NVIDIA unveiled DGX SuperPOD, the world’s first cloud-native AI Supercomputer fitted with Bluefield-2 DPUs, to provide a secure, reliable and accelerated infrastructure to cloud and edge. According to Reports And Data study, the global supercomputer market cap is expected to reach $13 billion by 2027, growing at a rate of 9.5% during the 2020-2027 period.
Supercomputers & AI
NVIDIA’s DGX SuperPOD is built to meet customers’ increasing security and scalability requirements. According to the company, BlueField-2 DPUs can offload, accelerate, and isolate users and their data, enabling companies to securely provide private cloud access to their DGX SuperPOD infrastructure while maintaining security through user traffic, firewalls, and multi-tenant storage access. DDN is the first NVIDIA storage partner to receive certification to support the DGX SuperPOD with BlueField-2.
NVIDIA Base Command enables AI developers and data scientists to quickly provision and schedule workloads on DGX infrastructure, from prototyping to output. The built-in telemetry can be availed to verify deep learning models, workload settings, and resource allocations to continuously improve performance.
“The 210 petaFLOPS Param Siddhi AI [supercomputer] equipped with DGX SuperPOD and indigenously developed HPC-AI engine, HPC-AI software frameworks, and cloud platform by C-DAC will accelerate experiments for solving India-specific grand challenges using science and engineering,” said Hemant Darbari, Director General, C-DAC.
Data centres
Image credits: NVIDIA
“The CPU is for general-purpose computing, the GPU is for accelerated computing, and the DPU, which moves data around the data centre, does data processing,” said NVIDIA CEO Jensen Huang. BlueField-3 DPUs are built for AI and accelerated computing. During the virtual summit, Jensen Huang said, “BlueField-3 DPU delivers the equivalent data centre services of up to 300 CPU cores, freeing up valuable CPU cycles to run business-critical applications.”
“Modern hyperscale clouds are driving a fundamentally new architecture for data centres. This is a new type of processor designed to process data centre infrastructure software, required to offload and accelerate the tremendous compute load of virtualisation, networking, storage, security and other cloud-native AI services. The time for BlueField DPU has come,” he added.
With 22 billion transistors, it can produce up to 10 times the compute of the previous-generation BlueField-2 DPUs, thus isolating apps from the control and management plane. BlueField-3 is the first DPU to support fifth-generation PCIe and time-synchronised datacenter acceleration, thanks to its 16 ARM A78 cores that can handle 4X the cryptography performance.
Mellanox’s technology powers Bluefield DPUs. NVIDIA acquired Mellanox last year to make inroads into the data centre space. Bluefield will pit NVIDIA against Intel and AMD. For instance, Intel has partnered with the Argonne National Laboratory, under the US Energy Department, for developing a supercomputer named ‘Aurora’, slated for release this year 2021 after several delays. The supercomputer is intended to provide unrivalled data analytics, AI, and simulation capabilities for advanced 3D modelling. Intel’s supercomputing-focused Xe architecture is built on a 7-nm process.
Meanwhile, AMD is working towards rolling out the “Frontier Supercomputer”, capable of performing quintillion calculations per second in 2021. As per AMD, the Frontier would have the computing capacity of the next 160 fastest supercomputers combined. With a bandwidth 24,000,000 times greater than the typical home internet connection, it will be able to handle an incredible amount of data.
