🐲 NVIDIA Is the Legit Raid Boss of the AI Industry

Hello, fellow nerds 🧙

The most hyped AI keynote of the year took place yesterday at NVIDIA GTC, and hot damn consider us impressed.

The speaker, Jensen Huang, CEO of NVIDIA, announced a plethora of new advancements and releases that had us drooling. So buckle up because today’s going to be chalk full of what NVIDIA released to the world, and why it all matters. Let’s dive in!

Here’s the daily quests in AI to complete today:

• 🐲 NVIDIA Is the Legit Raid Boss of the AI Industry

• 🤯 This Piece of Tech Blew Our Minds 20 Years Ago

• 🦖 A T-Rex Could Beat Michael Phelps in a Race

• 👎️ Sam Altman Thinks ChatGPT 4 “Kinda Sucks”

NVIDIA Is the Legit Raid Boss of the AI Industry

… and we don’t see anyone stopping them anytime soon.

If you’ve been curious about NVIDIA’s impact on AI and why their stock has been shooting to the moon, boy, do we have a treat for you.

The hype was unreal at NVIDIA’s GTC (GPU Technology Conference) when CEO, Jensen Huang, took the stage. He filled a stadium full of AI nerds and business professionals alike. The companies presenting workshops and trainings at the event are worth north of 100 trillion dollars, which goes to show how big AI truly is.

@DrJimFan compared him to the tech version of Taylor Swift, and we couldn’t agree more. Each time Jensen “dropped the mic,” we dropped our jaws.

So how much can one person say to change the course of the world in the matter of only two hours? Well, turns out you can say a helluva lot.

And that’s where our adventure begins.

Jensen’s presentation wasn’t as smooth and polished as Steve Jobs but he’s emulated Apple in a way most businesses only dream of. He’s created an NVIDIA AI ecosystem and built software and business use cases to get everyone on board.

In the beginning, NVIDIA’s focus was to create amazing graphics cards for video editing, gaming and 3D processing, but they soon realized how the brain of the graphics cards, GPUs, were meant for so much more.

Why do GPUs matter to me?
Today, almost every AI company in the world uses the NVIDIA Hopper GPU to train large language models such as ChatGPT. But Jensen just announced the new Blackwell GPU which is up to 25x faster and uses approximately 25% less power.

The Blackwell GPU is really two chips engineered side-by-side to operate seamlessly as one.

We’re Trying to Skip the Technical, but… one term we should get to know is FLOPs.

Since GPUs are specially designed to handle stressful and intense workloads, they’ve been perfect for Generative AI, which relies on extensive floating-point arithmetic calculations, aka computations, measured in FLOPs (Floating-Point Operations Per Second) frequently abbreviated as FP.

FP4, FP8, FP16, etc. are different levels of precision, just like 1.384937 is more precise than rounding up to 1.4 for example. Note: FP4 is the least precise here.

We were disappointed to recognize how the graph is using different units of measurement to exaggerate the illustrated difference 😢 

Introducing, the Blackwell Platform
Don’t let the chip itself fool you, where NVIDIA has truly blown us out of the water is through the ecosystem surrounding the chip. In order to take full advantage of the Blackwell GPU, NVIDIA had to figure out how to distribute the data, do the computations, and then put everything back together, quickly and efficiently.

So, they built six revolutionary hardware and software technologies that will work in harmony to allow thousands of GPUs to act together as one.

Blackwell Platform Technical Features (only for the real nerds)
- 208 Billion Transistor Chip (created from two chips, unified as one)
- 2nd Gen Transformer Engine (ability to handle 10 trillion parameters - ie. GPT4 uses 1.76 trillion parmeters)
- 5th Gen NVLink (the communication link between handling 1.8TB/s interconnect to distribute data)
- RAS Engine (internal self-testing and diagnosing to prevent problems)
- Security (encryption, privacy)
- Decompression Engine (accelerate the ability to put everything back together quickly at 900GB/s)

To read more about Blackwell, check it out here.

Fun Fact: The name NVIDIA is a combination of the Latin word invidia, which means "envy", and the acronym NV, which stands for "next vision"

Enough of the technical jargon, let’s get to the fun stuff!

The Omniverse

Now we didn’t really know the Omniverse existed until this keynote, but we’re never going to forget its existence.

Simply put, the Omniverse is a digital universe of the world that’s pretty much like real life. We’re talking actual physics, gravity, and exact replicas of buildings and scenarios all built into this world. You can honestly build whatever your heart desires.

It’s pretty much a video game on steroids.

And that brings us to this idea of a “digital twin”. Basically, everything in the world is going to have a digital version to match the physical version.

But why?

Well, that’s where things start to get exciting because the possibilities for this are endless. Here are just a few.

