, Data scientists there have got the bandwidth to run multiple training sessions, multiple data preparation tasks in parallel. to ultimately get the most sharpest and most accurate predictions in the shortest possible time.
Have you seen particular areas of industry or research kind of gravitating towards the mobile form factor more than others?
Yes, we have. You know, it's very much I think it's not just industry specific.
I think it's almost then user specific as well.
We do a lot of work in the oil and gas industry.
They've always been very much desk-based and locked to a particular area. office-type environment and uses very heavy datasets, so ultimately demands very powerful desktop workstations.
But the need now to be more agile, to do work, you leave a job site, you've captured data, you almost want to take a small packet of that. download it, use it on your laptop on the flight home before you get back to the office to do the big analysis.
Because ultimately, you can sort of, you know, you can chop off and weed out some of the inaccurate pieces, you know, enabling you to be more active.
And we see a big news one in things like retail and finance. where those datasets are so fast moving that by the time you've got back to your desk to ultimately do the big number crunching, the data is out of date anyway.
People don't want to move backwards, they want to make progressive moves forward. beyond the hardware, which is obviously a huge move forward.
And it makes me think in my own career, just this notion of, I remember when digital video, which is obviously quite different, but still this idea of something that was relegated to these big, super expensive desktop class machines or tower machines.
And then when you could do it on the go, it was remarkable.
And so, you know, obviously that's a huge step forward, but kind of more broadly beyond just the think pads.
Can you talk about how Lenovo is supporting data scientists and researchers who are putting deep learning to work kind of more broadly?
Yeah, so as you pointed out, we're a hardware company.
We design and engineer some of the the industry's leading desktop and laptop workstations, but we've kind of taken a step beyond just the typical hardware.
We're leveraging best-in-class NVIDIA Quadro RTX GPUs in both our mobile and desktop products.
But that's only halfway there. We've baked around at the latest processing technology, memory and storage, But we've taken the next, the final step to deliver a complete solution to prospects and customers where we're delivering a certified Linux operating system. actually loading into there the NVIDIA Rapids AI frameworks.
So something that's baked in with all the GPU accelerated frameworks So customers can really have a true out of the box solution.
Rather than buy a piece of hardware, install it and then go online to github and get all the various pieces and parts so there's so many moving pieces you know from hardware and software and drivers and updates Anybody can buy a laptop and download the appropriate operating system and then play around and GitHub and get all the Redis pieces and parts and frameworks in place. but it's not a very productive use of their time and isn't something they can replicate en masse. the larger enterprises, people who are taking AI seriously and want to ultimately deploy solutions.
We've created a true out of the box platform that can be managed and deployed by IT. just like any other Windows or client-based laptop on their network.
So we've removed those hurdles, ultimately speeding up the time to predictions from an out of the box environment.
You mentioned before that more and more businesses these days are AI enabled businesses.
And do you anticipate going forward that this will just this this notion of an AI ready computer, an AI ready mobile workstation, will this just kind of become the de facto, you know, the default configuration for, you know, enterprise class hardware?
Oh, I see whether customers believe it or like it or not, you know, every customer today is an AI enabled customer.
You know, everyone that use a modern smartphone, you know, there's AI enabled features in everyday life from, you know, the smart lock or, Tesla car you drive to work through to the smartphone you use to listen to this podcast.
And what we're seeing in our industry is that we're embedding more AI-enabled features into all our products to make them more reliable, to make them more optimized for performance, to make them more robust in certain environments.
So yes, very much so. I think we're going to see more baked in features into the product stack and also the software stack. that will make these more AI ready as we start to see AI adopted more broadly across the main industry verticals.
What are some of the challenges and whether internally – you talked about just the laws of physics and being able to figure out how to – get a high-end Nvidia GPU into the system and then how to take, you know, the power of a 17-inch behemoth that not so many people want to carry around and fit that into a 15 inch form factor that's a little more palatable.
So whether from the engineering side of the house, or maybe even from the business adoption side of the house, what were some of the challenges that the teams at Lenovo faced in getting this current generation, the P-Series, out to market.
