Intel® Liftoff for Startups - Accelerating Innovation Through Partnerships

Tony [00:00:04] Welcome to Code Together, a podcast for developers by developers, where we discuss technology and trends in industry.  

Tony [00:00:11] I'm your host Tony Mongkolsmai. 

Tony [00:00:17] Innovation is at the heart of what drives technology companies. In an effort to foster and grow technology solutions, Intel has created the Intel(R) Liftoff for Startups program. Today, I'm joined by the co-technical leads for the Intel Liftoff for Startups program, Ryan Metz and Rahul Nair. Welcome to the podcast, Ryan.  

Ryan [00:00:34] Thank you so much. I'm really excited to be here.  

Tony [00:00:36] And thank you for doing this Rahul.  

Rahul [00:00:39] Thank you very much, Tony. It's it's a pleasure to be here.  

Tony [00:00:43] So let's start off with what the brand new Intel Liftoff for Startups program is. So, Ryan, give us a little background about what is Intel Liftoff. 

Ryan [00:00:53] Intel Liftoff is a free, acceleration program for startups. It's membership based, not cohort based. And we're not asking for equity or anything like that. Before I get into kind of what we do, I should explain why we exist. And that reason is oneAPI and heterogeneous computing. So there are, there are a ton of of of new computing architectures kind of flooding into the market right now. There's you know, everybody is bringing in GPUs and DPUs and IPUs and wafer scale processors and all these different ASICs for deep learning. It is a very exciting time for computer architecture work. And Intel is a part of that, you know, we're branching out into heterogeneous computing ourselves. We have a worry though, and that worry is a fragmentation of the developer ecosystem into these little into these little wells of only being able to run on one of these architectures. And that would really go back to the bad old days of like minicomputers and mainframes. That's what we're trying to avoid and that's computing used to be there. And primarily because of Intel's efforts, especially in like the nineties and early aughts. Now you can expect interoperability everywhere. So with this new explosion of architectures, we were worried about a return to that world and that paradigm. So we created oneAPI and the idea of oneAPI is a collection of tools and toolkits for developing software for doing parallel computing with acceleration. And the difference between this and some of its competitors is that this is the open ecosystem for doing that kind of thing. It will run on all of our hardware, our our CPUs and our GPUs, etc., but it will also run on hardware from our competitors certified for ARM; RISC-V support is coming along and more. So the idea here is you will write once and run anywhere and you will not be limited by what a vendor supplies. If there's some way you can build technology to create value on some kind of computing platform, you should be able to. And with oneAPI you can.  

[00:02:50] So a lot of older and larger institutions when they first saw this, were immediately on board because they remember, they have the institutional memory of the bad old days of like my application runs on DEC or runs on IBM and that's it. And they don't want to go back to that world. So as soon as we announced this, we got great penetration there in the start of developer ecosystem. These companies don't have an institutional memory that can remember those days anyway. And also they're much more focused on getting to market, getting the product market fit, getting their first dollar, getting their first customer, you know, solving problems immediately and they don't have these sort of longer strategic worries. So uptake there was always going to be more of a challenge. And that is why Intel Liftoff was created. Intel Liftoff is a program that you can join for free and we will help you adopt these technologies so that you can bring your innovation to wherever you see fit. Any platform, not just Intel platforms, although, you know, it will run very well on Intel platforms. We do a lot of engineering work and there's there's a lot of training. There are hackathons. We do consulting, including like one on one calls for like advice on how to do deployments, how to tune workloads all the way up to joint engineering projects where an Intel engineer will even pair with you and work on work on your day to day problems getting things going.  

Ryan [00:04:05] The second benefit, of course, is getting access to all of our hardware and our software, really everything we bring to market. And so once we have a project like that done, once you've adopted the tech and you've built something cool with it, we want to show that off. You want to show that off. Co-Marketing! It's time, right? We'll get some you know, we'll get together a blog or a white paper or you can go on Tony's podcast maybe. We've put people up in front of conferences, including some pretty big ones, to show off what they've done because we want the world to know what is capable, you know, on this stack.  

