For the GPU Technology Conference (GTC), chipmaker NVIDIA revealed its intention to pursue a data center-on-a-chip strategy and to create a robot brain, said Venture Beat.
In a keynote address, NVIDIA CEO Jensen Huang disclosed plans for a new chip designed to make running cloud-based data centers easier. To be called NVIDIA BlueField-2, the chip is seen as a data processing unit (DPU).
A DPU functions similarly to a graphics processing unit (GPU), but instead of graphics, it will deal with networking, storage technology, and security. The chipmaker acquired such tech after purchasing networking company Mellanox in 2019.
NVIDIA also acquired Arm Ltd last month and will be using the firm’s artificial intelligence and computing power, Reuters reported.
According to NVIDIA head of enterprise Manuvir Das, BlueField-2 offers an enhanced data center infrastructure that integrates the functions of GPUs, DPUs, and central processing units (CPUs). This combination becomes a computing unit that is programmable and secure.
He added, “This is really about the future of enterprise computing, how we see servers and data centers being built going forward for all workloads, not just AI-accelerated workloads.”
Venture Beat explained that BueField-2 combines all processing units for a streamlined system that can be considered a data center-on-a-chip. This is a huge shift compared to traditional servers that use separate CPUs and acceleration engines for specific tasks.
The company’s EGX AI platform uses the Blue-Field-2 DPU to serve as “the new building blocks of accelerated data centers.” EGX AI-based server systems are available from various manufacturers such as Dell Technologies, Lenovo, Supermicro, and Inspur.
In the same keynote address, Huang also revealed that the company has developed the latest iteration of its robot platform, the Jetson AI at the Edge. This robotics kit, designed for hobbyists, students, and educators, can be bought for $59.
The Jetson Nano 2GB Developer Kit, which is different from Jetson AI at the Edge, also goes for $59 and is intended for teaching and learning AI. It offers free online training and online certification.
These two kits come in addition to the 99-dollar kit announced back in March 2019, which used earlier chips. Edge computing vice president Deepu Talla noted that it is currently being used by over 700,000 developers, making it “the ultimate robotics and AI starter kit.”
Aside from the data center-on-a-chip and robotics kits, NVIDIA also announced workstation chips, the RTX A6000, and the A40, which will be widely released in 2021.
