News
More from NVIDIA GTC 2024

Zeus Kerravala

Founder & Principal Analyst

ZK Research

Three insights you might have missed from the Nvidia GTC event

In recent months, the financial performance of Nvidia Corp. stock has put the industry on notice. The company has been making big moves, and it’s no surprise that more than 16,000 people wanted to see what’s next at this year’s Nvidia GTC, which ran from March 18 to 21.The jam-packed event featured a keynote from CEO Jensen Huang, who declared the event the “conference for the era of AI.” It’s an amazing time in the tech industry, but the event delivered in terms of setting the tone for where artificial intelligence systems are going, according to theCUBE Research executive analyst John Furrier (pictured, right)

Keynote analysis from Nvidia GTC 2024: Spearheading AI and accelerated computing innovation

Nvidia Corp.’s annual GTC event has evolved from a developer conference to an industry artificial intelligence trade show with a cultlike atmosphere and incredible momentum, and that trend was starkly apparent this week as more than 16,000 attend in person for the first time in years and as many as 300,000 look on virtually.That’s according to theCUBE Research analysts John Furrier (pictured, left) and Zeus Kerravala (right), who are both at the event this week in San Jose. AI is just another form of accelerated computing, and nobody does it better than Nvidia, they said.“It’s all things accelerated computing and that’s really I think the nuance that people miss a lot,” Kerravala said

Blackwell: Nvidia’s GPU architecture to power new generation of trillion-parameter generative AI models

Generative artificial intelligence models will become more powerful than ever, thanks to a new, more advanced architecture for graphics processing units announced by Nvidia Corp. today.The company unveiled its all-new Blackwell GPU architecture (pictured) at its GTC 2024 event today, saying it’s based on six transformative new technologies that will bring accelerated computing to the next level. With the new Blackwell GPUs, companies will be able to run real-time generative AI powered by more expansive, 1 trillion-parameter large language models, with an impressive 25-times lower cost and power consumption than Nvidia’s existing H100 GPUs, based on the older Grace Hopper architecture

15:15

16:37

Info

Day 1 Keynote Analysis with John Furrier & Zeus Kerravala

Zeus Kerravala

Founder & Principal Analyst ZK Research

>>Everyone, I'm John Furrier with theCube. We are here in San f, San Jose, not San Francisco, San Jose for the Nvidia GTC 24 with Zeus Kerravala, with ZK research, with keynote analysis. Got all the action going on, got trains going by, but nothing more impressive than a packed house at SAP Center. Jensen Wong was given the keynote, looked more like a Taylor Swift concert. Taylor can hold a candle to Jenssen's awesome performance Z It's great to have you on. Yeah,

>>It's a pleasure. It's been like, what, A couple weeks since I've been on the cube. So yeah, I thought it was a interesting how Jensen called it a developer conference because it is certainly much more than that. I think it's, in fact, the evolution of GTC has been nice too. 'cause it really was a developer conference for a while, but now it has become a, an industry, the industry AI trade show.

>>It's interesting how that sounds like an HPC event. Everyone's going crazy. It's very Steve Jobs, like on one hand, very cult-like in the coolness of this next generation AI wave, but also felt like supercomputing built in. So it was like cloud supercomputing, reinvent meets Apple Developer Conference. Really an incredible momentum. To me. It was a a point in history where I think the system's revolution is definitely happening. Well,

>>It's, it's all things accelerated computing and that's really, I think the nuance that people miss a lot. AI is just another form of accelerated computing. So is the high performance graphics, so is robotics, computer vision, all those things. It's all wrapped up in ai, but nobody does accelerated computing better than, than Nvidia. Well let's

>>Get into the highlights. Blackwell is, was announced. That's, that's a, that it replaces Hopper, but Hopper becomes a CPU, the AI super chip also ENV link switch the key ingredient as part of this DGX cloud and system 120 kilowatts in one rack and exo flus, it's in one rack was monster. They call it the spine. It's a neural network. It's, I mean, this is the future system. I joked on Twitter, it looks like a data center. I'm like, and everyone's like, you're wrong. Like, I was only kidding. What's

