Miss Minehan has served as an advisor for multiple companies on blockchain technologies and semiconductor engineering, and currently sits as Chief Technology Officer for Core Scientific, a US-based corporation with a focus on infrastructure at scale.

The Embedded Future of Monero’s Proof of Work
Kristy-Leigh Minehan, Core Scientific
https://www.youtube.com/watch?v=jDSKIr5EPiU

MoneroTalk w/ Kristy-Leigh Minehan of Core Scientific on lifesaving drones that mine XMR - youtu.be/gSlf0EpHQVQ

Abstract

Proof of work forces specialisation, and the choice of algorithms has caused certain chains to favour certain hardware. Join Kristy-Leigh Minehan, CTO of Core Scientific, as she explains the future of embedded mining - where blockchains will be secured by everything from IoT, to drones, to TV’s and toasters, and how this is uniquely suited to Monero’s future of decentralisation and privacy.

Transcription

Talking never gets easy, even after the hundredth time...Today, the embedded future of Monero’s proof of work.

I’m a big subscriber to chaos theory. Chaos theory, the mathematical definition, that is, is that systems are inherently senstive and a very small change in the starting conditions lead to very large changes over time. Now, the other common definition is more commonly known as the butterfly effect, which is the theory that the smallest change can set off a chain reaction of very large events that can change a system drastically.

The commitment to ASIC resistance by Monero is a brilliant example of chaos theory. Now, I wanna go into why. I want to tell you guys a little story. Like any great story of the twenty first century, it all starts with a tweet. Specifically, this tweet was by a CEO of a company called Astral AR. Astral AR builds drones, of all things, and I was approached over Twitter of all places, with the concept of mining with drones. That was the first tweet. And I stared at it for a few minutes and I kept flinking my eyes and was like “what, really?”. And rather to going into detail on Twitter, the CEO was like “no, no, let’s just get on a call and I’ll tell you, I’ll explain it”. I was like, “oh, OK”. She said, “so we build drones that are designed to help police, public service workers, you know, policemem, firemen, you name it. Specifically, they’re designed to stop bullets.” I was sitting here going “Hmm, uh-huh, uh-huh”. “No really, they have armour that basically disintegrates bullets on impact, and they’re deployed above school doorways, or on the cherries of the squad car”, and I’m thinking “OK, uh-huh, interesting”. “And so when they detect a gun, they’ll basically swarm the shooter and do their best to get in front of the bullet. Now, their job is to keep absorbing bullets and annoy the shooter until a human can disarm them, or until help arrives. They call this the Edna drone”. She goes on to explain that the Edna drone isn’t designed to harm people. Instead, it’s designed to absorb as many of the bullets as possible in guardian angel mode, so that it can end up saving the lives in active shooter zones”. I’m starting to think “this is pretty cool, but where does mining come into play?”. And I’m sitting there thinking “and you wanna mine with these things, did I hear you correctly?”. “Yes, well, they have a bunch of hardware inside of them and if we can figure out a way to monetize them while they’re charging, we’re able to use some of those profits to distribute the hardware, help the network, lots of good stuff. So what do you say, do you want to give it a shot?” I’m sitting here thinking “OK, interesting, interesting, processing, processing.” Then I think “you know what, you had me at mining”.

