Nvidia's tech jargon and what it all means: YF Explains

Nvidia (NVDA) is one of the most valuable companies by market capitalization, retaining this status after posting first-quarter earnings on Wednesday that topped expectations. However, the company's highly technical terms may confound some investors.

Yahoo Finance's Julie Hyman joins Wealth! to break down all the jargon Nvidia uses when referencing its products and services.

For more expert insight and the latest market action, click here to watch this full episode of Wealth!

This post was written by Nicholas Jacobino

Video Transcript

Well for much more on NVIDIA, NVIDIA being one of the most valuable companies by market cap in the world.

But it uses a whole lot of jargon to break down exactly what they're even talking about.

We've got Yahoo Finance's very own Julie Hyman here.

Hey, Julie!

Hey, Brad, thank you Indeed, what NVIDIA does is highly technical.

And so, even though CEO Jensen won in my conversation with he and Our and how last night, he does tend to explain things well.

But still, there are some terms that might confound you, so that's why I'm here to help.

So first of all, there's the graphics processing unit.

This is the bread and butter of what Invidia does.

What is a graphics processing unit?

It's a type of computer chip or semi conductor.

Those terms used pretty much interchangeably, and it's capable of rendering graphics for display on an electronic device.

Sounds pretty simple, right?

GP US are used for all kinds of stuff.

They're used for video games.

That's one of the first use cases and one of the bigger use cases in the past for invidious chips, for example, not to be confused with CPU Central processing units, which are mostly used in computers, for example.

So CPU versus GP US, the kind of GP U now that NVIDIA is going to be introducing later this year, is called Blackwell Now.

Blackwell was introduced at NVIDIA's GTC conference this spring.

And what is Blackwell exactly?

Well, NVIDIA calls it the world's most powerful chip, and I am no tech expert.

But from what I've gleaned about this Blackwell chip, it's actually two chips that are sort of, uh, connected.

And it has four times as much, uh, of speed as does Hopper, which is the generation of chip that is out from, uh, NVIDIA right now.

So it's faster, has more compute power, and that's really the big selling point here.

I should also mention Blackwell isn't just a chip sort of in isolation.

Like the recent generations of NVIDIA's products, it's the chip on a rack that sits in a data centre.

So it's the rack itself that NVIDIA also sells, along with software to sort of run all of these chips together to help them compute, especially with generative a i.

This Blackwell system also has a higher energy efficiency than its predecessors, so meaning when you're in a data centre using an enormous amount of power, as we've talked frequently about you want a chip that is going to be more energy efficient.

And then finally, the other thing is in those data centres.

It is hot people.

You are talking about chips again that are using an enormous amount of energy.

There are different ways to cool those racks that sit in the data centres.

You can cool them with air.

You can actually cool them with water that runs on pipes through those racks.

There are different ways.

Some chips in the past could only be used with a specific type of cooling system.

Blackwell is cooling system agnostic, so these are the various ways that it is different from past chips.

I should also mention it being a newer generation chip.

It's also going to be more expensive, and then another term or two terms.

You might have heard a lot about it.

When we're talking about generative A, I specifically is training versus inference.

What are we talking about there?

Well, training is when the A I model reviews a lot of different material text video images as it learns as it learns.

You know what search results.

It's gonna give you, whatever the task, it's being given.

And then inference is, then, when it actually starts to work, it is making inferences.

It's making addictions.

It's analysing based on what it has learned Now.

Some of the thinking going into this earnings report had been that invidious chips were very good for training, but they were maybe a little more than you needed for the infer part.

Training is where you need more power.

These are more expensive chips and some of their competitors.

Jensen One court sort of pushed back against that and said, Indeed, in video chips are being used for infer.

And here's what he told us last night in Fencing, as you know, is about generation of information generative a I.

And so whenever you're talking to chat GP T and is generating information for you, or drawing a picture for you or recognising something and then drawing something for you, that generation is a brand new, uh, infer.

Technology is really, really complicated and requires a lot of performance.

And again, he emphasised that those in video chips are indeed being used for infer.

We'll see how the whole landscape is gonna look going forward.

But just a few terms to get you started here, Brad.