NVIDIA's Grace promises a 10x performance increase in AI workloads compared to x86

NVIDIA has just introduced its new data center oriented CPU based on the ARM architecture. Promises 10x the performance of today's leading servers.

The new specialized microchip uses energy efficient cores, fourth generation NVIDIA NVLink interconnect technology, novel LPDDR5x memory subsystem, unified cache coherence and several other key developments aimed at optimizing the training and processing of AI models and targeting other areas of high-performance computing (HPC) at the same time is currently dominated by other manufacturers.

The new product is the first data center CPU in the company's portfolio and promises a lot. According to a press release released yesterday, NVIDIA Grace should outperform today's fastest servers by a factor of 10 and meet the demands of the most complex workloads such as natural language processing, data analysis and supercomputing.

For maximum performance, Grace CPUs must be paired with GPUs, also developed by NVIDIA. The fourth generation CPU-GPU interconnect technology can maintain a record connection of 900 GB / s and achieve 30 times the total bandwidth compared to today's fastest systems.

The Swiss National Supercomputing Center (CSCS) and the Los Alamos National Laboratory of the US Department of Energy have already announced that they will be building the first supercomputers based on the Grace architecture

Source: NVIDIA Newsroom

<! –

comment these news or articles


Leave a Reply

Your email address will not be published. Required fields are marked *