The first step in this crazy adventure was of course to get the hardware that was actually inside the drones. So these drones run on nVidia’s TX2 Jetson board. Which is actually quite a lot of power packed into a tiny 15 watt board. So inside of it it has a full quad core ARM as well as a Maxwell GPU, so it’s got quite a lot of beef. I spent about three days obsessing over this, trying to get first the miner to compile, “that’s not working”, wondering “OK, why are my libraries not working”, stressing, obsessing over this, I’m starting to let my other duties at work fall off, and I start looking a little bit like this guy. And finally after about three days of not showing up at the office, my boss calls me. And I’m dreading this call, because I really want to finish this project, I really just want it to get it out of the way, prove that it actually works, because I think “hang on, we’re on to something really cool here”. So my boss calls me, and it rings and it rings, and finally I pick up, and before he can even answer, I’m like “OK boss, hear me out, so there’s this project, and there’s gonna be mining with drones, and imagine you could also mine with TVs and fridges and...”. And my boss, you have to understand, is a very, very corporate guy. And I can almost hear him audibly facepalm on the other end of the phone, because he knows where this is going to go. He knows that I’m going to do this, no matter what. He knows that it’s going to be a little bit ridiculous and he’s gonna end up having to support it. And, well, like a good boss, he does.

So, we start working on this project collectively as a team at Core Scientific. And, many months later, we finally end up with a working prototype. So, we tested around an array of hardware to see “does this actually make sense, can you actually, you know, monetize a lot of this embedded hardware and actually do a useful form of proof of work”. So first we started with the Jetson TX2 board which is what the drone is based off of. And there’s enough room in this chip, as I said before, for a rhino to run around in. So at 15 watts total power consumption, it pulls 75 hashes per second on Cryptonight-R, which is around 5 cents per day. Now, that might not seem like a lot, but remember the purpose of this hardware is to be deployed already in school or in squad cars, you’re not actually buying this hardware to mine with. So any kind of cents you’re making, per day, it’s just a bonus. On top of that, on Ethereum, 3.1 MH, at 8 cents a day. Pretty impressive. So we’re like “OK, let’s keep going a bit further, what else can we do?”. So then we try with the Jetson Xavier. Now the Xavier board is actually found on most of the AI hardware today. It’s found in self driving cars, it’s found in some of the newer drones, it’s found in pretty much all of the smart TVs that nVidia does deploy with different companies. It’s actually quite a little bit of a beast. 30 watt total power consumption, it can pull 110 hashes on Cryptonight-R, at 7 cents per day profit. And, surprisingly, 19 MH on Ethereum, at 51 cents per day. Now we’re starting to talk. So then I’m like “OK”, and I talk to my team and I’m like “guys, what if, what if, just hear me out, we try it with the fridge”. Because it has programmable hardware in it, surely it’s working. So we hit up Samsung first, because we do not want to reverse engineer a lot of firmware on the fridge ourselves, and Samsung does not answer our calls for about three weeks straight. Obviously they think we’re a little bit insane. So I put my embedded engineer to work on this. The fridge is down for everyone in the office for a little bit of a while, and the whole office starts thinking the whole tech team is a little bit crazy, and maybe we are.

Finally Samsung answers our calls. “Sorry, what are you guys trying to do?” We explain the whole concept of “hey look, we’re just doing a bit of a research project, but we actually want to see if we can mine with your smart fridges. And potentially, if we can mine with them, you know, we might be able to help you guys out with some of the threats of IOT, and you know, criminals crypto-mining”, and trying to get our feet in the door with Samsung.

They finally decide “OK, we’ll help you guys with this.” So we try this with the fridge in our office, we end up getting 75 H/s on Cryptonight-R, 4 cents/day, it’s not that effective. And Ethereum is pretty poor at 2 cents/day. So that one wasn’t in our favour, but it was fun nonetheless. And finally we have of course three TVs in our office for different conferences. So at this point you just need to start assuming that anything Samsung makes is destined for mining when it comes to Core Scientific. So the Samsung start TV, the QU7000, does a whopping 220 H/s for Cryptonight-R. That’s 14 cents/day profit. And 1.4 MH for Ethereum at 4 cents/day. Now what’s really impressive to me, and where I started to think “hang on, we’re onto something here”, is there are 46 million smart TVs sold in 2018. That’s a potential for almost 6.4 million dollars daily revenue in mining. Hang on, maybe, just maybe we’re onto something when it comes to wealth distribution.

