Invezz explains: Nvidia’s Blackwell summed up in four questions

Nvidia’s GTC conference has only just begun its second day, and there’s already a lot to process (pun intended). 

Easily the biggest highlight so far has been CEO Jensen Huang officially unveiling the new Blackwell chip and platform… but it wasn’t all that easy to follow if you don’t have some sort of computing degree from an Ivy League. 

Here are four questions to ask (which we’ve answered below) to understand the significance of Blackwell for Nvidia – and for the world.

1. What is Blackwell?
2. How is Blackwell different from Hopper?
3. What will Blackwell do for AI and computing?
4. What will Blackwell do for Nvidia?

1. What is Blackwell?

The first thing that Nvidia CEO Jensen Huang did at GTC 2024 yesterday was pull out Blackwell – Nvidia’s latest generation superchip, officially unveiled at the conference for the first time.

The Blackwell is a more powerful GPU than Nvidia has ever created before, and the company is regarding it as the successor to its Hopper chip.

Simply put, it’s Nvidia’s latest and most technically impressive GPU chip (in terms of processing power and speed) to date.

2. How is Blackwell different from Hopper?

What’s important to note first off is that the Blackwell is brand new, and just barely introduced to the world.

This means that the host of independent reviews, testing and user experience cases we will no doubt get of the Blackwell simply don’t exist yet. No one outside of the company has tested it in any great length.

And so, we need to rely on Nvidia’s word for telling us what it does (for now at least).

That being said, Nvidia believes it’s a complete game-changer. According to Nvidia’s blog:

Blackwell delivers 2.5x its predecessor’s performance in FP8 for training, per chip, and 5x with FP4 for inference. It features a fifth-generation NVLink interconnect that’s twice as fast as Hopper and scales up to 576 GPUs.”

Simply put, that means the Blackwell is able to handle exponentially more data – and different types of data, from text to images and videos or more – faster and more intelligently than the Hopper could.

Hopper can process 4,000 teraFLOPS of data (one teraflop represents the ability to calculate one trillion complex mathematical calculations per second). That’s approximately 400x the computing power required for a Tesla Model S car.

Blackwell, on the other hand, can process five times that amount – 20,000 teraFLOPS of data.

And the best part? Blackwell is fully integrated into the same Nvidia systems and computers the Hopper is – simply slide the Hopper chip out and replace it with the Blackwell.

3. What will Blackwell do for AI and computing?

We’ve all witnessed firsthand, thanks to ChatGPT, the power of AI to change the world. But the truth is that transformative AI requires deep machine learning, tons of storage and memory recall, which most computing systems simply doesn’t have the horsepower or sophistication to deliver.

That’s where the Blackwell platform comes in. It’s not only bigger and more powerful than any other chip and processing system currently on the market (the biggest Nvidia has ever created by far), it is designed especially for the complex calculations required for AI machine learning.

This means that, in short, any organisation that can afford a Blackwell (operating at its full capacity, in the whole platform offering) can access the ability to create and run their own generative AI that is powerful enough to operate in real time.

Imagine having your own ChatGPT – not necessarily with text, but images or videos or anything else you require – for whatever you or your business needs, second for second.

Some of the fields Nvidia showcased this in included assistant robots, self-driving cars, running factories employing both humans and robots, gene sequencing and pharmaceutical R&D in medicine and even the graphics mapping and processing of geographical areas as big a planet.

4. What will Blackwell do for Nvidia?

Simply put? Blackwell cements Nvidia at the centre of every investor’s favourite new industry: AI and generative computing.

By doing this, it also puts Nvidia squarely amongst the biggest long-term plans of some of the world’s most powerful people and companies.

As Nvidia had to say on their blog:

The industry has already embraced Blackwell. The press release announcing Blackwell includes endorsements from Alphabet and Google CEO Sundar Pichai, Google DeepMind CEO Demis Hassabis, Meta CEO Mark Zuckerberg, Microsoft CEO Satya Nadella, OpenAI CEO Sam Altman, and Tesla and xAI CEO Elon Musk.”

It’s also set to make them a lot of money, if Huang is to be believed.

Huang said during the GTC conference that Amazon Web Services alone is building its own server system using about 20,000 GB200 chips, using two Blackwell GPUs apiece. And that’s just Amazon – reportedly, Google, Microsoft and Oracle are just some of the major tech players also interested.