Nvidia Blackwell announces ultra and vera rubin Ai Chips

Nvidia New chips were published in the annual GTC conference on the GTC annual GTC conference on the creation and placement of artificial intellectual models.

Jensen Huans General Director Landwell ultra, the supply of chips in the second half of this year, as well as Like vera rubyGPU of the company’s next generation graphic processing unit or delivered in 2026.

The sale of NVidia exceeds six times, as his business ended with the release of Chefan’s Cheuspy at the end of 2022. Because it has a lot of marketing, because it is a great deal of marketing, because it is a big GPUS.

Software developers and investors are carefully monitoring the company’s new chips, if they are offered enough additional work and efficiency for the largest customers of the company, including the largest customers of the company. MicrosoftWhere Goont and Amuson – Continue to spend billions of dollars to create data centers around Nvidia chips.

“It’s all over the world last year. The calculation requirement, the scaling law of AI was flexible, and in fact, hyper-accelerated, “Huang said.

Tuesday announcements are also a new cadence of NVIDIA’s annual issue. The company tries to update new chip families. In front of AI Boom, Nvidia has released new chip architectures every year.

The GTC conference in San Jose, California is also an indicator of strength for Nvidia.

The event, the second-come conference, believes that 25,000 participants and hundreds of companies have used the company AI of 25,000 participants and hundreds of companies since pandemic. It is WAMPO, Microsoft and Fordbetween others. General engines This stated that NVIDIA will use its service The next generation cars.

NVIDIA also provides other products and services at the event.

For example, NVIDIA has published two AI-oriented computers that can use new laptops and desktops, including large AI models such as Llama or Depesek. The company also united its network parts, they published hundreds of or thousand GPUs, as well as a software package called Dynamo, as well as Dynamo software package, which helps users to achieve maximum access to users.

Nvidia expects to start GPU in its next joints in the second half of 2026.

The system has two main components: the processor called Vera and the new GPU design, ruby. His name is given Astronomer Vera Rubin.

Vera is the first CPU design of Nvidia in Nvidia, the company reports based on the basic design called Olympus.

Previously used a design of Nvidia-standing design when it is necessary Armament. The existing ARM designed ARM, for example, said Qualcomm and Apple, for example, can give complete and good results.

Custom Vera design will be twice as fast as CPU, consisting of last year’s grace, the company reports.

When you pair Vera, rubbins can control 50 petaflops in position, which can double the 20 petaflops for modern black-colored chips. Rubin also supports 288 gigabytes of gigabytes, which is one of the main characteristics that AI developers are watching.

NVIDIA also makes changes to what is called GPU. Rubin actually said two GPU, N. Videlia.

Currently, the Blackwell GPU in the market is practically two separate chips to work with one chip.

When Nvidia, which starts with rubbin, says he is referring to them as a separate Gpus to create a single chip. In the second half of 2027, Nvidia plans to produce “Ruby’s next” chip, which will double four deaths, which doubles the rhuba speed and is called four GPU.

Nvidia Vera Rubin said he would go to a rack called NVL144. The previous versions of the Nvidia shelf were called NVL72.

NVIDIA also published new versions of Chips Chips Commun, called Blackwell Ultra.

This chip can produce more tokens per second, so Chip can create more content at its pre-briefing, NE.E.e.

NVIDIA says, it means that these cloud providers allow Blackwell Ultra to offer for sensitive applications, which allows you to provide Premium AI service for sensitive applications, which are 50 times more than in 2023.

Blackwell Ultra comes in two paired versions with Nvidia Arm CPU, and called B300. It also happens in versions with eight GPU versions with eight GPUs and 72 black bicycle chips.

The top four cloud companies have been placed three times more than the number of Blackwell chips, Nvidia reports.

Chinese Depeople’s DEPD can be threatened by Nvidia when leaving in January, but Nvidia has received software. The chipboard directs the model to the standard of several new products.

Many AI observers demanded that the Dethsek model, which requires less chips than models made in the United States, threatened to Nvidia.

But Juang said that at the beginning of this year there was a good sign for DEPKEK NVIDIA. This “de Detsek” uses the “judgment” process, which requires more calculations to give users good answers.

New Blackwell Ultra Chips is good for models, said Nvidia.

It has developed to effectively perform its chips, so when new judgments can process the Nvidia chips when you need a lot of calculation when using.

“In the last 2-3 years, has achieved great success, we have a major progress in artificial intelligence, we call it agent AI,” he said. “It can think about how to respond or how to solve the problem.”

View: Nvidia initiates its GTC conference: how the committee must sell it