The past week was Nvidia’s GTC 2023 developer conference, which featured a number of announcements including the metaverse, massive language models, robotics, automobiles, and more.
Before we discuss what is happening and what is new in technology this year, let’s briefly discuss the purpose of this conference.
A technology business called Nvidia Corporation is well renowned for creating graphics processing units (GPUs). Jen-Hsun “Jensen” Huang, Curtis Priem, and Chris Malachowsky formed the business in 1993. Its headquarters are in Santa Clara, California.
The creators of Nvidia anticipated that a dedicated GPU would be necessary for the advancement of computer graphics. Computer games used to be fully CPU-based. Yet, gaming technology was improving and gradually switching from MS-DOS to Windows. The CPU’s math coprocessor was just insufficient to handle the amount of floating-point math work required for graphics, especially 3D graphics.
Nvidia has grown into high-performance computing (HPC) and artificial intelligence since becoming the leading supplier of graphics chips for video games (AI). These various computational processes are performed on the same gaming CPUs.
The corporation is anticipated to report $26.9 billion in revenue for 2022.
History of Nvidia
When Nvidia made their entry into the GPU industry in the early 1990s, it was already highly crowded. ATI Technologies, Matrox, Chips & Technology, S3 Graphics, and 3Dfx were competitors. With the introduction of the GeForce card in 1999, Nvidia distinguished itself from its rivals. Compared to other manufacturers, it had more sophisticated 3D graphics and lighting techniques.
Nvidia tried to increase the usage of its GPU technology as the GPU market solidified around it and ATI, which AMD purchased in 2006. The business created CUDA in 2004, a C++-like language used to program GPUs.
Gamers typically use 3D graphics libraries, but CUDA enables programmers to write code straight to the GPU. This made it possible for them to create enormous parallel programs to carry out high-performance floating-point operations, such as simulations, visualizations, and other applications requiring the parallel processing of vast volumes of data.
Nvidia made a strong attempt to get CUDA taught in colleges after announcing the programming language in 2006. More than 200 colleges worldwide offer CUDA courses. This has aided in the growth of the Nvidia programming workforce.
Nvidia launched the Tegra family of systems-on-a-chip (SoC) in 2008; Tegra was primarily offered to automakers for in-dash systems. Tegra coupled an Arm Processor with a scaled-down Nvidia GPU. Nintendo, however, decided to use the Tegra for their portable Switch platform in 2017.
During the majority of its existence, Nvidia strategically purchased tiny businesses. Yet in 2019, the business paid $7 billion for networking expert Mellanox Technologies, which was thought to be Intel’s target for purchase. Data processing units (DPUs), which are chips used in SmartNICs, are Mellanox’s area of expertise. More intelligently than a typical networking chip or CPU, smart NICs can route data. As the CPU continues to do its primary task of processing data, the goal is to take over the processing of networking data. Massive data sets must be moved for HPC and AI, therefore Nvidia’s GPU processors gain from intelligent data processing and routing.
Nvidia tried to buy Arm Holdings, a company that designs CPUs, for $40 billion in 2020. Allegations that Nvidia would prefer Arm licensing led to instant criticism of the purchase in Arm’s native United Kingdom. Despite vigorous denials, Nvidia was unable to overcome the resistance and, after 18 months of work, abandoned the agreement.
Nvidia and AMD encountered difficulties in 2016 when cryptocurrency miners discovered that GPUs were remarkably effective at mining cryptocurrencies like Bitcoin. The entire supply of GPU cards from both Nvidia and AMD was used to build large systems. This resulted in a GPU card shortage, which was made worse by COVID-19’s supply limits and shortages.
NVIDIA GPU Technology Conference (GTC)
A global AI conference for developers, NVIDIA GTC (GPU Technology Conference) brings together programmers, engineers, researchers, inventors, and IT specialists. AI, computer graphics, data science, machine learning, and autonomous machines are the main topics. Jensen Huang, the CEO and founder of Nvidia, gives the opening keynote address at each conference. There are then a number of seminars and lectures with international professionals that follow.