Rahul [00:04:38] One of the important things is a programing model. So once you're creating a product, right, like Ryan mentioned, the first thing is to create something and get the product market fit. Eventually you'd be in a place where you have multiple source code for multiple architectures and there is a certain amount of technical debt that would be there to optimize. To maintain this code, you'll need more engineers and you know, many other things. One of the things we try to avoid is particularly that and to have a single source code that you can compile and target to different devices, it could be an Intel GPU, an Intel CPU, it could be one of our competitors in an NVIDIA GPU, an AMG GPU or even ARM in some cases. So you avoid that technical debt of maintaining multiple different types of source code. And I know that will go into the benefits of the programing in detail. But one of the key benefits is also that we do bring our startups and showcase them in many events specifically where there are customers. So they can find their market, and directly interface through customers with our help where Intel is sort of telling, you know, this company has created this great product and it's accelerated; maybe you need to talk to them, have a discussion. And we've done that multiple times with several of our partners, but I'll stop by the benefits here and we can go into detail it all.  

Tony [00:06:04] I mean, let's we can even go into the benefits now. So if somebody comes to you with a problem, they hopefully already have some level of source code. I guess one of the requirements to be in the Liftoff program is they need to be a company of a certain size, right? Can you outline that really quick?  

Rahul [00:06:19] Right. Yeah. So thank you, Tony, for that. Like the specific startup that we look at our early states and up to series B startups and we don't have a particular condition that you should be making revenue right now, but at least the minimum thing is that you need to have some sort of product. It could be pre-alpha, you know, running on your laptop, doesn't matter. If you have a product and if a core team of engineers and a core plan on making it work and scale, that's where we come in. The benefits specifically from the engineering side are you get access to hardware much before even some cases the hardware is released. For example, I'll take the latest Intel Xeon processors that you can use on AWS or Google today, the fourth generation Xeon. Our partners got access to those machines through our program at least six months before it reached general availability on CSPs. So the advantage there is if you have a product, you can make sure from day zero when it is available on cloud and hyperscalers, you can make sure your product works and is highly performant on that architecture and our engineers and our team help you to make sure that it works as expected on that hardware. Beyond engineering help, there are a lot of technical sessions exclusively designed for our startup members where you can participate. There are things like hackathons and workshop. Also, once we do a project together with you, those the marketing drive. We have a separate division that does marketing. We interact with internal Intel folks from the sales group, the marketing group, and also many other different divisions to promote your story through all Intel channels. You know, we stand behind you and we want your story, the awesome product you have made, is showcased to as much customers, as much people as possible. Also, when you're at a stage, maybe a stage, series A and beyond, if you want to have a meeting with Intel Capital, that is our venture capital wing. We do arrange those meetings for you. And I need a caveat that our two teams are different. So we do arrange the meetings and make sure that you can talk to an Intel capital folks, but the decisions and things whether to invest or what is advice is all based on Intel capital priorities. We do have engagements with the universities and different accelerators. So if you want to find like-minded startups or you want to talk to different VCs outside, we do also help you with that. These are the main benefits that I can think of. Did I missed anything, obvious Ryan? 

Ryan [00:08:51] And I guess the only other thing I would want to add is that just to go back to hardware access and how all this happens, like how the computers is made available and why we're able to grant access to next gen chips before they hit the market. It's this incredible partnership between us and the Intel Developer Cloud. They've been a great internal partner to us and getting access to that is is an exciting thing. Like that is going to be in the future be a production cloud where things can be put into production and drive revenue. And and in the run up to that happening, like for now, it's been an incredible place for companies to get access. You know, you want next gen hardware, you want it before it hits the market, you want to get ready for it and do the things you can do with it before your competitors do? You come to us.  

