During the GPU Technology Conference (GTC) this year, many people thought that the Nvidia was finally going to reveal its new graphics architecture and announce the video cards of the RTX 40 family. Despite this, only half of this desire came true in the brand’s presentations this Tuesday (22).
After all, the company really highlighted the Hopper architecture, but instead of the GPU for the final consumer, it focused on the Nvidia H100a graphics processing solution aimed at powering artificial intelligence (AI) data centers.
The power of Hopper architecture
According to the brand, the H100 is Nvidia’s first GPU based on the Hopper architecture – a tribute to Grace Hopper, a pioneering US computer scientist -, boasting nothing less than 80 billion transistors and delivering up to six times the performance of its direct predecessor.
“Data centers are becoming AI factories, processing and refining mountains of data to produce intelligence. The H100 is the AI infrastructure engine used by companies around the world to accelerate their business.”
Among the technological innovations of this GPU focused on artificial intelligence is an even wider bandwidth, capable of transferring up to 3 TB of data per second. According to the company, 20 units of the hardware could sustain the equivalent of the world’s internet traffic.
The H100 also: brings a new deep learning model, called Engine Transformer; allows you to partition the GPU into 7 unique instances; supports sensitive computing, protecting customer models and data, gives access to 4th generation NVLink, which connects the local server to an external network with 256 H100 GPUs; and have instructions capable of accelerating dynamic programming by up to 40 times when compared to traditional CPUs.
“We are all proud of the computing power achieved in this architecture. With this novelty, we will be able to serve several sectors, mainly those that demand the use of different scales”, emphasizes Marcio Aguiar, director of Nvidia’s Enterprise division for Latin America.
The Nvidia H100 GPU supports advanced software tools that enable developers and businesses to build and accelerate applications from AI to HPC. This includes major updates to workloads like speech, recommender systems, and hyperscale inference.
The product is scheduled to arrive in 3rd quarter of this yearraising expectations that other hardware based on the Hopper architecture – such as the RTX 40 video gamers – will also be released in the period.
Source: Nvidia / Advisory