Finally, none of those hash rates were actually optimized. That was us just getting the software up and running. So now as we’ve started to realize that hey, we’ve got something on our hands, we’ve started to go through the optimization rounds. And we’re expecting anywhere from a 20% to a 50% increase in performance. Now you might ask, why on earth are we doing this? Oh, and lest we forget, just in case people think I’m crazy, never forget Bitfury developed a lightbulb that automatically mines Bitcoin when you screw it in. The philosophy now that I’m going by is, you know, if it was an IP address, it’s mining for me.

So, like I said, that’s great and all, but why are we doing this? Why on earth would we go to all this trouble, all this research, into just proving that, yes, if it’s programmable hardware, it can mine. Well, part of it is because proof of work is powerful. See, proof of work has a very unique property of being able to monetize hardware while it’s idle. No other kind of compute marketplace can do that in the world today. And when you can monetize idle hardware, some characteristics can come into play and some unique, I would say synergies are also at play. Specifically some of the cool things that can happen is this will enable creative hardware financing. So going back to the Astral AR scenario, part of what we teamed out to do together was figure out how could we mine with these drones so that they could be given to schools free of charge and while they were idle, be making back their money, slowly. So that schools would not have to put the upfront capital to have this life saving technology.

Proof of work is the only form of compute where revenue is instantly recognized and repeatable. No other form of compute has that. AI doesn’t have that, and in an AI marketplace you do need a buyer and a seller. With proof of work, you just immediately deploy your compute power and you get revenue back for it. That is incredibly poweful. Now it might not seem like it’s very effective, but if hardware can pay back for itself over a period of five to ten years, all of a sudden this enables consumer discounts. So as an example, companies will be glad to give you $10 off your purchase if this made back over the lifetime of the machine. Further, companies like Nintendo, with the Nintendo Switch, would love to be able to do some creative gift codes, or in game discounts if all of a sudden, while the device was charging, the Nintendo Switch was monetized. This is a creative world where corporations are starting to intersect with some of the mining ecosystem.

Now this also enables network stability. Generally, dedicated hardware is not going to stay locked in to a network, it’s going to bounce around to whatever is most profitable. Now embedded hardware is going to stay locked in to one chain. There is usually not enough space on the actual embedded hardware to have multiple different kinds of kernels, and bouncing around to different chains has different energy consumptions. So for instance, if it’s locked into Monero, there’s repeatable energy consumption and predictable energy consumption. Which means people can estimate back what is their ROI and they also don’t have to deal with any issues with quality control, etc, extra heat generated, things that you can’t really account or predict for.

It also enables fairer wealth distribution. As the Core Scientific team started thinking about this, my chief product officer Ganesh, who spent a good portion of his life with Amazon Prime and Twitch started saying “hang on, if we could deploy this technology in India, all of a sudden people could have access to phones. This kind of monetization can enable a lot fairer wealth distribution. When I started thinking, hey, hang on, if a phone could afford could afford to pay off its phone bill, the one billion unconnected could join the connected world right today. And that’s possible.

ASTRAL AR and Core Scientific launched something that we’re calling project Palladium. Project Palladium is an initiative to change how hardware is financed in the world. Now we don’t know if we’re gonna make a big enough impact, but we’re at least trying. So project Palladium is only possible on chains like Monero. Specifically, it’s only possible on chains that decide to target consumer grade hardware. No manufacturer is going to put an ASIC chip in their consumer devices, it’s never going to happen. There’s all sorts of security reasons why that won’t happen, legal reasons why it won’t happen, and quite frankly it’s just a nuisance. But with the consumer grade chain, people will jump into that. Companies are more... I would say, often incentivized or eager to experiment in these kinds of different wealth distributions. It’s also a step towards useful proof of work. So one of the biggest complaints that I get in my job is people always come to me and say hey, Bitcoin mining is incredibly wasteful. Monero mining is incredibly wasteful, XYZ mining is incredibly wasteful”. Well, it’s only wasteful if you don’t find the monetization model of proof of work useful. Now, it’s already being used in the ecosystem but so far only ASIC manufacturers have utilized it in this manner, specifically ASIC manufacturers mine with the hardware before they sell it to consumers. We don’t actually want that, what would be way better is if all of a sudden the hardware is much cheaper and was deployed and was it was paying back over a period of X amount of years. And the other important thing to understand about project Palladium, it’s not about breakeven and it’s not about ROI. So in contrast to traditional mining where you’re trying to get back your sunk cost as soon as possible, this consumer grade hardware is already brought for other purposes and has other monetization strategies. So project Palladium is a way to extend monetization and provide sustained security to a different chain.