Rahul [00:09:35] When you said that, like I still remember it, my first in a going back to history a little bit, my first computer had a Pentium 4 processor and you know, there's always this fun and this excitement to work on this basically because of my history with computers and we get to work with the coolest hardware possible much before anyone has hands on. And the awesome workloads that some of those startups are building, you know, being in the AI or HPC space, to see it first run and optimize it and and going through that, that pipeline it's...I would say it's one of the it's more than a job, I still feel I'm in a lab hacking these things and I can do it for free. That's you know that's one of the best things that I could do in my job. And the good thing is that what we do, it just doesn't remain in the lab. At least one of our partners can take that at scale. So that is a benefit for the startup also.  

Ryan [00:10:32] Specifically like just talking about getting access to the next gen, how exciting it is. I will never forget the first time I saw a development team get their hands on like a four way GPU system where each GPU had 128 gigs of VRAM on it, so a total of half a terabyte in a four way box and started to do a machine vision training job on 8K images with GPU acceleration, which used to not, the batch size on a lesser GPU is too small to do it effectively, or it had been for them. They'd never been able to do it before and so seeing that happen for the first time ever for them gives you that feeling of excitement and discovery you can't really get anywhere else.  

Rahul [00:11:11] Oh yeah. And even the CPU side, right? Like you can get a bare metal CPU, with Intel 4th Gen Xeon with 224 cores and half a terabyte of RAM, so that is an incredible machine. And going back to that 8K resolution images, some machine learning workloads you are okay with doing it as a batch job, but your main thing is that the input data size is so large that you cannot move it to GPU. So those sort of special use cases you get a machine with have half a terabyte of RAM and these many cores; it's beautiful to see it run. And specifically with our AMX matrix extensions on the CPUs, if you're doing mixed precision BFloat-16 training it's impressive. It's beautiful to see it running.  

Tony [00:12:06] Yeah. So while we're talking about next gen hardware and developer cloud, obviously you guys are working with a bunch of startups. We probably should mention, like you said, there's the Intel Data Center Max GPUs there, codenamed Ponte Vecchio. We have the Habana Gaudi 2 in the developer cloud right now, so that's actually pretty cool. And like you said, I'm sure all of the all of the developers in the world are really excited about that kind of stuff. Can you give us a rundown of what types of problems the startups are tackling right now?  

Ryan [00:12:33] Oh, man. We've got anything from people using drones to, like, fly around autonomously and look at windmills and inspect them from damage to health tech. You know, things about doing an AI for analyzing X-rays or CAT scans, you know, looking at strokes to people trying to use like alarms to do prediction. Couple of generative startups doing auto automated like gamification. On the non AI side, we've had some very cool computational fluid dynamics on the cloud, approximations of the algorithms that will run someday on quantum computers when those are ready being able to approximate those algorithms on GPUs at scale so that you can get most of the benefit of running them. All kinds of cool stuff.  

Rahul [00:13:16] We have companies that are doing battery recycling using AI, all the way to quantum simulations, and in between there are generative AIs using, you know, 3D reconstructions in real time. Companies that create virtual avatars using LLMs for creating text. The great thing is because because our program is global, we do get folks who are building for different geos that is particularly needed in that geo. We have companies all the way from working with us, from India, China, the African continent, Nigeria, a lot of companies from Europe, few from the U.S. So it's fascinating to see different geographies and these innovators and like this incredible people building for their market, you know, these different types of products. And it's great to have that connection. And I feel that we have talked to maybe 300 or 400 startups over the last 1 year, and it's been it's been really interesting. Each time we go into a call with a startup, it's always interesting and there's something always new, you know, something that we have never thought of that someone came up with. So it's fascinating to be in those calls.  

Tony [00:14:28] Can you talk a little bit about the different types of technology? Are you helping them enable technology or are you just giving them advice? I think I remember something about you guys helping people like leverage AVX-512 extensions. But can you can you talk a little about kind of the detail of the actual engineering support you guys provide?  

