In the latest episode of Dinis Guarda YouTube Podcast, Erik Pounds, Director of Product Marketing at NVIDIA, discusses the impact of the technological advances made by NVIDIA in AI and accelerated computing. The podcast, powered by Businessabc.net and Citiesabc.com, also highlights NVIDIA’s innovations and discusses the applications of their GPUs in AI and data processing.
Erik Pounds is the Director of Product Marketing at NVIDIA, where he plays an important role in promoting the company’s advanced technologies. NVIDIA is a leader in developing powerful computer chips called GPUs, which are essential for artificial intelligence (AI) and data processing. These GPUs help computers run many tasks at once, making them ideal for training AI models and handling large amounts of data. Erik’s work focuses on showcasing how NVIDIA’s technology can help businesses and researchers make the most of these innovations.
“NVIDIA, as you know, is a semiconductor company or they bucket us that way. Well, the semiconductor, or the chip is just the bottom layer of the stack, it’s a very important layer of the stack, but it’s the bottom layer of the stack.
What sits above that are acceleration libraries to take advantage of that amazing computing power and then what sits above that are our domain specific software to apply that accelerated computing across various industry applications. Then, as you go further up the stack, you eventually end up getting to a specific application that takes advantage of everything that is below it and you do see NVIDIA climbing up the stack.
As we climb up the stack, we ensure that it’s integrated with and it’s compatible with all our partners and to make accelerated computing easier to use.”
Accelerated computing at NVIDIA
Erik explained NVIDIA’s approach to addressing the evolving needs of businesses with advanced technology:
“Every business is trying to figure out how to adopt these new generative AI capabilities to reach new levels of productivity and efficiency. NVIDIA is seen definitely as a technology powerhouse.
NVIDIA is focused on accelerated computing, especially in business, taking some computing tasks that are common and accelerating those to make them more efficient and drive cost down. Accelerated computing is a full stack challenge.
We’re making a lot of investments to make accelerated computing easier to use. We’re trying to make this full NVIDIA stack as simple as possible where you can just code to a very easy-to-use API. When you look at our products, we keep our platform as open as possible so you can engage with NVIDIA at any layer of that stack.”
Enhancing large language models with proprietary data
During the interview, Erik emphasises the importance of integrating proprietary data into AI models and make it more active:
“I think this applies to governments, nations, communities, and societies just as well as it applies to businesses. Whether you’re a nation or a business, you have information and knowledge that is close to you, often proprietary to you. So if you’re going to deploy an AI to work on tasks or generate new knowledge on behalf of your community, organisation, or business, it should be the most knowledgeable to you.
If we go back a decade, a lot of data was cold. It was sitting on some storage system, in the corner. If you had to access it, you had to go talk to somebody and they had to pull the data. Now, the data is more online, more active, it’s more cloud native. So, essentially a lot of data is now just an API call away.
During the last decade there has been a lot of work into making data more active. So, now it ensures the right applications and the right users have access to that data and now we have these information retrieval or pipelines that can can connect these amazing large language models and other types of generative AI models with this data to produce amazing content, provide very valuable insights into your into whatever you’re trying to do.
Over the last two years, we have seen a lot of research into how we augment the knowledge of these AI models with external data, or proprietary knowledge.”
The GPU technology of NVIDIA
GPUs by NVIDIA are designed to handle intensive parallel processing tasks, making them ideal for AI, machine learning, and data analytics.
Explaining the concept of GPUs and parallel processing, Erik tells Dinis:
“Traditional Computing is very serial: You do one task at a time and you do those as quickly as possible with GPU Computing.
With parallel computing, you’re doing many things in parallel at a single time and it just turns out that tasks like processing a large amount of data or training an AI model or running that AI model and production across thousands and thousands of users are just amazing applications for to accelerate computing
Although it’s a new type of computing for many, it has been worked on for multiple decades by some amazing innovators and a very large ecosystem. So, for a lot of development teams and businesses and application owners that want to take advantage of accelerated computing, it’s now right at your fingertips and it’s much easier to leverage.
We’re trying to make it so the important things you need to leverage are just a simple API call away through a microservice or a cloud service.”
Its cutting-edge architectures, such as Ampere and Hopper, power GPUs that excel in both training and inference of AI models. Technologies like NVIDIA CUDA enable developers to leverage GPU acceleration easily, enhancing the capabilities of software in fields ranging from scientific research to entertainment.
NVIDIA’s innovations in ray tracing and DLSS (Deep Learning Super Sampling) have significantly advanced the realism and performance of graphics in gaming and professional visualisation.
Highlighting the potentials of NVIDIA’s GPU technologies, Erik further adds:
“The impact that Hopper made on generative AI is tremendous. And then, Jensen explains Blackwell as the first platform built for the ground up for the era of generative AI. This is just tremendously exciting, the progress that is being made from platform to platform.
If you adopt NVIDIA software and build your applications on top of our platform, you have a track record of forward and backwards compatibility. This makes it easy for you to adopt these solutions as well as run them across different types of systems.”
NVIDIA NIM and NVIDIA AI Foundry
NVIDIA NIM, part of NVIDIA AI Enterprise, offers a suite of accelerated inference microservices designed to run AI models on NVIDIA GPUs across various environments, including the cloud, data centres, workstations, and personal computers. NIM simplifies the process of implementing scalable AI inferencing, enabling organisations to develop powerful AI applications like chatbots, AI assistants, and more with minimal code.
Recently, NVIDIA launched the NVIDIA AI Foundry with NVIDIA NIM™ inference microservices, designed to empower enterprises and nations to develop bespoke ‘supermodels’ designed to specific industry needs using the latest Llama 3.1 collection of openly available models.
Commenting on this part of innovation, Erik tells Dinis:
“It’s open, and can generate data that you can own and leverage in a commercial type application. One of the new interesting things about this model is that you can use it as it is and customise it.
One of the new applications is synthetic data generation. So, as you’re building a custom model, you can take your proprietary data and use that to alter the model to have the knowledge of your domain, or the knowledge of your business, or you can even trade it, do certain skills in a lot of scenarios.
Nvidia AI Foundry gives you a service that you could use to take your proprietary data, bring that into The Foundry and customise these powerful models like Llama. We have evaluation tools as well so that you can evaluate the model across, whether it’s industry benchmark or cross-benchmarks that are specific to your application. So, you ensure that you are building the best possible model and guardrail it so the model behaves in the way that you need it. We output models out of The Foundry in a stack that we call NVIDIA NIM that takes an optimised runtime and an easy to use API, wraps it all up into a container that you can deploy anywhere in the world, in the cloud, at the edge, in a device, on a workstation, and so on so forth.
So, now it makes building custom generative AI, its deployment, and use a whole lot easier.”
With a driving passion to create a relatable content, Pallavi progressed from writing as a freelancer to full-time professional. Science, innovation, technology, economics are very few (but not limiting) fields she zealous about. Reading, writing, and teaching are the other activities she loves to get involved beyond content writing for intelligenthq.com, citiesabc.com, and openbusinesscouncil.org
