Just like that Apple gave its M1 chip, NVIDA runs Intel directly with its own arm-based CPU. But don’t expect it on PCs anytime soon. Named after pioneer computer scientist Grace Hopper, Grace NVIDIA’s first datacenter targets huge workloads such as CPU, AI supercomputing and natural language processing. It is powered by ARM Neversity Core and will be firmly integrated with the company’s latest GPU technology.
According to NVIDA, the Grace-powered system is ten times faster to train natural language processing (NLP) models with one trillion parameters than its X ৮ x DGXTM machine. Like Apple, NVIDIA can hit the limitations of the current x86 hardware at the moment, so the only choice is to take things into your own hands.
NVIDIA founder and CEO Jensen Huang said in a statement, “Top AI and data science have pushed today’s computer architecture beyond its limits – processing unimaginable amounts of data,” Jensen Huang, NVIDIA’s founder and CEO, said in a statement. “Combined with GPUs and DPUs, Grace gives us the third core technology for computing and the ability to rebuild the data center to take AI forward. NVIDA is now a three-chip company. “
NVIDA wasted no time in finding buyers for its new hardware. Both the Swiss National Supercomputing Center (CSCS) and DOE’s Los Alamos National Laboratory have announced plans to launch a grace-powered system. At CSCS, it will be the hub of its new Alps supercomputer, built by Hollet Packard Enterprise. Both companies plan to launch their grace system in 2023, which will then be widely available to other customers.
Grace systems will rely heavily on NVIDIA’s Envilink technology, which will be able to deliver speeds of up to 900 GB / s between their CPUs and GPUs. NVIDA claims that thanks to almost 30 times faster and LPDDR5x memory than today’s servers, Grace machines will be 10 times as energy efficient and will offer twice the bandwidth of DDR4 RAM.
In another push against Intel, NVIDAO announced a partnership with Amazon Web Services to integrate its GPUs with AWS ‘ARM-based Graviton 2 processors. This arrangement shows just how flexible the company can be and adding another ARM-based processor can only do more damage to Intel. The company says these NVIDA-powered AWS examples will enable Android games to run locally, as well as speed up streaming and rendering and encoding games on mobile devices, the company says.
For those whose fingers need to be immersed in ARM-based High-Performance Computing (HPC), NVIDAO has announced an ARM HPC Developer Kit (above) announced that it has an Ampere Ultra CPU (including 80 ARM Neversity cores); Two NVIDA A100 GPUs and two NVIDA Bluefield-2 DPUs for speeding up networking, security and storage. Early customers include the Oak Ridge National Laboratory, the Los Alamos National Laboratory, and the National Center for High Performance Computing in Taiwan.
Given the growing importance of power-efficient supercomputers, Grace was able to succeed more instantly than NVIDIA’s last ARM-based hardware, the Tegra system-on-a-chip. The product eventually found a home on both the Nintendo Switch and NVIDIA Drive motor computers, but it appeared on HD in June and on a lot of failed convertible PCs like the original Microsoft Surface. This time, NVIDIA is actually providing what the industry wants.