Dell Technologies, NVIDIA partner on Generative AI through Project Helix

Michael Dell thinks the commercial potential of Generative AI is greater than PCs and smartphones, although what the market seems to want from it, and what influencers think the market wants, are not the same thing.

Michael Dell, at Dell Technologies World 2023

LAS VEGAS — Today, at Dell Technologies World, Dell Technologies and NVIDIA are announcing Project Helix,  a joint initiative which combines the two companies’ technologies to make it easier for businesses to build and use generative AI models on premises. It will result in a series of full-stack solutions with technical expertise and prebuilt tools based on Dell and NVIDIA infrastructure and software.

“Generative AI is a really hot topic of discussion these days, ever since the advent of ChatGPT,” said Varun Chhabra, Senior Vice President, Product Marketing, Infrastructure Solutions Group at Dell. “Now it just it seems like there’s a new generative AI solution that everyone’s talking about every single day. Not surprisingly, enterprises are now actively looking at how they can put Generative AI to use for their businesses, whether it’s to improve their back end operations to transform their customer experience, etc., etc. There are so many different use cases here.”

Chhabra said that what Dell is finding as they talk to enterprises about generative AI  is that enterprises are exploring generative more and uncovering very, very specific needs.

“First and most important, enterprises need to be need to be able to trust the output of generative AI applications, especially as they put it in front of customers and partners,” Chhabra indicated. They need to be able to trust that the output is going to be accurate. It’s going to be appropriate and following enterprise guardrails and is consistent with their own brand voice.”

Chhabra emphasized that future Generative AI initiatives will need to be more specialized than early iterations of the technology like ChatGPT.

“Given that a lot of these models or the models that are driving these generative AI technologies are going to be using proprietary data or high-value data sets that customers organizations have, they need to make sure that their data is secure,” Chabbra said. “They also need to make sure that they don’t fall afoul of any privacy or data regulations specific to their industry or to their local geographies as customers are going to be using their own proprietary data here. It’s very clear that there is going to be a big market for purpose-build generative AI models that are very different from more general purpose models like ChatGPT.”

Chhabra said that Dell is already starting to see is enterprises creating models or tuning existing models for very specific domain-specific use cases.

“It could be virtual sales assistance. It could be marketing, content creation, or for a simplified back end operations, or technical support for customers,” he stated. “We think that the most efficient way for enterprises to do this is to actually take models that are either built specifically for those needs or tuned with existing models with some of this data for that specific need. So purpose built models are going to play a huge role in enterprise adoption of generative AI. Agility and time to market is also a key concern for enterprises. They want to be able to get to market early.”

Chhabra emphasized that this is clearly going to be a competitive differentiator for enterprises that get out early and are able to transform their customer and internal or back office operations faster than their competitors.

“Being able to train these models quickly, not having to wait for months before the model is updated, and having that positive cycle of continual reinforcement happening for these AI models is going to be very, very important,” he said.

“And finally, as with any transformative technology, we certainly think generative. AI is a transformative technology, so expertise and know-how are going to be a bottleneck at first,” Chhabra said. “Technology doesn’t mean that you and customers, have the know-how how to either build these models from scratch or take existing models, and then fine tune it with their data, or once you have the model ready, actually deploying it at scale. So it’s very clear to us that enterprises have these very, very unique needs when it comes to generative AI, and what they’re really looking for are solutions that are going to help them do that.”

Having set the stage for the announcement, Chhabra announced that he was very excited to be able to announce that NVIDIA and Dell technologies are collaborating together on an initiative that they are calling Project Helix

“Project Helix is really about helping enterprises adopt generative AI, and put them to use for their businesses as quickly and efficiently as possible,” he said.

“What Project Helix will do is deliver full stack scalable solutions for our customers, using the best of Dell infrastructure, whether it’s our compute infrastructure or storage infrastructure, our software, our expertise in delivering AI solutions for customers as well as NVIDIA’s market leading accelerators. their unique AI software capabilities as well as foundational models and the unmatched experience that NVIDIA has in delivering AI for enterprises.”

Dell PowerEdge servers, such as the PowerEdge XE9680 and PowerEdge R760xa, that are optimized to deliver performance for generative AI training and AI inferencing will work with NVIDIA H100 Tensor Core GPUs and NVIDIA Networking as the infrastructure backbone for these workloads. Project Helix will also include expertise and guidance for enterprises that are looking to deploy these solutions.

“It will really help our customers,” Chhabra said. “It’ll  assist them in every single stage of whatever they are in the Generative AI lifecycle, whether they are  building a model from scratch, or taking a foundational model and tuning it with their high value data  or proprietary data.  It could be that they’ve got the model figured out, but they need to deploy Generative AI applications using that model at scale or in production ,or they need support with their existing applications, whatever. Wherever a customer is in their Generative a life cycle, the solutions coming out of Project Helix and the expertise and blueprints will help our customers wherever they are.”

Chabbra also stressed that Project Helix solutions will have capabilities within them to make sure that customers are able to deliver.

“Organizations will be able to deliver trusted AI outcomes with integrity, making sure that the answers are as factual as possible, hallucinations are reduced to a minimum, and that there are enterprise guardrails, and solutions to make sure that that these generated AI applications don’t veer into inappropriate territories,” he said.  “The solutions will also have capabilities to ensure compliance with data security as well as privacy regulations.”

The first of the solutions that are going to be delivered under Project Helix will be available starting in June 2023. We will have more solutions coming quickly thereafter. They will be available through traditional channels and APEX flexible consumption options, beginning in July 2023.

Dell’s senior executive team thinks that while the opportunities with this are enormous, there has been a lot of hype which has confused what customers really want from Generative AI to a degree.

“This is a massive unlocked source of data,” said Michael Dell, Dell Technologies CEO and Chairman. “I think it will accelerate the economy. It is bigger than PCs and the smartphone in terms of impact. For us, it’s a very comprehensive reimagining in terms of our business. Every single product line we are developing has AI embedded in it in some way.”

At the same time, Dell emphasized that customers are looking at more limited use cases for Generative AI than many have implied.

“I don’t think one model to rule them all will happen,” he said. “They want it with specific data, not everything that’s ever been written. That’s irrelevant, and likely counterproductive. On the show floor here, the use cases in many instances are very narrow.”

Jeff Clarke, Dell’s vice chairman and co-COO, said this isn’t really what Dell wants to do either.

“We don’t want customers to train a big global AI model, a large language model,” he stated. “We want to train them around what we do.”