One question around that when I am. Maybe I'm thinking about this the wrong way, but when you think about proof of work, and you think about okay, well, people Engineers came together and said, in order to do more of this more efficiently. We need to build custom built processors A6, things like that, right? That's the path that I went down to, to. The mining side. When you think about the performance side of Solana, you're using equipment that the engineers at all, the companies you named are going out in manufacturing and bringing to Market not specifically or purpose-built for Solana, but just because they're improving the hardware. Do you see a world where Hardware manufacturers say? Hey, we can actually do a few tweaks or, you know, we can improve performance in this way or that way that will actually benefit salon. - oh, there's there's custom Hardware design for Solana. Was that just crazy thinking? No,
like we're a bunch of X Qualcomm guys. Are always itching. Like, what, where's our chance to go for their own system on a chip? Like what can we do it? So, the way that's done and like, even a Qualcomm is, let's say you have like a new augmented reality or something like that Loop that you really excited about because it tracks eyesight much better than anything. You try to take that software. And optimize it for the current system of the chip which has dsps and has gpus and it's a lot of work to form to restructure code that so runs efficiently there. But doing that work forces you to really kind of chop away, all the unnecessary bits of the of that algorithm and get to a level where it's almost structured to the point. That okay. Do we need that extra 5x performance Improvement for that part? If we do, then we ship it as an Asic and making that decision at something like welcome at that. This is going to be a use case that out of the, you know, 4 billion. People that have cell phones, that hundreds of millions of them were get I want to use, right. So, this is why it's so important. It's like, the next-gen use case is going to be like, people will buy a phone because it has that thing, right?
And understood. This is why Apple does their own custom chips. It's like, yeah, they know that they can optimize things for their specific. Vic hardware and they're used to specific use cases, right?
Exactly. And like, a lot of the stuff like everywhere else in Sac is hyped driven, and I'm kind of like, goes by the wayside, like, do we need neural network accelerators and mobile phones, right? It sounds exciting. Like, we're going to have a I right hand, but maybe they use cases in there. Right? Like maybe there isn't actually an application that needs it. It's
great for marketing though, man. Like they tell you more artificial intelligence and you know, there's there's always got to be a slight upgrade. To get you to buy new hardware
and and as a hardware company, you're making this bad, like almost like a 5-10, your bat. This is where the industry goes. And it's a big commitment because it is hundreds of millions of dollars worth of engineering effort and it takes space on the chip. You can always use that space for L2 cache and just make the chip better. But so those are you know, those are trade-offs that companies live a diet by for us it's easier because we can build a card that plugs into CIA bars that maybe accelerates like, you know, pools that work packets fetches state from a bunch of ssds and then has like a bajillion optimized Digital Signal dsps or something like that, that can execute smart contracts. Maybe that's something that we'll get to, but we're not there yet. And like the advantage for us is that thing could be maybe like that card itself could be 5,000 bucks, you made, it doesn't need like a hundred million units of it. Right? Like it only needs, maybe 10,000 or 20,000, but it makes it Harder for us to deploy these systems in like financial institutions which are the true users of this, right? If this is a network that connects the world of Finance, then Finance Circle, coinbase. They don't run dedicated Hardware. They're on their stuff in the cloud. They want to run it on generic systems. So how do we make sure that they can do that? But also have the fastest possible, not work, right? These are like,
yeah, that's tricky, right? Because the some say, They want to build a trust, the hardware. You want it to be almost like commodity mass-produced so that anyone could just pick it up and you know, there's with anything Finance are there there now auditing these motherboards at the chip level right to make sure there's no malicious code or anything. Actually tucked into the chip fabrication process, which is just insane. So I can see that would introduce a bunch of new additional complexity that when you're not running into any limitations today, it's not like you need to optimize the Get an extra, you know, 50% performance. Like you don't you're not having those problems, right? But in a world
where this is the fabric of Finance, you could see like us defining this. Here's the, here's the design for a risc processor that implements Alana and then Nvidia Cisco cetera, right? They see that every business that dude does Finance needs one of these six. Maybe at that hundred thousand units, you know, then they're like, okay, let's go. Let's go build it. Like, maybe we that happens, but maybe that's 10 years from now, you know.
Yeah, I mean it's happening at the proof-of-work level. Right? I mean, Invidia has dedicated cards now that are just done for GPU mining. That is purpose-built for that activity. So it's not, it's not crazy to imagine that. What happened. What's,
what I think is likely to happen? Because our art problems overlap. So much with traditional databases. Is that the problems of Oracle or whoever? This building of XJ of databases that are deployed. A lot of times in the cloud. They face the same IO bottlenecks the same computational bottlenecks is us a lot of the engineering effort is already spent and making those better and we get to leverage that. So that, that is kind of like part of the benefit is that will likely see dedicated like cheap you to networking layer, bypass write something like that. That's just a ship a ship from Nvidia get deployed of like Google cloud and the sure and something about them. Maybe like a dedicated Solana,
right? Make sense.
Will Hardware Eventually Be Custom Designed for Solana?
The Solana Blockchain: Ethereum Killer or a Piece of a Larger Puzzle - Anatoly Yakovenko