Rahul [00:14:46] We do have a engineering team that is expertised in all stacks of software. So if there is AVX-512 SIMD that you're working on and you need us to help you with optimizations on that, we have engineers working a couple of startups on that. But if you want to scale your application on on a hyperscaler and you want our help with how to architect the right way, we do help with that. And if you're a AI startup and want to figure out what is the best GPU for this token length input for this kind of model to create a text data out. Yeah, that's also something that we help. There are a few startups where we help them to reduce their cost when they're deploying this model for inference. There are things were startups want performance and they don't really care about the cost, there we do help with that. And from day to day engineering work, basically how it goes is me and Ryan will be having calls if it's startup comes in and they need to do some generative AI, for example, let's take that. We'll have a couple of discussion with them on the architecture and we'll assign one or two folks from our engineering team to have at least two pair programing sessions with them per week, where engineers from our side, would be on a call with them and working with them to make sure that what they wanted to achieve. It could be either creating an extension to their existing application or optimizing something that they have already. They help with that. In these cases, mostly we want or we prefer one of the engineers from the startup take ownership of the project because we don't want them to be in the same position when they're doing a new project. We want them to understand what exactly, what kind of optimizations we are doing and essentially control the IP and control the ownership. And in no way we want to dictate what the architecture should be. So I was it's mostly a consulting and a helping hand rather than taking ownership of the the engineering solution.  

Ryan [00:17:26] Yeah. I mean, they need to be able to go forward. They need to know every inch of their code base. Right. We don't want to deliver them, like, completed work. But I will say, like on the pair program. So we are happy to do that. We do that with a lot a lot of our partners. That's the very highest end of of that's the highest touch, if you will, that we have. There's a lot of levels below that as well, where it's like sometimes it's like, okay, you've got it. There's, there's a new feature, you want to build, a new capability you want to add. You can hop on a call with us, show us your plan. You know what I mean? Like your architectural diagram that you got like and we'll you know, kick the tires and tell you what we think. And there's their different approaches. It's like, okay, don't use that. Use this. Or like, let's make this a little bit more modular, you know, or oh, there's a better model. And it might just be like, Oh, well, we were thinking we were going to use like a ResNet for this, but we're not getting good results. It's like, okay, well, have you tried anything more modern? Here's some of your here's some like, here's what SODA like the cutting edge of open source in terms of vision models you can try and this is and this is how you can deploy them. This is how you can get good speed. So there's a lot of consultative calls and advising that happens to. And that goes beyond purely engineering stuff, I mean, that's our strong suit and that's what the the basis of our partnership is in engineering partnership with our members. But we do go beyond that because of the size of our of our engineering team. There's a very broad set of different experiences and expertise, just knowledge within the team that we can use to help you do that. Intel Liftoff is a new program and so even though, you know, we're at launch point and we're very proud of what we've built, we're still growing and still trying to figure out things to add. So the other thing I'd say is, if we have a new member who's like, "Well, can you help us with this?" And it's a good idea and we can, we're going to say yes. One of the benefits of getting in with a program when it's young and new is that, you know, we're a little bit more flexible and open to new ideas than something that's been around for ten years, you know.  

Tony [00:19:33] So yeah, it's a new program. I mean, but you guys have been kind of doing this work for a while. So just to give people an idea, I think at the end of the year you had a goal of getting how many start ups into the program 

Ryan [00:19:45] End this year it's going to be...end of 2023, the goal is 150.  

Tony [00:19:49] 150. Okay. So there's a lot of opportunity there. It's not like we're saying ten.  

Rahul [00:19:55] Yeah, we are. We are open to having discussions. But, you know, if you have a cool idea and if you have a plan to get it realized, the best I can promise, the worst I can promise is that you'll get a face to face time with us, you know, both of us, and we'll have a discussion with you. And how we see startups and working with folks working in this project, right, it's more than the startup they are working on it. We are always focused on people. So if we see that, you know, maybe it might be the seventh startup or the eighth startup, that that is becoming the next, you know, greatest product or greatest company ever. Always, what we have felt is that if the person of the people working behind that idea is passionate and is committed, we're really happy to work with them and help them in any way we can. And, you know, if you just want to have a discussion, have a chat and you don't have any particular technical queries or technical notes, that's fine and we can we can have a call with you and have a discussion with you. We have around 50 to 60 startups already partnering with us, so we are happy to add more. 150 is our is our limit, but they are hoping to add much more startups than that.  