I mean, the ThinkPad P53 is our 15-inch mobile workstation that can run the NVIDIA Quadro RTX 5000.
The challenges we faced around that product was historically those types of GPUs, the 5000 class GPUs, could only physically fit in the larger 17-inch platforms because of the power and thermal requirements around it.
So using our R&D teams, using some of those intelligent tools, we were able to design an enhanced thermal solution into the 15-inch platform. that say something that was unique in the market and has enabled us to kind of not break the laws of physics, but ultimately adapt our platform to better support the modern GPUs from NVIDIA
And we work closely with them to optimize the GPU for this footprint as well.
But it was very much a joint partnership to make sure that, The market needed something.
People were screaming out for more performance in a slightly smaller footprint.
We were pleased to deliver that. We've done it very well.
Let's turn the lens around for a second, so to speak.
Can you talk about how Lenovo is using deep learning internally to advance some of the work that your company is doing?
I can't talk about everything. Some of the secret sauce, unfortunately, has to stay secret.
Absolutely. We're using data analytics and AI in particular around the global supply chain, optimizing our manufacturing pipelines, making sure that we have the right components in the right place at the right time.
We're using it baked into our own product to deliver a more enhanced solution for the customer.
That might be smart enabled features like Wi-Fi or Bluetooth or webcam so your webcam isn't switched on during certain hours of the day or night?
Sure. and making sure that the system runs optimized for battery efficiency driver downloads or performance enhancements.
There's no short of maybe half a dozen separate features baked into most modern ThinkPads that can be enhanced and obviously we're working hard to make those better using artificial intelligence.
We have a number of platforms obviously related to Lenovo, the Motorola smartphone division, They're using Lenovo workstations and NVIDIA Quadro GPUs to enhance the learning around computer vision for facial recognition on the camera.
There's numerous things we're using with inside our business to not only make ourselves more efficient or more cost-effective to deliver that to the customer, but also delivering better products to the market in light of obviously everything that's going on in the world.
And this is kind of a question born as much from my personal background.
I used to report on the mobile phone industry particularly.
And I'm not asking you to divulge any secret sauce, as you said, but just kind of curious, how far out is a product like the P53 you know, planned and in the works before it hits the market.
Because I think people sometimes have, and maybe not people listening to this podcast in particular, but people sometimes have a skewed idea of thinking that things take less time to actually develop and hit the store shelves than they really do.
Yeah, and I think everybody has a slightly warped view of, you know, what's reality.
You know, people just arrive on the shelves and, oh, there's a new one next year and a new one the year after that.
I mean... Development that goes into every mobile workstation is years in the making.
So the products that are on the shelves today, They are best in class, bleeding edge technology that's taken us months and months and months and months to ultimately design and refresh. and ultimately tweak to make it the best that it can be.
But realistically, we're looking at a design window of probably 18 to 24 months to deliver these types of platforms out, knowing that customers out there want the best.
And technology never stands still. And the industry and the market and the The challenges, the data sets are growing, you know, nothing stands still.
So as a business, you know, we're keen to make sure we keep pushing that envelope and delivering the best that we can do.
So I would imagine that we're recording this in, what, April 2020.
There's probably some cool stuff in the works for summer, fall of 2022 at this point already.
Yes, there is. I can't tell you about any of it.
But yes, we've always got lots of cool stuff in the works.
So looking forward to those next new products when they come out.
I'm speaking with Mike Leach. Mike is Solution Portfolio Manager at Lenovo, and we're talking about hardware, specifically a new breed of mobile workstation that are built with AI in mind, ready to go, capable of doing the data processing, the high-end GPU work, all that good stuff that we've come to expect from desktop class workstations now available in a thin and light form factor. which is good news for everybody listening to this, I'm sure.
Mike, let's switch gears a little bit and talk about your background.
How did you come to be Solution Portfolio Manager at Lenovo?
I think it can be inferred, but what does that mean, solution portfolio manager?
And, you know, what was the path you took to get here?
So, yeah, I've been at Lenovo, you know, for nearly seven years now.
My background before my current role was I was a senior product technologist at My Twitter handle is at MrWorkstationUK.
I was working in EMEA, I was chief technologist for workstation products.
