- Nvidia presents Blackwell Ultra at GTC 2025
- The new flagship GPU offers more power and efficiency than previous generations
- There is also a new NVIDIA RTX Pro 6000 Blackwell Server Edition designed for business workloads
Nvidia has removed the wrapping of her latest GPU Blackwell Ultra Ultra hardware, since she seems to affirm her place as the world -computing leader of AI.
Revealed in its Nvidia GTC event in San José, the new Blackwell Ultra hardware offers more power and efficiency in the data center than previous generations, since the company seeks to establish itself as the backbone of the development and deployment of the future AI.
Declaring that it is “built for the era of the reasoning of AI”, Nvidia says that Blackwell Ultra can help democratize the adoption of AI throughout the world, which makes it possible for more organizations to enjoy the benefits that such a computer can bring.
Nvidia Blackwell Ultra
Nvidia pointed out that the rapid growth of the cases of use of AI worldwide in recent years has led to a great demand for computing, since companies and consumers claim equally for more.
“The AI has taken a giant leap: reasoning and orders of magnitude of AI of AI in computer performance,” said Jensen Huang, founder and CEO of Nvidia.
“We design Blackwell Ultra for this moment: it is a single versatile platform that can make an inference of the prior to pretratenation, after training and reasoning easily and efficiently.”
The increase in reasoning models in particular has led a boom in the requirements for a complete pile offer that is profitable but also does the job.
Built in Blackwell’s initial architecture presented at GTC 2024, Blackwell Ultra will offer 1.5X more FP4 inference, and will be configured for indecs available by Nvidia Partners in the second half of 2025.
It will be present in a large number of new NVIDIA offers, including the GB300 NVL72 Rack, which says it offers better energy efficiency and service capacity, with bandwidth speeds up to 130TB/s.
Nvidia RTX Pro 6000 Blackwell Server Edition
However, that was not all when it comes to New Blackwell Hardware, since the company also revealed the NVIDIA RTX Pro 6000 Blackwell Server Edition edition, designed for business workloads such as Inference of Multimodal, creation of immersive content and scientific computing.
Nvidia says that the new launch offers great advances on the GPU of the Ada Lovelace architecture of generation of previous generation, providing an LLM inference performance of up to 5x higher for agent AI applications, almost 7x faster genomic sequencing, sequence of 3.3x for the generation of texts to videos, almost 2x of faster influence for systems of recommendation of recommendation of 2x recommendation and overene Redereciousness.
Each RTX PRO 6000 can also be divided by up to four instances fully isolated with 24 GB each to execute simultaneous work loads of AI and graphics, with 96 GB of ultra -granted GDDR7 memory and multi -instance GPU support.
They can also be configured on high -density accelerated computer platforms for distributed inference work loads, or used to deliver virtual work stations using the NVIDIA VGPU software, for graphic intensive applications or to feed the development of AI.
“With the Blackwell RTX Pro 6000 server edition, companies in several sectors, including architecture, automotive, cloud services, financial services, games development, health, manufacturing, media and entertainment and retail trade, can allow innovative performance for workloads such as multimodal work loads, simulation of data data, engineering simulation and engineering simulation Visual “, notable Nvidia Gutepte.