Project Palladium is actually run by two other projects that we’re pretty proud of, now we’re giving you guys a sneak peak of here, we haven’t shown this to anyone and we’ve been in stealth for about eight months at Core Scientific. So first is Minder OS. Minder OS is our embedded operating system, that kind of powers our entire data centres. And what Minder OS does for project Palladium is it actually allows the tokenization of hardware. That is, tiny machines’ unique identifier and other metrics of the machine to a digital token for identification and access control deployed on the Monero blockchain. Now each token has its own pair of access keys that allow the owner of the device to modify and manage the device. It also has a custom APIs so embedded hardware running Minder OS can only receive commands that are signed by a private key associated with the digital token, and, you know, subsequently the machine’s unique identifier. And also has inherent optimization built in. So Minder OS automatically adjusts clocks, memory timings and other hardware characteristics to suit the algorithm it is working with. And version 2, which we’re scheduling for public release in 2020, has its own compiler specifically for cryptocurrency mining. It’s also powered by Minder. So, Minder is what we call our intelligent infrastructure. Think of Minder like what lives in the cloud and Minder OS like what lives on the machine. Minder is designed and built to prevent 51% attacks. Specifically, it monitors for hash rate dips in the network, and it alerts human workers. It has a mode called Lancelot mode where minder will automatically fill lost hash rate with hardware that it’s currently managing. So if it sees a massive dip in Monero hash rate, because a bunch of GPUs have gone offline, it will immediately deploy hash rate under its control to fill that gap. This helps keep a sustained network going forward. It also has automatic conversion in place so most of the participants that we’ve signed up to project Palladium cannot hold cryptocurrency for various legal reasons or whatever. So instead, through smart contracts on the Ethereum blockchain, mining rewards are instantly converted to a handful of fiat currencies. So this is a really great way to get people sort of into cryptocurrency without forcing them. Because then, when you do it like this, when you start converting mining rewards into fiat, they start getting really curious. We’ve often found they start reaching out to us and say “hey, you know, I’d actually like to convert a bit of my payout into Monero instead, could you do a 70/30 percent split?”. Then you get them hooked.

And just for some amusement, we did try mining with a few other things, so I do not recommend mining with these three things: one is your pacemaker. We did research into that, it doesn’t have WiFi or bluetooth, and actually being actually gonna damage your heart muscles, so do not do that - that was an interesting conversation with the hospital. Speakers, terrible latency and continuously interrupts the sound quality, we had some pretty beefy bluetooth speakers, they didn’t produce anything, it’s terrible. And also your Roomba. So the Roomba is the thing I really want to tackle. For some reason its processor is actually at 90% utilization while it’s charging. I haven’t cracked that case but I wanna figure out how do we mine with Roombas.

So final conclusions is that project Palladium is only made possible because of Monero’s commitment to decentralised and fair mining. I want that to sink in, because what Core Scientific and Astral AR are going to do is ensure one of every single one of these drones are in schools across America. And that is seriously only possible because Monero exists today, because Monero is doing research to proof of work algorithms that are for consumer grade hardware. No other chain can support this, no other chain has that dedication. So when the media asks you why you’re fighting ASICs, you can tell them it’s for the drones.