Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
NVIDIA CEO Jensen Huang, Taiwan, Taiwan, Taiwan, Taiwan, Taiwan, Taiwan, Taiwan, Taiwan Tan Ke Plant Site (Spil) ‘s Tan Ke is coming to attend the opening ceremony.
Ann Wang | Reuters
Nvidia New plugs were announced for the establishment and installation and placement of artificial intelligence models at the GTC conference on Tuesday.
CEO Jensen Huang revealed Blackwell Ultra in the second half of this year As a vera rubinIn 2026, the next generation is expected to be a graphic processing unit or GPU.
NVIDIA sales are more than six times because of the release of Openai Chatgpt in late 2022. Because His Great GPUS has a process that has called the market, most of the market for the development of an advanced EU.
Software developers and investors are closely following the company’s new chips, offer enough additional performance and efficiency to convince the company’s largest end customers – cloud companies Microsoft, Google and Amazon – To continue spending billions of dollars to create information centers around NVIDIA chips.
“This is the place where it is almost all over the world last year, the requirement of the EU’s scale law, and in fact, hyper is accelerated,” Huang said.
Tuesdays are the test of the new annual annual release of NVIDIA. The company is trying to announce new chip families within each year. Before the AI BOOM, NVIDIA has released a new chip architecture annually.
San Jose, the GTC conference in California is also a power demonstration for NVIDIA.
Since the pandemic, NVIDIA is expected to discuss 25,000 participants in the second person conference and hundreds of companies used by the devices for the AI. This is Waymo, Microsoft and Fordbetween others. General Engines He also announced that NVIDIA will use the service The next generation of vehicles.
After the ruby, the Chip architecture will be named after Physicist Richard Feynman, and Nvidia continued to name the chip families after scientists. NVIDIA’s Feynman chips are expected to be available in 2028, according to a slide displayed by Huang.
NVIDIA will also demonstrate other products and services at the event.
For example, NVIDIA called new laptops and desktops, including two AI-centered PCs using the chips DGX Spark and DGX station It will be able to run great AI models such as Llama or DeepSeek. The company also announced the updates with a program package called Dynamo, which helps to connect hundreds or thousands of GPUs to each other, as well as a program of Dynamo.
NVIDIA Corp’s co-founder and CEO Jensen Huang, NVIDIA GPU Technology Conference (GTC), USA, USA, San Jose, the United States, San Jose, in the United States, “GTC”.
David Paul Morris | Bloomberg | Getty pictures
Vera rubin
NVIDIA, in the second half of 2026, the next generation GPU family expects to start shipment systems.
The system has two main components: a CPU, a new and new GPU design called Rubin. Bears its name astronomer vera rubin.
Vera Nvidia’s first special CPU is designed, the company and this was based on a basis of a basic design called Olympus.
Used a design from NVIDIA to previously needed CPUs Weapon. Companies developing basic designs such as Qualcomm and Apple, say they can be able to unlock more suitable and better performance.
The company said that the special vera design will be twice as much from the CPU used by KPU last year in Grace Blackwell.
When combined with vera, Rubin can manage 50 petaflopes, the company doubles 20 petaflopes for current Blackwell chips. Rubin, one of the main features watching AI developers, can also support 288 gigabytes.
Nvidia makes a change of what it calls a GPU at the same time. Rubin is actually two GPUs, said Nvidia.
Blackwell GPU in the market is actually harvested and two separate chips that are designed to work as a chip.
Starting the rub from the rub, NVIDIA will say that when they combine two or more dead to make a chip, they will appeal to them as a separate GPU. In the second half of 2027, NVIDIA, four will refer to one chip, double GPUs, twice the Rubin speed.
Nvidia said he would come in a rack called Vera Rubin NVL144. The previous versions of NVIDIA rack were called NVL72.
NVIDIA Corp’s co-founder and CEO Jensen Huang, NVIDIA GPU Technology Conference (GTC), USA, USA, San Jose, the United States, San Jose, in the United States, “GTC”.
David Paul Morris | Bloomberg | Getty pictures
Blackwell Ultra
Nvidia also announced the new versions of the Blackwell family of plugs called Blackwell Ultra.
This chip means that can produce more verses in seconds, that is, the chip can create more content as the predecessor.
Nvidia says cloud providers can use Blackwell Ultra that can use Blackwell Ultra to offer a premium AI service that allows you to earn 50 times income from new chips such as Hopper generation sent in 2023.
Blackwell Ultra will be in two pairs and two couples connected to a NVIDIA ARM CPU called B300. You will also be in versions containing eight GPUs and 72 Blackwell plugs in a server knife.
NVIDIA, the top four cloud company, placed three times the number of Blackwell plugs like hover chips.
Depth
When China’s DeepSeek R1 was broadcast in January, Nvidia scared investors, but the NVIDIA program hugged. The ChipMaker will use several of the model to evaluate a new product.
Many AI observers, the DeepSeek model, which requires fewer chips from the models made in the United States, said that NVIDIA threatens the work of NVIDIA.
However, Huang said he was a good sign for Deepseek’s Nvidia earlier this year. DeepSeek is because it uses a process called “Foundation” that requires more computing power to give users better.
NVIDIA is better to think of new Blackwell Ultra chips models.
New thinking models require more computational models during placement, NVIDIA chips will be able to handle it.
“In the last 2-3 years, there was a great progress, a foundable progress in artificial intelligence. We say this Agentic AI,” Huang said. “It can cause you to answer or how to solve a problem.”
Follow: Nvidia begins with the GTC Conference: How to trade the Committee
