Nvidia adds Arm support to its 5G kit and a 16-core BlueField DPU to accelerate Edge, HPC • The Register
Mobile World Congress Nvidia has decided to bet on the Arm and x86 platforms for the 5G edge, with a new data processing unit – aka SmartNIC – which contains 16 Arm cores.
The upcoming BlueField-3 A100 will feature 16 Arm Cortex A78 processors, but will be happy to work inside Intel or Arm-flexing servers.
Wherever it operates, Nvidia believes it has sufficient computing power to serve in software-defined 5G base stations and manage their network-based needs so that server processors can be left to others. tasks.
Nvidia’s vision is also for these base stations to include their small form factor EGX GPUs and for these units to be commissioned to run AI applications.
To make things easier, the company’s Aerial software platform has been infused with a 5G RAN.
Just why the 5G Edge needs AI hasn’t been explained at The register. Perhaps this is why Google Cloud and Nvidia decided to create a joint AI-on-5G lab to explore the opportunities.
Nvidia also advanced the BlueField-3 A100 as a good choice for high performance computing (HPC), for which the company introduced a new 80GB PCIe A100 GPU. The new machine has 2TB / s of memory bandwidth and 80 GB of HBM2e high-bandwidth memory.
Also for HPC types, Nvidia introduced InfiniBand 400 Gb / s switches and the new “Magnum IO GPUDirect Storage” which provides direct memory access between GPU memory and storage.
The combination of the BlueField-3 A100 and the other new kit would improve Nvidia’s HGX supercomputing platform so that it can propel humanity into a fourth industrial revolution that will use AI everywhere to produce a fulfilled future. unicorns, rainbows and general intelligence.
The University of Edinburgh has acquired: Nvidia has announced that HGX will power a new super called “Tursa” there, a machine described as allowing researchers “to perform the very high-precision calculations of the properties of subatomic particles necessary to interpret the data from massive particle physics experiments, such as the Large Hadron Collider. ”
The folks at Nvidia who explained all this new stuff to The register were pleased to have recognized the potential offer of DPUs to 5G and HPC. When we first considered DPUs in September 2020, The register noted that they are a good idea, but integrating them into actual production systems was not for the faint-hearted.
Nvidia has made great strides in ensuring that its platforms, at least, are fully DPU compatible. ®