Nvidia CEO Jensen Huang will unpack a range of new tech for 30,000 registered engineers at the company’s GPU Technology Conference (GTC) 2020 starting first thing Monday.
The news includes everything from a new JetsonNano 2GB developer kit for $59 to Nvidia’s plans to build the U.K.’s most powerful supercomputer for healthcare research.
Huang is also expected to describe in a keynote a new data center chip architecture and Maxine, a suite of AI videoconferencing software to enhance streaming video.
Here are some GTC highlights based on what Nvidia executives outlined in advance for reporters:
Healthcare news includes plans by year end for Nvidia’s new Cambridge-1, an AI supercomputer capable of 400 petaflops of AI performance to rank it 29th of the world’s most powerful computers. It is based on the Nvidia DGX SuperPOD system.
GSK and AstraZeneca will use Cambridge-1 along with researchers from university researchers.
Nvidia also said it has partnered with GSK and its AI group to apply computation to the drug and vaccine discovery process. It currently takes more than a decade and $2 billion to develop a new drug with a 90% failure rate, said Kimberly Powell, vice president of healthcare at Nvidia.
Enterprise Computing news includes a new category of processor called Data Processing Units (DPUs). Huang plans to discuss the new Nvidia BlueField-2 family of DPUs along with an Nvidia DOCA software dev kit for building apps.
Nvidia said that Asus, Dell Technologies, Lenovo and others have plans to integrate Nvidia DPUs into their servers. VMware has also announced Project Monterey to include Bluefield-2 with VMware Cloud Foundation.
RELATED: Nvidia partners with VMware for ease in managing AI infrastructure
DPUs have evolved from Nvidia’s recognition that data centers have become the new unit of computing, Huang said in a statement. Manuvir Das, head of Nvidia’s enterprise computing, said that DPUs can work alongside CPUs and GPUs and function in much the same way as NICs (Network Interface Cards). In fact, the BlueField-2 has all the capabilities of the Nvidia Mellanox-6 DXSmartNIC combined with powerful Arm cores, according to Nvidia documentation.
“DPU is a new concept that goes along with the CPU and GPU and puts all software defined infrastructure on a chip in the same server,” Das said. “The DPU belongs in every server going forward. Security is now table stakes. The GPU moves down to the DPU, freeing up the host CPU. The benefit is you get isolation and if the host is compromised then it doesn’t compromise the security layer.”
BlueField-2 is now sampling with partners and will go into production in late 2020. Das also showed a roadmap for a BlueField-4 DPU appearing in 2023 that is 600 times more capable than the Bluefield-2.
Edge AI news includes the new JetsonNano 2G dev kit, priced at $59, for teaching and learning AI and robotics. It will be available at the end of October through Nvidia distribution channels.
Nvidia also said its Nvidia EGX AI platform is expanding to combine Nvidia Ampere GPU and BlueField-2 DPU capabilities on a single PCIe card for enterprises to build a secure, accelerated data center.
AI-enabled collaboration news includes Omniverse Open Beta to let designers collaborate in real-time. The open beta begins sometimes this fall.
Nvidia also announced the Maxine platform to allow videoconference providers ways to improve streaming quality for people working and studying remotely. Streaming data will be processed in the cloud rather than on local devices to offer the benefit of AI effects such as super-resolution, noise cancellation, face relighting and gaze correction. Gaze correction helps give the appearance that person is looking at the camera instead checking the keyboard or looking in another direction.