One of the most-awaited developer conferences, Nvidia GTC, is just around the corner. Scheduled between March 20 and 23, this year’s GTC will focus on the latest advancements from Nvidia in areas of AI computing systems, generative AI, industrial metaverse, and robotics. 

The Nvidia GTC would feature a keynote by chief Jensen Huang and hold over 650 sessions with researchers, developers, and industry leaders in almost every computing domain. The keynote announcement by Huang will be live-streamed on Tuesday, March 21, at 8:30 PM IST (8 AM PT). 

The event will feature a special fireside chat with Huang and OpenAI co-founder Ilya Sutskever, alongside talks by DeepMind’s Demis Hassabis, Stability AI’s Emad Mostaque, and many others. 

Click here to register. 

Over the past few months, Nvidia has come to be synonymous with AI. This is evident in the company’s latest quarterly earnings report, where AI accounted for much of its revenue. AI for enterprise, the company’s new business model, leaves the air thick with anticipation of what more can Nvidia bring into this segment. 

Generative AI

Previously, CEO Jensen Huang had noted how generative AI will be the front and centre at its GTC event. Nvidia has already offered some cues on what to expect, with NeMo and BioNeMo. Currently, in early access with customers, these pre-trained generative AI model layers are designed for enterprise customers looking to develop proprietary generative models and services to enhance their businesses. 

Doubling down on the strong revenue numbers from hyperscale customers like AWS, Azure, Oracle, and GCP, Huang called Nvidia DGX Cloud “the fastest and easiest way to have your own DGX AI supercomputer”, which will help deploy generative AI applications for enterprise on Cloud. 

At GTC 2023, there will likely be a display of generative AI used in various industry vertices such as healthcare, and biology, etc. This will be part of Nvidia’s AI cloud service offerings, which will allow enterprise customers to be able to access full-scale AI computing across their private to any public cloud. The AI as a service model for AI computing, training, as well as, for software layers is in line with the ‘generative AI for enterprise’ business model for enterprise. 

Furthermore, there is also 3D rendering, which is central to Nvidia’s overall plans for metaverse. Last year Nvidia had released models like get3D and Magic3D, which could render 3D models from 2D images using a pre-trained text-to-image diffusion model, which is better than the existing benchmarks. The company also released 3D MoMa, which allows creators to edit 3D representations.  

Alongside this, Nvidia has been doing research in areas that include training AI agents in video game environments, diffusion models for high-quality image generation, 3D construction with approaches like latent point diffusion modelling and PeRFception, a variation of NERF, which can convey 3D information better. Research has also been done in creating animation from a set of source images. 

AIM Prediction: AI is showing no signs of slowing down. There are constant updates by OpenAI, Meta, Google, and the likes on the work being done in text-to-video generation, multimodal AI, consistency models, approaches to training on Indic language datasets, and much more. In this light, Nvidia will likely be headlining some big announcements not only on 3D modelling, but also in other generative AI applications. 

At CES 2023, Nvidia also announced the Omniverse Avatar Cloud Engine (ACE), a set a cloud-native AI microservices – which include the likes of text-to-speech and 3D animation – to build and deploy interactive avatars. We will likely see many more additions to the services considering how Nvidia is keeping up pace. 

AI Computing

Previously, at the 2021 GTC event, Nvidia detailed the successor to Bluefield-3, a 400-gigabit data processing unit, to be released by 2023. Bluefield-4, with a traffic throughput of 800-gigabit per second, will have an AI acceleration hardware added to it. This AI-assisted DPU, Gazettabyte reports, will support tasks such as video analytics, 5G, and robotics. We can expect Nvidia to give an update on Bluefield-4. 

Meanwhile, the Nvidia Hopper H100, announced at GTC 2022, is now in full production. A follow-up to the very successful A100 accelerator, H100 has already exceeded the A100 in revenue. 

AIM Prediction: It’s likely that Nvidia will provide use cases of how the H100s will be used in training and inferencing of transformer-based large language models, and how enterprises will use language models, whilst keeping the data secure.  

Additionally, as per Liftr Insights, the first sighting of the H100s used in the public cloud is yet to be seen. In the case of A100 as well, it took up to a year for it to appear in public cloud after the initial production. We can expect Nvidia to accelerate the deployment of H100s in the public cloud, given their focus on enterprise.  

Omniverse

No conversation on Nvidia goes without mentioning the Omniverse – it is that big to its plans. Earlier this year, Nvidia announced the Omniverse Avatar Cloud Engine (ACE), which will provide generative AI services to create digital avatars in the virtual world. 

The Nvidia Omniverse platform is already being used for several digital twin applications such as Mercedes-Benz and BMW in their manufacturing and assembling facilities. The company also revealed it will collaborate with Fujitsu to develop a new AI-on-5G solution, combining 5G vRAN, edge AI and digital twin workloads on an all-in-one, hyperconverged and GPU-accelerated system. 

At GTC 2023, we can expect Nvidia to announce more industry collaborations to facilitate digital twins and AI-on-5G for Metaverse applications and computer vision. 

Moreover, at the recent CES event, Nvidia also announced that Isaac Sim, its robotic simulation application, will be available on the Cloud, while also upgrading Isaac Gym for reinforcement learning and Isaac Cortex for collaborative robotic programming. It also introduced Isaac ORBIT for simulation operating environments and benchmarks for robot learning and motion planning. 

We will also probably see announcements related to delivering AI robotics at scale, as well as development of AI models with faster learning from human behaviour. 

Hardware, gaming and others

Nvidia has been making waves in the PC GPU market with the variants, RTX 4090, 4080, 4070, 4060, and 4050, starting to ship and power some of the best gaming laptops. It is highly possible that Nvidia will announce GeForce RTX 4070 since the rumours have it being released in April. The GPU will be a blend of optimal price and performance, likely providing a mid-range option for end-users. We can expect more updates on the release of 4060 and 4050 to arrive.  

Last year, Nvidia announced the release of CUDA’s software toolkit update 12.0, which focuses on new programming models and accelerating processing capabilities. The company has also leapfrogged into building a platform for hybrid classical-quantum computing. The QODA architecture which powers this platform will be crucial to achieving all future quantum advantage. There are more reasons to believe that Nvidia will announce potential collaborations and use cases with QODA. 

On the autonomous technology front, Nvidia Drive, an end-to-end modular development platform, announced at GTC 2021, is leveraged by companies like Arrow, Foxconn, and Polestar, for designing autonomous vehicles. With a surge in automobile companies looking up to driver-assistance systems for future development, GTC will also involve more such partnerships and customer case studies in this area.