Enlarge image
Nvidia boss Jensen Huang at the presentation of the new “Blackwell” technology
Photo: Josh Edelson / AFP
Nvidia had an ice hockey arena converted to demonstrate its latest products.
16,000 people watched as company boss Jensen Huang presented a new computer system and what he described as the “most powerful chip in the world” in San Jose, California on Monday: “Blackwell” was “the engine for a “new industrial revolution,” promised Huang.
That sounds cocky.
But unlike many other company bosses, there are many reasons to take the Taiwanese-American founder seriously.
Since ChatGPT was released at the end of 2022, Nvidia's value has increased around sixfold - to more than two trillion dollars.
In the past twelve months alone, the company's shares have gained 240 percent.
Nvidia is often referred to as a chip manufacturer.
But that's slightly misleading because the company doesn't produce chips at all.
To put it simply, the company's founder's achievement is to have developed graphics processors into brains for computers.
These computer brains are now used worldwide to train artificial intelligence.
And now they should work even faster and better with “Blackwell” and the “Blackwell B200” chip.
With current Nvidia technology, an AI chatbot like ChatGPT can be trained within three months, but this requires 8,000 Nvidia chips with a power consumption of 15 megawatts, says Huang.
With "Blackwell" you can create such a chatbot in the same time - with 2000 chips and only four megawatts of electricity.
More efficient, faster, better, easier - that's how Huang's presentation of "Blackwell", named after the American mathematician David Blackwell, can be roughly summarized.
The initial reactions from the expert audience were positive.
“We assume that Nvidia will not only continue to lead, but will also increase the gap to its competitors in the area of AI,” the Reuters news agency quotes industry expert Kinngai Chan.
During his presentation, Nvidia boss Huang also brought a cute little robot onto the stage.
“The ChatGPT moment for robotics could be imminent,” said Huang and also explained how he imagines it: In the future, robots should learn simply by observing people.
The future, as Huang described in his presentation, looks like this: Web content will no longer be retrieved prefabricated from storage, but rather generated fresh by AI software.
Companies have “digital twins” on which individual decisions are tested before someone implements them in the real world.
And in general: Before something is built in the real world, it is first simulated digitally.
vet/dpa/Reuters