5G promises to be one of the greatest advances in mobile networks today. Anys see an opportunity to improve performance via an improved physical interface, including the use of millimeter waves, albeit at short range. Latency could be reduced by a factor of ten, bandwidth at best multiplied by a thousand, the number of clients per square kilometer multiplied by several million.
Others are looking at the virtualization side of the network. In fact, one of the advantages of 5G is that each component will be virtualized, which follows the current trend in data centers: in addition to the physical antenna, all the components of a 5G network can be distributed in virtual machines, hosted anywhere in the world.
Eg in a computer center, everything can be virtualized except the network connection external: each component of an application (database, application server, etc.), but also the parts of the network (load balancing, routing, etc.), commonly referred to by the acronym SDN (defined network by software).
All these changes do not just wander in ideas or prototypes if not that have already been brought to reality and is whate Nvidia CEO Jensen Huang recently launched a series of news at the Mobile World Congress in Los Angeles, announcing a new supercomputing platform, a software developer kit and new alliances with Microsoft, Red Hat and Ericsson, all intended to showcase Nvidia's graphics processing units (GPUs) and how these can drive new 5G services and systems.
Aerial is a development kit software that allows telecommunications providers start with the two key elements of Nvidia's 5G offering- A 5G signal processing engine that contains data in the CPU memory for higher performance and a Mellanox network interface that directly connects the GPU memory with the network interface.
uVNF focuses on virtualization of network functions, that is, a processing optimized input-output operations and packet management through a direct interface to the network card (GPUDirect).
cuBB is more oriented towards the physical layerhence the signal processing (for which the GPU has been great for a long time), particularly with cuPHY for the 5G physical layer implementation.
There are already network cards that are compatible with GPUDirect technology, for example, the Mellanox CX-5. For GPUs, the entire NVIDIA family can be used from Jetson Nano to EGX servers.
The combination of Aerial and EGX makes gpu acceleration possible within a telecommunications network infrastructure built on Kubernetes. The EGX stack is equipped with an Nvidia driver and company plugins for Kubernetes and container runtimes. Finally, the stack contains gpu monitor tools.
Nvidia's offer for Mellanox was accepted earlier this year, so IThe first users of the new solution are Walmart, BMW, Protect & Gamble, Samsung Eletronics, NTT East, Ericsson, Microsoft and Red Hat. The US cities of San Francisco and Las Vegas have also been added.
NVIDIA CEO Jensen Huang said that:
The solution is indispensable as we enter an era where billions of IoT devices that are online non-stop will use 5G connectivity and will not function without artificial intelligence.
According to Huang, this requires a new class of high-performance computers, with solid protection and easy remote control. Indirectly described NVIDIA EGX Edge Supercomputing. The performance of the solution should be well beyond today's personal computers and cloud services.
Nvidia wants to be part of the 5G future because you believe connectivity is an essential part of tomorrow's computing landscape. The hardware designer previously recorded a vision for the future in which the computer landscape evolves into a hybrid model with hardware close to where the data is, connected on a global scale.