Datacentres on a Chip?
NVIDIA is revolutionising the way we look at datacentres – here’s how
One of the world’s leading firms in terms of chip-based processors is US-based NVIDIA, having shipped over a billion graphics processing units (GPUs) till date. Its new-age Ampere GPUs are also being regarded as the ‘fastest ramp in [NVIDIA] history.’ Hence it should come as no surprise to find NVIDIA consistently pushing the boundaries of chip-based computing and developing industry-breaking products on a regular basis. The newest in their line of processors is the BlueField-2, set to completely transform the running of cloud-based datacentres globally.
The BlueField-2
Over the last few months, the world has taken a massive step forward towards software-defined datacentres, where fixed-hardware infrastructures have been converted into flexible software that can be deployed across application servers. Traditionally, these servers have separate CPUs and acceleration engines to complete various tasks. NVIDIA, however, are taking it a step forward – and developing adventitious Data Processing Units (DPUs) that will function as the first-of-its-kind datacentre-on-a-chip.
Figure 1: The BlueField-2 DPU. Source: NVIDIA
NVIDIA’s BlueField-2 is set to be the first-in-line DPU chip that will provide accelerated datacentre infrastructure services wherein central processing units (CPUs), GPUs and DPUs will come together to deliver a holistic computing unit that is programmable, AI-enabled as well as secure. It is set to have eight 64-bit cores and several other ingrained components to accelerate networking, storage processing as well as security.
The BlueField-2 also contains an Ampere GPU and can be used for aspects such as automated responses and anomaly detection, real-time traffic analysis, dynamic security, malicious activity detection and online analytics. The applications of a DPU is set to be found on every server going forward, especially in putting software-defined infrastructure on the same chip on the same server.
NVIDIA’s BlueField-2 DPU is set to deliver the same datacentre capabilities that can currently consume up to 125 cores. By carrying out almost 0.7 trillion operations per second, the DPU is able to free up valuable space for other cores to run a wide range of other enterprise applications.
Several major global players are set to adopt DPU technology in their products in the coming years as well, including, but not restricted to, Asus, Dell, Gigabyte, Fujitsu and Lenovo.
The EGX AI Platform
The NVIDIA EGX AI platform, which has already seen widespread adoption by several tech companies for its edge datacentres or enterprises, is set to get a refresh with the release of the BlueField-2 DPU – with the DPU and the NVIDIA Ampere GPU both being coalesced into a single computing chip. Tech giants such as Dell, Lenovo and Supermicro, which have already adopted this platform into their businesses will be able to immediately reap the benefits of having massively accelerated datacentres.
With massive computing power now moving from the cloud to the edge, the need for competitive action to stay relevant is louder than ever. Datacentre capabilities need to be brought to the point of action, especially ion sectors which have already started widespread adoption of edge computing technologies, such as manufacturing, retail, healthcare, logistics and public safety, among others. The EGX AI platform is set to allow efficient deployment of AI at the scale needed to match needs.
Figure 1: The EGX AI fleet. Source: NVIDIA
According to NVIDIA’s edge computing vice-president, DeepuTalla: “Rather than having 10,000 servers in one location, NVIDIA believes future enterprise datacentres will have one or more servers across 10,000 different locations, including inside office buildings, factories, warehouses, cell towers, schools, stores, and banks. These edge datacentres will help support the internet of things (IoT).”
