US gaming and computer graphics giant Nvidia Corp. unveiled on Monday a new and “very powerful” processor that enables the transfer of all data center infrastructure services onto one chip. The technology was initially created by Mellanox Technologies Ltd., the Israeli chip maker that Nvidia acquired in 2019 for $7 billion.
The idea is to increase the workload capacity of servers by freeing up resources within the server by moving data center infrastructure services — which can take up some 30 percent of server resources — onto one chip. The new chip design that is the basis of the new processor enables “breakthrough networking, storage and security performance,” the firm said.
For many years Nvidia has worked on using the graphic processing units it has created to “accelerate applications for particular workloads like AI and data science that are used in the data center,” said Manuvir Das, the head of enterprise computing at Nvidia, in a briefing to reporters on Monday.
The company is now ready to get to the next “phase of acceleration,” which is to accelerate every server, every application running in the data center, “based on work that was done by the Mellanox team for many years, even before they joined forces with Nvidia. Now, with the power of the two companies together, we will take it to a whole new level.”
Enterprise data centers are made up of servers that have applications running on them. Over time functions that were previously done by hardware devices — infrastructure functions like storage or networking or firewalls that were placed in the perimeter of data centers — have become “software defined,” he said, meaning they are now implemented in software that is running in the servers, alongside the applications.
Studies have shown that more than 30% of the resources available on the servers are now being consumed to run the infrastructure services in software – which means that that amount of resources is not available for use by applications. And this means “more servers are needed to run the same workload. And this situation is only getting worse,” Das said, because the requirements are “growing exponentially.”
Because the amount of data and traffic and applications running together is “growing dramatically,” going forward “more and more of the resources” of the central processing units of computers are going to be consumed by infrastructure needs, “leaving less and less room for the applications themselves,” Das said. This means that customers will have to deploy more and more servers to run the same size workload.
“Our solution to this is to introduce a new piece of hardware,” a new chip called the Data Processing Unit, or the DPU, “which takes all of that software-defined data center infrastructure and hosts it on a chip.”
A single BlueField-2 DPU can deliver the same data center services that could consume up to 125 CPU cores, Nvidia said in a statement. This frees up valuable CPU cores to run a wide range of other enterprise applications.
And because the acceleration engines built in the DPU are specifically built for those functions, “they can be much more efficient than the original CPU and take on a much larger load,” Das said.
The BlueField-2 DPU is being tried by server manufacturers and software infrastructure firms and is expected to be integrated into new server systems in 2021. Leading server manufacturers worldwide — including ASUS, Atos, Dell Technologies, Fujitsu and Lenovo — have plans to integrate the Nvidia’s DPUs into their enterprise server offerings, Nvidia said in a statement.
The new BlueField-2 DPU will also have a smarter version — the BlueField -2X, which comes with artificial intelligence capacity. BlueField-2X DPUs are under development and are also expected to become available in 2021.
“This is just the beginning of our journey,” for the new product, Das said, laying out a strategy for Nvidia to come out with versions three and four of the new processing unit by 2023. The Bluefield 4 will be 600 times as powerful as the 2, he said, forecasting that every server in data centers will have DPUs eventually.
Nvidia founder and CEO Jensen Huang unveiled the company’s new processor and its development roadmap in a keynote speech at on Monday at the Nvidia GTC 2020 conference. The conference, which opened on Monday and will run through Thursday, will be attended virtually by some 30,000 participants globally. At the conference, the company will also announce a new platform that will allow the processing of artificial intelligence data in edge devices locally, without the need to send the data to the cloud.
At the event, Nvidia will also announce the launch of a new computer for developers for use as an AI and robotics starter kit, and the Santa Clara, California-based firm will also unveil a new platform for real time collaboration and simulation of 3D visual effects, to bring about a “revolution in the world of 3D content creation,” the firm said.
In the field of healthcare, Nvidia said it is setting up a partnership with multinational pharma company GlaxoSmithKline Plc. (GSK) to speed up the use of artificial intelligence for the development of new drugs. The partnership will create the first AI-based computational drug discovery lab for the pharma industry, and will be based in London, said Kimberly Powell, VP of healthcare at Nvidia in the briefing.
Nvidia will also create an AI-based supercomputer called Cambridge-1, which Powell said will be the fastest supercomputer in the UK and among the top 30 of the 500 most powerful supercomputers in the world, to solve large scale problems in the medical field using AI capabilities. The supercomputer will be used for science research, the UK industry, universities and startups. Nvidia will invest around £40 million pounds ($52 million) in Cambridge-1, the firm said in a statement.
“Tackling the world’s most pressing challenges in healthcare requires massively powerful computing resources to harness the capabilities of AI,” said Nvidia’s CEO Huang in his GPU Technology Conference keynote. “The Cambridge-1 supercomputer will serve as a hub of innovation for the UK, and further the groundbreaking work being done by the nation’s researchers in critical healthcare and drug discovery.”
Nvidia has previously said that it intends to create an AI Center of Excellence in Cambridge, through its acquisition of UK chip maker Arm Ltd., which will serve as a hub of collaboration for AI researchers, scientists and startups across the UK. As these plans develop Cambridge-1 will become a part of that Center of Excellence, which will expand to include further supercomputers, and support more industries across the UK, the statement said.