It began in 2009 in San Jose, California, with an initial emphasis on the potential for using GPUs to potentially solve computer problems.
The conference’s emphasis has recently changed to a variety of AI and deep learning applications, such as self-driving cars, healthcare, high-performance computing, professional visualization, and Nvidia Deep Learning Institute (DLI) training.
Almost 8,400 people attended GTC 2018. The COVID epidemic of 2020 forced the conversion of GTC 2020 to a digital event, which attracted about 59,000 registrations. The Nvidia Omniverse real-time rendering tool was used to create a piece of the 2021 GTC keynote, which was broadcast on YouTube on April 12. Because of the event’s photorealism, which featured a model of the CEO Jensen Huang, news organizations claimed they were unable to tell that a component of the speech was CGI until it was later made clear in a blog post on August 11.
The GPU Technology Conference, also known as GTC, was launched by NVIDIA on Monday, March 20. NVIDIA is continuing, as in previous years, its assault on the AI hardware industry. The hardware manufacturer has shown a number of products at NVIDIA GTC 2023 in addition to its recognizable computing and processing units.
This year, accessibility appears to be the unofficial theme. We are at the iPhone moment for AI, NVIDIA CEO Jensen Huang declared during his keynote speech, citing ChatGPT’s enormous success and its quick acceptance and integration across multiple existing goods and services.
As generative AI apps and enterprise-grade cloud-delivered AI infrastructure appear to have blown open a new vertical, the company is making sure it’s AI technologies are available for it.
The GTC this year was, in my opinion, one of Nvidia’s best ones yet, and we are seeing Nvidia define its strategic position—which lies halfway between cloud service providers and SaaS providers—in front of us. Here I want to share some of the most important things I learned about Nvidia’s Omniverse technology, which promises to integrate and synchronize platform as a service (PaaS) and infrastructure as a service (IaaS) activities.
Nvidia Omniverse’s power
One of the most significant revelations made at GTC 2023 is that Microsoft Azure would host the Isaac Sim digital twin application within Nvidia’s Omniverse Cloud. Nvidia’s full-stack cloud platform for developing, deploying, and administering commercial metaverse applications is called Omniverse Cloud. Nvidia is extending access to its service portfolio, which includes Isaac Sim as well as Replicator and Drive Sim, by hosting them on Azure.
Although CAD software has been used for decades in product design and engineering, it only accounts for a small portion of the total resources and time spent on product development. Nvidia is providing a full-stack IaaS to businesses with Omniverse, reducing many obstacles and integrating all product workflows, including not only design and engineering but also programming, validation, production, and product service.
Every team participating in the life cycle of the physical product, including the design, engineering, and retail teams, can utilize the same CAD data and digital twin environment in this scenario. Some special capabilities are made possible by this digital infrastructure. In order to train robots and AI models in a controlled environment, developers can simulate factories and warehouses with Isaac Sim, while Drive Sim is utilized for platforms for ADAS and autonomous vehicles. Nvidia revealed that it is now supporting more sensors and lidar support for mimicking real-world performance in Isaac Sim at CES 2023 a few months ago.
AI inference platforms
Three computing platforms for AI inferencing, which uses trained neural networks to evaluate and predict outcomes, were announced by NVIDIA. They consist of the H100 NVL, DGX H100, and NVIDIA L4.
NVIDIA claims that the DGX H100 offers nine times the performance, twice as fast networking, high-speed scaling, and more.
The NVIDIA L4 is a universal accelerator that is 99% more energy-efficient than the fastest CPU platform and built for effective video, AI, and graphics. It can be used on a variety of servers and can produce 120x quicker performance than the fastest CPU platform.
The H100 NVL is built with an accelerated Transformer Engine and 94 GB of memory for real-time inference of huge language models. The Hopper architecture-based H100 NVL outperforms the A100 by 12 times in GPT3 inference, and it was first unveiled at NVIDIA GTC 2022. NVIDIA L40 is also used to create 2D and 3D images.