Natural Disaster Prediction

Currently, our knowledge of natural disasters is just north of an educated guess. We have some semi decent models to predict where a hurricane will hit, when it’ll hit, and potential category it’ll be. But nothing is concrete, and that’s where things get dangerous.

The solution? How about creating an exact digital replica of the world? That’s exactly what NVIDIA did.

Introducing Earth 2
A digital twin of the actual Earth. It’s powered by a brand new generative AI model called “CorrDiff” that blows current weather prediction models out of this world with:
- 12.5x higher resolution images
- 1000x faster
- 3000x more energy efficient

Tsunamis, hurricanes, earthquakes, you name it. The precision level is night and day, and the impact it’ll have on natural disaster preparation… priceless.

And this can all be simulated thanks to the Omniverse.

Teaching Robots How to Do “Adulting”

And by “adulting” we mean walk, talk, pay bills, and get a job. Well, the last two maybe not so much, but hey, maybe.

What the Omniverse does for robots is actually pretty sweet. You can use the AI brain that will one day operate a physical robot and train it in a virtual robot’s body.

How it learns is by navigating a playground of sorts. A jungle gym for robots, if you will. Walking, obstacle navigation, hand manipulation, and depth perception are just a few things the AI can learn effectively without ever setting foot in the physical world (Google is doing something similar).

The kicker to this is that a digital robot can run these learning simulations millions of times.

The ultimate goal is to train robots to perform at an adult level right outta the box instead of having the AI learn and grow starting at the toddler phase. Although it may be cute, ain’t nobody got time for that.

Customizing The Perfect Vehicle

Imagine having the power to pick any vehicle in the world and customize it exactly the way you want. Everything from the color, the trim, the wheels, and any other accessories you wanna slap on it. All without leaving the comfort of your home, or being hassled by a used car salesman. The world could be your oyster.

There’s a pattern emerging here. This is reality.

Leave it to NVIDIA. They just partnered up with Apple and integrated the Omniverse into the Apple Vision Pro so you can have the power of car customization at your fingertips. Oh, but it doesn’t end there. Not yet.

When you finish your customizing business, go ahead and pull up a chair. Have a seat. Then watch as your freshly customized car generates itself around you. Every detail becomes generated. Inside and out. How about opening the door and walking around the exterior or taking a peak in the backseat to see how much legroom you’ve got?

It sounds crazy, but it’s all possible.

GROOT

No, I know what you’re thinking. This isn’t the cute and cuddly woody GROOT you’re thinking of. We were mildly disappointed when we first found out as well, but then the potential of this GROOT turned our frown upside down rather quickly.

The world loves acronyms, so real quick, this is what GROOT stands for: Generalist Robot OO Technology (we have no clue what the “OO” stand for), but it aims to take humanoid robots to the next level.

Now normally, robots could only learn by the tried and true method of trial and error to fine tune movements and thinking. With GROOT, some different abilities have been unlocked. Humanoid robots now have the power and capacity to learn from human demonstrations (small movements like waving, high fives, fist bumps etc.) aka imitation learning, leveling up in the Omniverse, and generating robot movements from watching videos.

The one that stands out the most to us is robots can learn from watching videos. Plop little tin man down in front of the cooking channel, and he’s definitely gonna pick something up. Or a YouTube black hole on blackbelt origami. To be honest, either or would be cool as hell.

💻️ Tech Tuesday

An interesting tidbit about technology.

With NVIDIA taking up the tech space as of late, I wanted to go back in time and unveil the pinnacle of technology. The apex of human invention. At least in my brain. When I saw this appear in the wild for the very first time in elementary school, I was taken aback. I asked myself, “What genius designed such a majestic device?” We may never know…

WARNING: Brace yourself, this might be almost too much to handle.

Daily Delight

Just something fun and interesting around the web.

Listen, there’s a time and place for everything, and I feel like talking about dinosaurs couldn’t be more appropriate in this moment. The fact that the infamous T-Rex, Ruler of Land and the King of the Cretaceous Period, was great at swimming… blows my mind. This is mildly unsettling to me.

What’s happening inside the Realm

A list of side quests to explore and more.

• Sam Altman, CEO of OpenAI, said ChatGPT 4 “kinda sucks”, but ChatGPT 5 will blow our minds like 4 did.

• Pfizer introduced the world to, Charlie, a generative AI platform that specializes in marketing drugs.

• Apple looks to expand its AI even further with the help from its archnemesis, Google.

• Mercedes-Benz partnered with robotics company, Apptronik, to bring their Apollo robot to the manufacturing floor.

That’s all the quests we have for today. Check back tomorrow for more!

Looking to become a sponsor?

Was this email forwarded to you? Subscribe here.