Firstly, the BlueField family has been expanded with the introduction of the BlueField-2X, which combines "all the key features" of the BlueField-2 with an Nvidia Ampere GPU. Using a PCIe Gen 4 interconnect, the GPU can be employed to add AI functionality to data centre security, networking and storage tasks.
For example, real-time security analytics could identify abnormal traffic indicating theft of confidential data, encrypt traffic analytics at line rate, identify malicious activity through host introspection, and dynamically orchestrate security responses.
Another possibility is to use video analytics to block inappropriate content at the time it is uploaded.
BlueField-2X GPUs are under development and are expected to ship during 2021.
Secondly, Nvidia has announced the availability (but only to early access partners at this time) of the DOCA SDK. Just as Nvidia's CUDA helps developers build GPU-accelerated applications, DOCA helps them build applications on DPU-accelerated data centre infrastructure services.
Thirdly, the BlueField roadmap has been extended to the BlueField-3, -3X and -4.
Expected in 2022, the BlueField-3 will support data transfers at up to 400Gbps, and the -3X will add a GPU with 85 times the AI performance of the -2X.
The 2023 followup will be the BlueField-4, which will put the GPU on the same chip as the DPU (the two functions are separate on the -3), and also include tensor cores for even higher performance. The BlueField-4 is rated at 1000 SPECINT compared with 70 SPECINT for the BlueField-2X and 350 SPECINT for the BlueFIeld-3X
Nvidia intends to continue making DPUs without GPUs for customers who do not need the AI functionality.
"The DPU applies to every workload," said Nvidia head of enterprise computing Manuvir Das, predicting it will be a part of all data centre servers in the future.
Asus, Atos, Dell Technologies, Fujitsu, Gigabyte, H3C, Inspur, Lenovo, Quanta/QCT and Supermicro have all announced plans to integrate Nvidia DPUs with their enterprise servers, and VMware, Red Hat, Canonical and Check Point have all announced support for BlueField-2 DPUs.
Edge AI. The large and growing number of IoT devices means there is a pressing need to apply AI at the edge, observed Nvidia vice president and general manager of edge computing Deepu Talla.
So the EGX range will expand to include a card that combines Ampere GPU and BlueField-2 DPU capabilities on one PCIe card.
To help deploy and manage EGX servers, Nvidia has announced Fleet Command, a hybrid cloud platform that centralises the management of servers spread across vast areas.
Initially available to participants in an early access program, Fleet Command offers one-touch provisioning, over-the-air software updates, remote management and detailed monitoring dashboards.
Priced at just US$59, the Nano 2GB is "the ultimate AI robotics starter kit," according to the company. The main differences are that it has half as much RAM as the original version, one camera port instead of two, and three rather than four USB ports. But it is still supported by the JetPack SDK and the Cuda-X stack.
It is expected to appeal to individual developers and secondary and tertiary students who want to learn about AI, said Talia.
Nvidia is providing a free AI certification program using the Jetson Nano 2GB, he added.
Jetson Nano 2GB will ship at the end of October.
Collaboration. Nvidia Omniverse has moved into open beta.
Originally conceived as a way to support internal collaboration, Omniverse can be used by individuals to integrate the diverse tools they use (eg, Blender, SketchUp) as well as Nvidia technologies such as AI Pose and Audio2Face to generate photorealistic images in real time.
It also allows the use of those same tools and technologies in a collaborative space. So architects, animators or engineers can work together on 3D models just as easily as people can co-edit a document or spreadsheet.
"Physical and virtual worlds will increasingly be fused," said Nvidia founder and CEO Jensen Huang.
"Omniverse gives teams of creators spread around the world or just working from home the ability to collaborate on a single design as easily as editing a document. This is the beginning of the Star Trek Holodeck, realised at last."
Early adopters include visual effects company Industrial Light and Magic, architecture and engineering firm Foster + Partners, and telecommunications company Ericsson.
Nvidia Maxine is a cloud-native streaming video AI platform for video conferencing providers. It provides a range of capabilities including AI-based compression (to one-tenth of H.264 requirements), face alignment, gaze correction, conversational avatars, virtual backgrounds, de-noising, and auto framing.
The bandwidth reduction is achieved by analysing the key facial points of each person and then intelligently re-animating the faces in the video on the other side of the call.
The work is done on GPUs in the cloud, so users do not need special hardware.
"We're really proud to be launching it," said Nvidia head of worldwide developer relations and general manager of media and entertainment Richard Kerris.
Healthcare. Nvidia Clara Discovery is a platform for computational drug discovery, said Nvidia vice president of healthcare Kimberley Powell.
Available immediately, it combines accelerated computing, AI (including deep learning) and natural language processing with the aim of improving the efficiency of drug discovery.
The natural language capabilities provided by Nvidia's BioMegatron are "transformative," she said, as the language model is able to extract information from scientific literature and electronic health records.
Healthcare company GSK and Nvidia have established an AI hub in London to apply the technology to design and develop medicines and vaccines.
"Because of the massive size of the datasets we use for drug discovery, we need to push the boundaries of hardware and develop new machine learning software," said GSK senior vice president and global head of AI and ML Kim Branson.
"We're building new algorithms and approaches in addition to bringing together the best minds at the intersection of medicine, genetics and artificial intelligence in the U.K.'s rich ecosystem. This new partnership with NVIDIA will also contribute additional computational power and state-of-the-art AI technology.
Nvidia's Cambridge-1 supercomputer, based on the DGX SuperPod, is being built in London. It would be number 29 on the current Top500 list, and in the top three of the Green500 thanks to its 80 DGX A100s giving 400 AI petaflops. Nvidia is investing around £40 million in Cambridge-1, which will be part of the company's already-announced AI Centre of Excellence in Cambridge, which will also feature a new Arm-based supercomputer.
Along with GSK, Cambridge-1 will be used by AstraZenica, as well as researchers from Guy's and St Thomas' NHS Foundation Trust, King's College London and Oxford Nanopore.
"The use of big data, supercomputing and artificial intelligence have the potential to transform research and development; from target identification through clinical research and all the way to the launch of new medicines," said AstraZeneca head of data science and AI James Weatherall.
UK Secretary of State for Health and Social Care Matt Hancock said "Today's announcement from Nvidia is an exciting moment for the UK's world-leading healthcare industry and a tremendous vote of confidence in the UK as an international centre for research, AI and innovation.
"Accelerating drug discovery has never been so important and it is investments like this that can make a real difference in our fight against countless diseases. I care about technology because I care about people and Nvidia's new supercomputer will aid the UK's best and brightest to undertake research that will save lives."
Once established, startups and other researchers will be invited to use Cambridge-1.