Anything high-end desktop or mobile, It was understanding the key industry verticals that we target.
So knowing for us that AI is very much a horizontal technology that links into each of these key industries. vertical.
So telecommunications, finance, retail, medical and life science, oil and gas, energy, media entertainment, architectural engineering, construction, manufacturing.
They're all individual use cases for us where computer workstations fit.
You might be a CAD designer, you might be an architect, you might be a geophysicist.
And essentially workstation products fit into there very easily.
AI has kind of come on board as a very fast paced horizontal technology that links into each of those.
So my role at Lenovo is to kind of support customers, build out our strategy to make sure that we're fitting those requirements so that the solutions that we bring to market be it mobile workstations, desktop workstations, the enabled software pieces and data sets that appropriately go with those are best suited to customers.
So there wasn't necessarily an easy path, but I'd say for us, coming with many years of experience in the workstation market and understanding how AI is just the latest breed of technology that's come on board.
You know, in listening to you talk about and kind of ticking off and just some of the many industries that use computers, first of all, but now are using and ramping up their interest in use of AI technologies, it makes me think about how we sort of take for granted how flexible the modern computer is.
And that, you know, the software, the hardware is sort of one size fits all in some ways because you can load the software on it and then do what you need to do as an architect or as a an oil industry executive or what have you.
I'm wondering, given how long you've been at Lenovo and your depth of experience working with hardware, How much of a concern or issue or challenge is it as you're developing these roadmaps and then the individual products, trying to make things that will sort of fit every industry, kind of this one computer model fits all kind of thing.
Is that part of the thinking Or are there specific industry requirements that kind of have to take precedence over others?
Or just kind of wondering what goes into kind of planning out a modern computer roadmap.
Yeah, so we don't just have one product.
The great thing is we've got the industry's largest desktop and mobile workstation portfolio.
So we look at the market and what their main requirements are and what we see those requirements are.
But the key thing for Lenovo is we work very closely with our customers.
We have customer advisory councils, multi-customer briefings where we get a lot of feedback from those customers that ultimately then help us define what that next gen or next but next product will ultimately look like.
So if everybody will start to want something that's pink with white spots because that's where everything is going, then we'll take on board those that information and that feedback and ultimately build that into what the next-gen products will be.
For us, there's over 3.5 million different configurations of each product we have.
In the portfolio of desktop and mobile workstations, there is no one-cap fits at all.
And the great thing is with the portfolio of GPUs from NVIDIA, processors, the memory, the hard drives, we can custom build solutions to precise customer requirements.
So they're never paying for things they don't need.
The workstations are flexible enough to meet any type of requirement, be it project or application, data set, or maybe budget, we can build something to suit.
Often on the podcast, either I'll ask or the guest will kind of offer up, something about how the past two to five years in particular, let's say, the acceleration of hardware performance has really made a huge difference or even enabled whatever the guest has been working on relative to deep learning and machine learning and AI.
I'm wondering though, I'm trying to think of a way to kind of flip that question around to you, given your work, and again, without divulging any secrets.
What do you think might be the next big leap on the hardware side that's going to nudge AI in particular? forward and whether that's in the next couple of years, five years, 10 years, however far out.
I mean, technology moves forward. There's a key change every year.
Moore's law enables us to deliver processor clock speeds, more cores, lower power consumption. counteract that by then fueling in more performance and more cores and more GPU power.
I mean, on the horizon, you know, we'll see changes from every major component vendor over the next sort of 12 to 18 months with their next generation refreshes we do every year.
And for us, it's packaging in those solutions so that customers can A, make best use of them.
It's pointless having those features in there if they They don't translate to return on investment or a lower total cost of ownership.
And then making sure, mainly for us, I think the key change is going to be around storage and memory.
At the moment, we can support up to 128 gigabytes of RAM. on a mobile workstation.
If you go back two or three years, that was data center compute cluster type memory.
We can deliver that to you something you can operate on an aeroplane.
So that's a huge amount of data sets. You know, we can support six terabytes of storage now in a mobile workstation to complement that.
So- Which is insane. You mentioned back to the video, the analog to digital video shift.