Ryan [00:21:22] I mean, I kind of want to blow 150 out of the water, but that's just me. And we'll see what happens with some of our infrastructure availability, but we may be able to go higher, I would very much like to go higher than that this year. We'll see what happens.  

Rahul [00:21:38] 300 is a good number. 300 is the right number. 

Ryan [00:21:42] Right.  

Rahul [00:21:42] Yeah.  

Ryan [00:21:44] It's not a competition. Should it be?  

Rahul [00:21:47] Yeah and we don't, Yeah, that's good. And we don't want partners just for the sake of numbers, right? We don't want to say our program has 10,000 members. 20,000. But that's not a real aim. And both of us are hardware geeks. Like we really love the hardware. We like the things you can do with software on it and we want the right kind of partners and right kind of people who are committed, passionate and has an idea and want our support to make it, you know, to get into the final stages or realize their ideas. And we have patience and we do wait for the right startups to come in. And, you know, the number 150, 200, that's that's not important.  

Ryan [00:22:36] Yeah, Yeah. I mean, I would say we are looking for builders like we've and you see this in startups sometimes, there's people with maybe a maybe a really great idea, but like they don't have that, don't have that builder drive to get things done where it's like, you know, you'll try to, you'll try to put things together for them, but they just they're, they're not they're not able to get to get started and to go from 0 to 1. And we're looking for people who have that drive and like, you know, the grit and determination and at least on their team, the chops, the technical chops to really build something and really make something new, to really like, you know, bring something new to the world that's really going to make a huge impact. That's that's our goal, is that we want industry and world changing startups to come and build, to build with us, to build with Intel.  

Tony [00:23:28] You know, what's funny is we wanted to have this podcast to kind of talk about the benefits of the Intel Liftoff program. And you guys are talking about the benefits that we can provide to startups. And as I'm listening to you guys talk, because obviously I'm a builder, too, and I just want to build cool things. It actually almost feels to me like we're having a riff about trying to get people to come work for us. Right. It was actually trying to get people to come work with us, but it's almost like we said, Hey, come work for our team. You're going to work on all of these startups and do cool stuff.  

Rahul [00:24:01] Yeah.  

Ryan [00:24:02] No, no, This is a this is a riff on how to live a good life. Build something, do something that matters, have a purpose, something.  

Rahul [00:24:12] That you get free hardware.  

Tony [00:24:15] Yeah. I mean, that would maybe that would have been a good tagline. Something about Intel Liftoff, build something with purpose. That's actually a good one, too.  

Rahul [00:24:23] I don't think that. I don't think that will go beyond the legal review.  

Tony [00:24:26] So, you know, let's try this podcast still probably has to get legal reviewed. So we'll see what happens.  

Rahul [00:24:34] All right.  

Tony [00:24:36] Is there anything else you guys think is worth calling out about the Intel Liftoff program or something that you'd like to say about the program?  

Rahul [00:24:44] Yeah. The one thing I want to talk about is, you know, the whole purpose of a partnership program, right? Intel has several partnership programs, and this one Intel Liftoff for Startups is specifically for early stage startups. And there is Intel Ignite program for startups. There are many other partnership driven programs also. All these programs exist so that we can act like a conduit to Intel. You know, there are different teams working inside Intel, creating awesome stuff, you know, in the AI space, HPC space, hardware. So the results that we get and the way we are able to help is sort of the collective Intel mind and the opportunity that a startup gets is that they can, you know, talk to the team who created, for example, the Intel Extensions for PyTorch, and they want to ask us a specific question to them and we facilitate that. So I consider us as ambassadors here, and there's like 40,000, 50,000 people behind us helping us to do this, what we are doing. So I just wanted to just shout out to the the hundreds of engineers I have pinged to help that, you know, different things when we are helping our startups.  

Tony [00:26:17] All right, guys. So we're getting close to the end of our time here. So I'll do what I always do, which is ask you guys, where do you guys hope technology goes in the next couple of years? And what's what's really getting you excited?  

