Decentralized ZettaFlop Uber GPU/XPU Network for Al and Spatial Computing - TOKEN2049 Dubai 2024

okay um thank you welcome all you know the the title of this talk uh reads like a PhD thesis you know Zeta flop scale Uber CPU GPU Network for AI and spatial Computing how many deep tech computer people in the audience no one good so you know my role as Chief scientist for this new initiative which is the first time we are talking about here so you're the group of you here are the first to hear this um so when I heard the word Chief scientist or some people hear the word Chief scientist you may have a picture like this right like you know is some magician that creates uh you know some new thing and an alchemist sort of thing um now that's really not the kind of uh you know science that I do I deal with numbers I deal with computation I deal I deal with transistors designing gpus all my all my life 25 uh uh plus years designed many many gpus on which you know the first uh crypto mining happened uh way back in 2012 2013 time frame um and um when um uh you know so what I wanted to talk about is something very very fundamental okay it won't be a PhD thesis I promise you okay um the fundamental thing about the gpus and fundamental thing about Ai and blockchain and everything is the hint is in the welcome here so can it does anyone know what this number represents welcome to token 45001 0 so that number is 2049 in the floating Point format okay so the core to computation in a computer in a GPU is this number called floating Point format right and uh see 2049 you and I can read it understand it the computer notation we created this notation called floating Point notation and store it in this format why did we create a floating Point format because the integer format we couldn't repres represent right you know the scale of things that we needed to simulate from a size of an atom right like you know a small thing1 10 the^ of you know - 9 M to like this huge you know 4936 to the^ 10 ^ of 13 we couldn't represent with integers so we needed to come up with a new format and that's what the floating Point format is so that from the uh size of uh an atom to the size of the whole solar system we can represent with these numbers right and that's the magic right with this magic what did we do you know this floating Point enabled us to simulate the entire world right from submicron scale to Cosmic scale okay and and this is why our lives depend on Computing right now right so what did we do with this we have our entertainment in the form of games movies you know coming through this floating Point operations education you know if you look at chat GPT as an education mechanism right it is because of this floating Point operations exploration you know you can do simulated space travel you can simulate a robot on a computer and then of course the central to this conference economics right you all used computation to do validation of uh you know various transaction sections so whether it's games digital twins neural Nets and crypto they all depended on this fundamental fundamental thing called a floating Point operation okay that is what we call a flop right a floating Point operation so I'll give you a scale of this operations kind of that you see right so if you they're called teraflops that is 10 ^ of 12 1 and like 120 floating Point operations you find this literally in a billion gpus gaming gpus from Apple AMD Intel Nvidia okay this teraflop scale you know when the first teraflop GPU shipped it was 208 okay 2008 so that's roughly 16 years ago every year there are over 50 to 80 million of this gpus that ship new gpus so up to this point if you just look at the install base of a teraflop GPU you have over um you know billion of these gpus out there now it's really interesting if you multiply billion by uh a trillion which is terlop you get 10 ^ of 21 that's a Zeta flop okay that's a huge number so if we can collectively leverage all these flops that are sitting with the gamers that are sitting in your MacBook that are sitting in your iPhone that you're sitting in your iPad right or an Android tablet you have a Zeta scale Network right and this number is a huge huge Network and I'll I'll tell you why and similarly another important one is a petaflop okay which is 10 10 power 15 this is the scale of compute you see in AI gpus like Nvidia h100 B100 or b200 that they just launched these are petaflop gpus you mainly find them in servers and mainly most of these gpus are sitting in Big Data Centers you know that owned by Microsoft open AI Facebook and all there are a few maybe some of you have some of these gpus as well right that you could utilize and again interestingly a million petaflop gpus is also a Zeta flop right so you know a a a a million of these clustered together can create a Zeta flop Network now open AI the leader in AI said that the amount of compute required for this large AI models is you know it's doubling every 3 to four months okay so you know 5 years ago nobody thought we would be needing jeta scale compute when I said by 2025 everyone needs a zet scale compute people laughed at me I said it in 2020 so what what's happening the super rich are building this Zeta scale data centers it cost one zeta scale Data Center cost you over $70 billion okay and over a gwatt of power okay $70 billion now if you look at Nvidia revenues every quarter they are shipping zets scale compute every two quarters every 6 months they're shipping like $70 billion worth of compute and you know several big companies like I said have the skill today and building more now my dream my dream and I've said this six years ago that I wanted to have Peta flops of compute petabytes of storage less than 10 milliseconds away from every person in the planet right and this dream can only be enabled with a decentral network okay um and with incentives with incentives for everyone who contributes this compute whether it's your gaming GPU whether it's your data center GPU if you contribute this to the whole uh Network that's when we can enable um this zet scale compute for everyone so I'll show you kind of illustr illustration because there are many gaming GP that are that cost only $300 $400 that are 10 to the^ of 14 right that are 100 teraflops and you have some expensive gpus at 10 power 15 you can buy at $10,000 or so so if there is a network of all of this across the world and they are connected with one protocol you get your zets scale Network and that was the uh like our our foundational architecture behind what we call the new network that we are developing called the axle flops Network and again ladies and gentlemen I'm really proud to introduce this new network that we have been working on called Axel flops and uh you know the this is the network to deliver Zeta scale at a global scale right at for everyone okay and what are some key attributes of this network right it is it is heterogeneous realtime GPU scheduling what do I mean by by that that means that you can use gpus from all vendors not just Nvidia you can use an AMD GPU you can use an Intel GPU you can also use other people's AI Hardware okay uh on this network and more importantly our network is not just limited to AI you can do games you can play games on it you can do pixels that gpus are very very good at you can do simulation you can do digital Twins and all so we have this low latency Network that we created and we have in terms of the tokenomics and all which I'm not the expert in but you know we have a white paper coming it's a dual token architecture we have white paper coming very very shortly uh on this network uh detailing all of this stuff but we areen stopping there because we want to get this is a a 10year Initiative for us right to get this network going we are also announcing a deep Tech fund called Axel fund with the goal that we want to get Zeta flop level Computing the new new gpus to less than a billion dollars cost so new transistor technology new packaging technology um new memory technology so our fund will be investing into many of this Tech new container technology and as I heard many talks today security is a big issue right we haven't really compl completely solved you know privacy and security issues so investing into those areas as well is very very important for us and U so again we are taking a com comprehensive view going to the foundation of this computation which is the Flop right an accelerated flop is what created this whole thing right people use buzzword like AI or this or that they come and go but what stayed with us plus is the mathematics right math is the new fire right new electricity and that is what is you know powering our future our past our current and our future and we are building a network based on mathematics which is the floating Point operations per second and uh you know we're super excited and we are also taking a an approach where we're bringing everybody along and also investing into the future uh and into to the future of math with this deep T fund so thank you for giving me this opportunity to present Axel flops