>>Your, well, no, it is a data center that's in a box. I think one of the interesting aspects of it with Blackwell is now what, what is a, what is a GPU? Like what is a chip? Because it's actually two chips bound together. But then if you NVLink 'em together right now you've got multiple ones. And then if you put 'em in a bigger NVLink, NV switch, right? You create a bigger one. And so the, there's a real blurring going on here about where the chip ends and where the system starts. And in fact, I think you really can't differentiate between one of the, between them

>>Anymore. And I think, I think a AI systems are gonna come fast. I call it clustered systems. I think this all started, in my opinion, with AWS when AWS created that hyperscale performance. Google Azure follow Oracle's now trying to follow the idea of James Hamilton. When you put together those data centers, they really innovate on the network. If you look at what's going on here in these engineered clustered systems, the connection points between GPUs, NV link switch makes it happen. So he said on stage, GPUs talking to each other, this is now a new system. So I think you're starting to see the evolution of not just the GPU, but everything around GPUs, plural. Now GPU racks plural, it's a AI factory. They call it, I call it cluster system super cloud, whatever you wanna call it. This is change. This is a different market.

>>It is. And I think it's becoming more network centric. And so that's, it's gonna be interesting to see we, the the tailwinds by Nvidia obvious, right? You just look at their stock price. Who else is gonna benefit from this? John? I, I do think pure storage on the storage side. I think Arista on the network side, I think Cisco's got a lot of possible upside as well. And then of course Nvidia builds their own networking gear. But the network plays a critical role in ai and I think that's been an underappreciated aspect of ai, AI systems for really up until now. Right.

>>When I saw Jason one on stage, I wanna get your reaction to this next comment. 'cause I think this is kind of maybe a key point that not really being talked about, but I felt very much like that was a reinvent wannabe pitch. Like he's saying, we're the cloud, we're the system, we're the AI factory. That's very much kinda the narrative of the way Andy Jazz used to talk about Amazon Cloud Jenssen's. Not saying he's gonna be an Amazon wannabe, but it's the same narrative. It's the same movie. We've seen that before. Do you see that same thing that that on there? 'cause look at Michael Dell's out there in the audience. All these hardware vendors out there, core weaves out there. This and separate cloud could emerge.

>>Yeah, I think the, the tone of it to me was they're trying to now become the center of this AI universe, right? And then they're gonna have this huge ecosystem surrounds them. And frankly, when you look at the expo hall, it's already become that way. But I think along with that becomes the responsibility to become a little more visionary. And so when you think back to those early Andy Jassy keynotes and, and certainly every Steve Jobs keynote, it wasn't just about product. It was a lot. There was a lot of vision. And I think Jenssen's trying to take that role now for AI to be not only the manufacturer of the infrastructure that makes ai, but the visionary to set the direction of where this goes.

>>I think he had a home run with a vision, big fan of what he's doing. I love his speech. I love what they're doing. I love this AI system. I like how he called it a new category that's kind of a shot across the bow, again to the existing incumbents. But you hear words like AI foundry and the ecosystem names. Yeah. I I mean, I gotta say, I mean that sounds very cloud-Like to me, you're enabled not just developers, I mean 1.3 mil million robotics developers Yes. Alone. All the systems coming together, there's horsepowers there and you got an ecosystem developing classic enterprise move combined with the developer chops. I I mean, you gotta look at this as saying if you're Amazon Web services, you gotta go, Hmm, what's our relationship with Nvidia?