Rahul [00:26:28] So to preface, I started my programing career, you know, maybe 20 years not even career, like when I was kid. I got a couple of floppy disks with Linux on it and I learned a lot from it. The beauty was that I could change anything. I can see what the source code is and break my system multiple times and reinstall things. That sort of openness and software is what really excited me. And, you know, in even if you look at all the popular software, all of them have been open, there's open contributions and open building of this projects. That is incredible. If you see from a software site, I can see that with, you know, things like RISC-V from hardware level, also that openness is finally coming out. My wish would be a complete open stack from top level to all the way down to the hardware. And, you know, I can tweak anything. I can create interesting hardware, software, it doesn't matter anymore. You know, things are not locked down. At least that's my wish list. If you could ask me, You know what? What would be what? What is the greatest thing I would like to see? Because freedom to build is very critical.  

Tony [00:28:07] How about you, Ryan?  

Ryan [00:28:08] Wow. So moving. I won't go quite all the way back to, like, my first experiences for computing. But I will say that, I ran my first, what we used to call a multilayer perceptron in 2003. And back then, you couldn't really do anything with it. So we didn't. And I was working as a computational nanoscience back then, scientist back then, or training to be one, and you couldn't really do much with them back then. So we moved on to doing more like HPC style simulations of  different brain networks. And it was so fun to see, you know, little over decade layer later, like the ImageNet movement, you know, when, when, when convolutional neural networks hit the scene, and everybody became instantly aware that this was going to be a big deal, that this like long known mathematical model that had nobody like, you know, had for years had any faith in. I hadn't, you know, I that's not what I spent my the last ten years working on was going to be a big deal. And now what we see with this, with the release of GPT, of ChatGPT, is that everybody knows now. It's not just not technical people, everybody knows that this technology is going to fundamentally change: industry, society, life. It's going to be transformative and we're in this incredible, exponential moment, where it's like there's a new important paper that's out every other day in the field. It's hard to keep up and it's so exciting. And my wish is to see these capabilities very broadly distributed, not limited to just a few, you know, like to see to, to unleash like human ingenuity and creativity and passion, to see them to see them empowered by these technologies. There are worries about about things that could happen with these things. We're not going to get into it, but everybody knows what people are saying, how terrible this might be. I am an optimist and I hope to see more people get to use AI tech and computing and accelerated computing, not less. And to see what happens and to see what people build and to be a part of it myself. So I'm hoping that, you know, this tech spreads into lots of different people for human flourishing instead of getting locked down and very limited, you know, based on who like, you know, which a select few are allowed to use it, that would be that would be my wish.  

Rahul [00:30:54] There are. On a closing note, I would say there are hundreds of thousands of real problems that can finally be solved with AI or elements of AI. It's important to solve them Three or four or even ten companies cannot really solve this. We need a global audience. All the innovators across the world to come together, get access to this to solve it. I really feel that maybe in you know, it's been about in a year that AI has finally transitioned into being a commodity. I always compare AI right now the tools, different tools and things in AI to for example how the sort function is right now in our programing language. You would study how a sort is implemented maybe in your undergrad, but you don't care how it's implemented. You use sort function in Python on C, whatever programing language you use and AI today is like that. That opens up the space incredibly. You don't need to be an expert on creating a multilayer perceptron. You can just call a HuggingFace embedding API and just create some sort of a service out of that that opens up so much of possibilities. And I'm really hopeful about this, this revolution of AI.  

Tony [00:32:23] Yeah, that's a good one. That's a good closing note. So thanks, Ryan and Rahul for joining me.  

Ryan [00:32:30] Thank you. Thank you for having us.  

Rahul [00:32:33] Thank you so much for doing this.  

Rahul [00:32:36] And if you guys are interested in the Intel Liftoff for Startups program, we'll link the website on our page here. And also in the future, we'll have more podcasts from the Intel Liftoff for Startups Partners. So the various different startups will come on. We've already had a couple of them on at the end of last year, but we'll have more throughout the year. I'm hoping to have at least one a month. And if you guys are interested in partnering and working with Ryan and Rahul, I would recommend that you check out the website and see if you can sign up. Thanks for listening.