For us, back to the original questions, The more data you can analyze, the more accurate those predictions can be.
And ultimately, everybody wants... a 100% accurate prediction.
No matter what industry use case, no matter how you'll train that neural network, it's making sure that they say it's accurate accurate as can be in the shortest possible time so i mean we're looking at say memory and storage i think are going to be those key changes for the coming the coming years ahead makes sense So obviously Lenovo is not the only hardware player out there in the mobile desktop computing space, but you've got things that make you unique.
What's something that makes Lenovo's solutions unique in the marketplace and specific to addressing some of the customer challenges that are out there?
Yeah, so we're very good at making hardware.
We believe we make and manufacture, design the best mobile desktop workstations.
We don't just make workstations. We are one of the very few IT vendors out there that have a full desktop to data center portfolio.
So working with Lenovo, we have a data center group, we have the workstation team, we have a commercial IoT and edge AI enabled team. right through to the Motorola kind of smartphone division.
So we have everything from those smart devices, those smart edge pieces, edge enabled GPU accelerated devices, then the mobile workstations, desktop workstations, and then high performance compute clusters, AI enabled data center servers and storage.
And we can connect all of these together.
So Lenovo have a unique piece of software called LICO, L-I-C-O, it stands for the Lenovo Intelligent Compute Orchestration Software.
And it's a cluster management suite of software that will allow people to scale up and scale out their AI infrastructure very quickly and very easily.
So for those out there that maybe invest in a new ThinkPad P53, they have all those AI enabled features their requirements may grow.
So they then might grow and purchase a ThinkStation, a P920 data science workstation.
Two of the very powerful NVIDIA Quadro RTX 8000 GPUs But they might grow again, and they might grow again.
Their products might get bigger and bigger.
And Leeko are nice to scale and create smaller carpet clusters. will then scale up to their non-premise data center class products.
So you can then use this end-to-end investment of hardware, manage it all from that single mobile workstation.
So everything can start on a workstation, everything can stay on a workstation, but then you can tap into more high-performance compute power should you need to, using Leeko in the backend, maybe in your server or storage infrastructure.
That's fantastic. As you said, and you know better than I, There are so many devices out there gathering data, so much data, as you spoke to, thinking about the future of data and RAM capacity going up, that having that kind of baked-in-from-the-beginning solution to orchestrate it all.
And I like the name with orchestration in the Leeko name.
It really makes sense, and I would imagine is something quite useful to a lot of your customers.
Very useful, actually. So for me, within AI, all AI-enabled projects, they will start in a sandbox environment on a workstation.
But obviously people see cloud, various undernames, web services enabled teams out there that can deploy almost on top this ubiquitous, unlimited compute power, but it doesn't come for free.
A lot of customers in certain markets have issues around things like data security, GDPR.
They want to maintain productivity. We've got a customer in the retail space that has some nearly $800,000 into a cloud-based solution of hardware they don't own.
That's just the cost of using somebody else's hardware.
So for us, by utilizing things like our workstation architecture with Leeko and then our data center group partners, Customers can actually own the hardware they use to continue to get productivity and much quicker return on investment from it.
So although that attractive ubiquitous as a service type model works, because you can start from zero, zero dollars down, zero dollars up front. it does cost you a lot of money to ultimately have the very high performance hardware.
So by enabling it on a lease or as a service to procure the product, We do find Leeko works very well with that because it's a much cheaper, a much lower token cost of ownership for the organization.
For folks. who are now have their interest peaked, where can they go to learn more about everything from the p series mobile workstations to lico to all of the other lenovo solutions yeah for more information if you go to lenovo clientai.com and they can see information on all our data science workstations and further products.
Excellent. Well, Mike, hopefully in the not too distant future, we'll be able to do this in person. with the hardware in front of us, we can get the full wow factor.
But you painted a great picture of what's possible now and certainly setting the stage for what's possible with all of these reams of data going forward, even from an airplane on your way back from the client side.
It's exciting stuff. Thanks for taking the time to come and tell us about it.
It's been a pleasure. I'd say I look forward to maybe meeting in person soon.
Absolutely. Thank you. Thank you.
