Nvidia introduced its first processor for supercomputers. This is reported on the company’s website.
Equipment release took place at the NVIDIA GTC 2021 Keynote event. The new processor was called Grace, it contains more than a trillion parameters and is focused on working with supercomputers and data centers. It is reported that when interacting with NVIDIA graphics processors, a new chip provides more than 10 times greater performance than modern computer stations based on NVIDIA DGX.
The released processor is characterized by the support of the Fourth-generation NVLink interconnect technology, which provides the data transfer rate at 900 gigabit per second. The equipment also allows you to increase server bandwidth 30 times. Grace is based on the LPDDR5X memory subsystem, which guarantees 10 times greater energy efficiency compared to the memory of the DDR4 standard.
According to the company, Grace includes energy-efficient CPU cores with low-power memory subsystem to ensure high performance with great efficiency. “Advanced artificial intelligence and the assembled amount of data output a modern computer architecture beyond its limits, which allows you to handle inconceivable amounts of data,” said the founder and CEO NVIDIA Jensen Huang (Jensen Huang).
NVIDIA does not disclose the cost of the component, but clarifies that Grace is focused on a narrow consumer. Intent to acquire a new processor already identified representatives of the Swiss Supercomputer Center and the Los Alamos National Laboratory of the United States. NVIDIA Grace supplies are scheduled for 2023.