Dell pumps out reference designs, plumps services, to bring AI on-prem
Go ahead, spend up big. But be warned: many models 'will never make it to widespread use because they are quite frankly rubbish'
Dell can't wait to tap into the frenzy of spending on … er … interest surrounding generative AI, so has cooked up hardware bundles and consulting chops to bring it on-prem.
The Texan IT titan on Monday announced a three-pronged offer comprising:
- Validated designs for inference systems using Nvidia accelerators;
- Professional services that will see Dell help customers inject generative AI into their workflows and applications;
- GPU-equipped Precision workstations – the desktop tools needed to work on AI.
Varun Chhabra, senior veep of Dell's Infrastructure Solution Group (ISG), told The Reg Dell sees plenty of orgs have "a unique sense of urgency … in terms of adopting and or trying out generative AI and applying that for the benefit of businesses."
The Validated Designs therefore use Dell kit such as the PowerEdge XE9680, announced earlier this year, which can be configured with up to eight Nvidia H100 GPUs, plus Dell storage, with software comprising Nvidia's AI Enterprise suite and NeMo framework for large language models (LLMs).
Chhabra said the reference designs include detailed specifications and guidelines around how to tune the infrastructure management software that's included.
The rigs were jointly engineered with Nvidia and aimed at taking existing models that customers have been either building from scratch or tuning with their own data and scaling them up and putting them to work.
Which sounds a lot like what hardware makers always do when a hot new workload comes along: build bundles so that users can use a LEGO-like approach to build infrastructure. That's faster than figuring out bespoke rigs.
How do you integrate generative AI into existing processes for maximum effect? How do you automate complex processes?
Chhabra said Dell's approach will be handy for "anything that requires natural language generation – chatbots and virtual assistants are another example. Software development is another popular use case, as well as marketing and content creation."
The veep added that Dell's AI-related professional services folks stand ready to help with AI advice and expertise whether you're just starting out with a binary brainbox or building a big one. Consultants, eh? Always eager to help!
Dell is especially keen to help with data management and business integration.
"One is about how to handle your data, what data sources to use, as well as the security of that data, and making sure that it doesn't fall in the wrong hands," Chhabra said. "The other aspect is, how do you integrate generative AI into existing processes for maximum effect? How do you automate complex processes? How do you connect the generative AI infrastructure and software with existing applications?"
Chhabra acknowledged that while many users believe there to be a lot of potential gains from generative AI, an equal number perceive risks. And they are quite right, as Reg readers are no doubt well aware.
"We all know the risks of generative AI, of sharing company data and tools that have some tools that have mixed privacy records and best unclear rules around what happens with any data that's entered,” Chhabra said.
And this is why Dell wants to set itself up as a trusted partner that can help customers with generative AI initiatives and make the most of any high value datasets that they have in their business.
The Dell Validated Design for Generative AI with Nvidia is available globally through Dell's traditional channels and its APEX IT-as-a-service scheme.
The Big D has also tweaked its workstations so developers can work on AI without delay. The Precision 7960, 7865, and 5860 models can now fit up to four Nvidia RTX 6000 GPUs in a single machine, which Chhabra claimed will allow them to run AI software frameworks 80 percent faster than the previous generation of Precision workstations.
- Your boss tells you to build some generative AI. Dell and Nvidia are already knocking
- AI mentioned 175 times during Microsoft's Q4 earnings call
- Dell down under dinged for dodgy display discounts
- To quench AI's thirst, the way we build, operate datacenters needs to change
Is Dell doing anything truly interesting here?
We asked some experts.
"There is hype like with any new technology, but actually there's a lot of purpose and projects," Canalys chief analyst Matthew Ball told us, adding that he estimated that AI services (consulting, design, implementation, integration etc) would hit $15 billion in value in 2023.
"This is being driven by many of those organizations that failed to embrace digitalization early on, which was highlighted during the pandemic, and are now rushing to adopt generative AI to avoid being left behind by their competitors," he said.
Dell is too big and too smart to miss out on that rush.
Omdia Chief Analyst Roy Illsley told us that generative AI is now becoming table stakes," so every vendor needs an offering that enables customers to build bespoke models based on their own data.
"I think if this technology follows the typical hype cycle we are still on the way up to the top of the curve," Illsley said, but added that he believed the decline would be smaller than for other technologies and will pick back up quicker. AI technology has been around for years, and the LLMs – or foundational models – have become a better way for people to access AI.
"Yes, we will see more of these models appear for people to use, and many of these will never make it to widespread use because they are quite frankly rubbish. So, the next 12 months will see a flood of models, people using them and either being happy they work, or disillusioned because the results are rubbish," he explained.
There are sufficient good examples that the market will quickly weed out the rubbish and people will gravitate to the good known working models, Illsley told us. ®