Microsoft (for the DGX H100) and Google Cloud are two early adopters (for NVIDIA L4).
DGX Cloud
The cloud version of NVIDIA’s DGX AI supercomputing service is now available. The next stage in providing AI supercomputing services to a larger consumer base is usually considered to be cloud-based data collecting, model training, and inferencing.
The unnecessary costs incurred by on-premise AI systems are eliminated by DGX Cloud. It makes using a web browser to access AI supercomputing infrastructure and related tools easier. Of course, the $36,999 price tag for one DGX Server box that has 640GB of memory and eight Nvidia H100 (or A100) GPUs might seem a little excessive.
Nonetheless, it is significantly less than the several hundred thousand dollars that would typically be needed for on-premise AI infrastructure. Additionally, it makes advanced model training at scale simpler without requiring a customer to have their own resources or infrastructure. It includes frameworks, rapid data science software libraries, pre-trained models, and NVIDIA AI Enterprise 3.1.
DGX Cloud is now accessible on Oracle Cloud, and it will eventually be available on Google Cloud, Microsoft Azure, and other clouds as well.
NVIDIA cuLitho
In addition to developing products with an AI focus, NVIDIA wants to hasten the development of a new generation of chips. A software library called NVIDIA cuLitho for computational lithography was created in cooperation with the top foundries TSMC, ASML, and Synopsys. They’ve all adopted cuLitho as well.
“The chip sector is the backbone of almost every other industry in the world,” Huang saidOpens a new window. As lithography approaches the physical limitations, fabs can enhance throughput, lower their carbon footprint, and lay the groundwork for 2nm and beyond thanks to NVIDIA’s launch of cuLitho and partnership with our partners TSMC, ASML, and Synopsys.
CuLitho, according to NVIDIA, will perform 40x better than current lithography methods. CuLitho fabrication equipment could manufacture 3-5 times as many photomasks per day with 9 times less power as the ones now in use. As a result, the business claimed that 40,000 CPU systems could be replaced by 500 NVIDIA DGX H100 systems.
According to Dr. C.C. Wei, CEO of TSMC, “the cuLitho team has made commendable progress on speeding up computational lithography by moving expensive operations to GPU.” “This breakthrough opens up new opportunities for TSMC to expand the use of deep learning and inverse lithography technologies in chip manufacturing, significantly advancing semiconductor scaling,” the researchers write.
Peter Wennink, CEO of ASML, stated that the business intends to incorporate GPU capability into all of its computational lithography software solutions.
Modulus framework
Nvidia has been well-positioned as it has expanded GPU technology beyond graphics cards and video games to include supercomputing, crypto-mining, and now enterprise AI.
Nvidia stated that its GPU technology has uses outside graphics processing even when it first emerged in the 1990s. At the time, standalone floating-point coprocessors were being included into CPUs, thus that was kind of outdated. The goal of Nvidia has always been to speed up computing.
The business took care to cultivate products with immediate applications while also sowing long-term ones. Software that facilitated such progress had to be continuously developed by Nvidia. The CUDA programming environment is an outstanding example.
It paid off when the business world focused on AI and deep learning, two fields in which GPUs’ extremely high memory bandwidth thrived. Nvidia chips have a considerable hardware advantage in data centers for AI. Little has slowed down the general GPU enthusiasm for deep learning, despite the abundance of specialist ASICs.
Nvidia keeps adding new features to its product lineup. One of Nvidia’s significant announcements at GTC was the open-sourcing of their Modulus framework, which was made in the background of various ChatGPT advancements.
Under the straightforward Apache 2.0 license, Nvidia is making its Modulus framework accessible for use with physics-ML. This could help with efforts to incorporate numerical simulation and physical modeling.