>>Well, right. And they had a, a joint announcement here and I think they wanna be the enabler of more Nvidia. I think, let's face it, Nvidia stuff's not cheap, right? So not everybody can go out and buy a DGX and stick it in their data center. And so unless if you're not a hyperscaler, I think your path to that largely will be through an Amazon or Google or Microsoft. Now you could argue that their value gets squeezed a little bit and they all wind up looking the same. Yeah. But that's where some of the other things that they bring to bear and need to be part of that. You think there's

>>A tan expansion opportunity for Nvidia without telegraphic to Amazon. Any competitive things like, 'cause I can see a scenario where Nvidia can say, I wanna partner with Amazon, you're a public cloud. I just wanna be the AI cloud. Yeah. I mean, it's a tan expansion, isn't

>>It? Oh, yeah, yeah. No, I think the idea here for Nvidia would be to make sure everybody has access to AI everywhere. Whether it's OnPrem through their own systems or through the cloud, and the cloud's gonna allow 'em to scale. And the countries where they can't reach down to smaller businesses, right? Even the, some of the stuff they're talking about, digital twin, anything that moves right? Any system should have a digital twin. Well, if you're a mid-size bank, right? You, you can't go afford to build your own digital twin, but you can buy that outta the Amazon cloud and then use that as a way to rebuild your, your bank office of the future. So,

>>Well, here's my notes from the five takeaways from the keynote. One is an industry revolution around ai, whole nother way of do business. Number two, new way how software is gonna be built with generator. That's the Blackwell connection

>>Software, right? And

>>Software. Software, right. And software and new, the, the new computer will create this new paradigm, this new software and it's how it's distributed. This whole nims Nvidia inference machines, no, in, was it Nvidia inference? Machines. Machines, I think. Yeah. Machines. Yeah. And NI creates a whole new application. That's the foundry idea. And of course robotics. Robotics, yeah. The physical world. I don't the convergence between digital and physicals here with ai. I don't

>>Know if the industry though John has figured out the modernization aspect of ai. 'cause we, I think Nvidia has. Yeah. But I think when it gets down to some of the other companies, you, you look at the contact center companies, the security companies that are trying to use AI to make their products better. I, I, when I talk to 'em about how they're gonna monetize it, they're not really sure. So one thing that concerns me here is from an industry perspective, is people are building ai, spend a lot of money building AI into their products, but they haven't figured out how to make money off it. And if they can't, then we want it with the deflationary effect for everybody. But Nvidia, and that's ultimately bad for everybody.

>>I mean, it checked all the boxes on all the hype had the nims, which is the new, you know, new way to do that. The API for the cloud, they call it. Then they have this thing called the NEMO retriever, which is really taking advantage of the whole rag market. Yeah. Retrieval, augmentation generation saying if you use Nvidia, we're the home for rag or retrieving. I love, that was my pet pet announcement. Yeah. Pun intended. Of course. DGX cloud was the key. So this whole AI foundry ecosystem is really built on this, the hype of retriever engine, API for ai. So he, they're, they're calling the nims, the API for ai.

>>Yeah. Historically use the APIs. But this is, I think, an easier way to do it.

>>All right. Final point. What do you expect for the week? We're gonna be all week. We'll see around Dave's coming in tomorrow. This is just Monday. What are you expecting to see for the rest of

>>The week? Well, we'll see a lot more partner announcements now that NVIDIA's have released all of theirs. And so I, I think they, he Jensen talked about some on stage Cohesity, Amazon, Microsoft, Google. But I expect to see a lot more. One of the aspects I'd like to see more of it that show sustainability, I think Nvidia and maybe unfairly gets beat up because the GPUs use a lot more power than traditional CPUs. But there is an argument to be made though that the more GPU enabled systems you have, the less CPO enabled systems you have. And so that brings down the overall cost of cooling and power. And so I think, but he, he alluded to that on stage. But I would like to see Nvidia hit the sustainability point a lot harder.

>>G, good to see you. We'll check in with later analyst angle later in the show. I'm John Ford, the Cube. You're watching the Cube. More coverage. Stay with, go to the cube.net. Silicon angle.com. W Walter Walco will be on Twitter, LinkedIn, all the channels. Look for us here. And let's go inside.

>>As always, great being here.

Day 1 Keynote Analysis with John Furrier & Zeus Kerravala

>>Software, right? And

>>Yeah. Historically use the APIs. But this is, I think, an easier way to do it.

>>All right. Final point. What do you expect for the week? We're gonna be all week. We'll see around Dave's coming in tomorrow. This is just Monday. What are you expecting to see for the rest of

>>As always, great being here.