Why is that crucial? The family of neural operators created by Nvidia Modulus in recent years now includes physics-informed neural networks. Better outcomes in modeling physical systems may result from success with physics-ML. That would improve fidelity, for instance, for avatars across metaverses or digital twins across industries.
Nvidia and the upcoming computing tyrant
Huang is not the only one who believes that genuine quantum computing will not be practical for at least 10 years. Given Nvidia’s comments on the current limits of classical computing, that is not entirely unexpected.
Nvidia appears in a position to embrace quantum computing if and when it surpasses GPU-based devices because of its concentration on accelerated computation.
Every major enterprise technology trend of the past 20 years, including big data, virtualization, and databases, has taught us a vital lesson. The infrastructure that supports AI is no different. Standardization, cost control, and governance are necessary to obtain the traction and broad adoption that can spur innovation. Sadly, many businesses today have trouble with all three.
Many businesses use a diverse and expensive range of technology, models, and tools. Individual data scientists and engineers may have different preferences. There is therefore no constant experience. It might be challenging to collaborate across groups and scale prototypes into production.
Nvidia Piacasso
A cloud service called NVIDIA Picasso is used to create visual applications driven by generative AI. To generate picture, video, and 3D content from text prompts, businesses, software developers, and service providers can execute inference on custom models, train NVIDIA Edify foundation models on confidential data, or begin using pretrained models. On the NVIDIA DGX Cloud, the Picasso service simplifies training, optimization, and inference and is completely optimized for GPUs. In terms of governance, AI initiatives are all too frequently divided up among teams, organizations, and departments without any IT control. Because of this, figuring out what technology is being utilized where and if models, valuable IP, and customer data are secure and compliant is difficult or impossible.
NVIDIA Picasso creates new opportunities. Make your application stand out by using personalized generative AI models. Provide top-tier generative AI tools to internal teams and clients so they can begin their creative journeys. Use the robust inference optimizations in NVIDIA DGX Cloud to deliver interactive experiences while reducing cloud inference expenses. And utilize NVIDIA OmniverseTM to create 3D virtual environments that are lifelike by bringing all of your creative assets.
Employ cutting-edge Edify models that have already been trained, or train your own models with a specific dataset, host them on the NVIDIA DGX Cloud, and execute inference using APIs.
To speed up the workflows of the top creators and marketers in the world, Adobe and NVIDIA will jointly create generative AI models with an emphasis on responsible content attribution and provenance. Via Picasso as well as Adobe Cloud flagship products including Photoshop, Premiere Pro, and After Effects, these models will be jointly developed and commercialized.
With Picasso, a leading global provider of visual content, Getty Images is collaborating with NVIDIA to offer specially built picture and video generating models that were trained on fully licensed data. Using API requests, enterprises can access these models.
Together with NVIDIA, Shutterstock is creating models to create 3D assets that are trained on fully licensed Shutterstock content. These models can be used to create high-fidelity 3D objects from straightforward text instructions, which can then be used to 3D workflows for animation, game development, and other applications.
As you can see, Nvidia announced a lot of revolutionary and sector-reforming things at the event. We can be sure that with these things alone, the company has made a huge contribution to the evolution of the AI space and will be a pioneer in leading the industry forward into the future.
Naturally, as an Nvidia partner, we will be following the changes and potential collaboration situations very closely.
We already know that the release of Picasso will reenergize the Web3 gaming industry, which is a tremendous plus and a benefit for us as a Guild.
AMG is a decentralized autonomous organization, bridging the gap between players and web3 game developers, bringing the next billion users into the metaverse.
AMG acquires non-fungible tokens (NFTs) used in blockchain-based games and acts as a trusted party advocating for games in the metaverse, thus bringing more people in.
We are leveraging strong partnerships with games developers and Yield Guild Games to launch our organization to the forefront of the Central/Eastern European gaming market.
AMG DAO has been building a game-focused crypto wallet, guild management, and other tools to keep contributing to the ecosystem, in order to aid the creation of games with sustainable economies.
AMG DAO Social Media Links:
