Generative artificial intelligence (genAI) use cases are practically limitless — organizations see opportunities everywhere. But the industry is moving so fast, and in these early days, enterprises struggle with AI development, deployment, and management.

Continuing its hard push into software, Intel is looking to establish itself as a top provider across the AI stack. The company recently introduced its Tiber portfolio of software and hardware tools for AI and edge workloads, trust and security, and cloud optimization.

“Software is really exploding,” Caitlin Clark-Zigmond, Intel’s senior director of software and software-as-a-service (SaaS) marketing, told SDxCentral. “Customers are not only looking to Intel for the hardware solutions we provide — we can help them get value out of software.”

The Tiber software and hardware suite was initially introduced at Intel Vision Americas in April. The new offerings come as the company jockeys with other large incumbents to establish dominance in the rapidly growing AI space.

“We’ve moved out of do-it-yourself — we’ve simplified into ‘do-it-with-me’ or ‘do-it-for-me’ models,” Clark-Zigmond said. “Instead of having to figure it out yourself, you can prompt, and an answer spits out at you.”

Simplifying AI deployment and management It’s estimated that somewhere between 60% to 80% of AI models never make it into production — and those that do typically take several months to deploy.

“It’s hard to manage all those resources,” Clark-Zigmond said, adding that while organizations have access to open-source tools, they don’t always have the capabilities to take full advantage of them. “Just because you have the tools in the garage, doesn’t make you a plumber or electrician.”

Intel's Tiber Developer Cloud is designed to give organizations those tools: Developers can leverage both hardware and software to perform model training, optimization, inferencing, and deployment. The platform uses advanced CPUs, GPUs, and Intel Gaudi 2 AI accelerators, as well as open-source software tools, libraries, and frameworks.

The platform is based on an open-source software foundation with oneAPI, allowing users to learn, test, and run applications on Intel clusters. Developers can perform pre-launch development and testing and can build and deploy at scale. They also get hardware choice, Clark-Zigmond pointed out.

The platform can be used by small- and mid-sized businesses (SMBs) looking to train models without increasing costs, or larger companies can perform bursting techniques that support peak loads.

“It’s everything you need to deploy AI at scale, you’re spending less time on technical complexity,” Clark-Zigmond said.

Complementing the developer cloud platform is the Tiber AI Studio MLOps automation tool. This helps simplify elements such as cluster management, software packaging dependencies, pipelines, and monitoring. It also reminds users when models need to be retrained, Clark-Zigmond explained.

Organizations “can manage all their training and inference from a single pane of glass,” she said.

Prediction Guard: Accessing faster, more cost-effective outputs and processing needs Although there are countless possibilities for AI, there are many challenges, as well — including ‘hallucinations’ (incorrect responses) or prompt injections (when attackers use crafty inputs to get around safety mechanisms).

“All of these challenges can make it tricky for organizations to gain traction with LLMs,” said Daniel Whitenack, founder of Prediction Guard, which uses Intel Developer Cloud. “They want to avoid liabilities, and they don’t want to risk the loss of sensitive data such as personally identifiable information.”

Prediction Guard uses Intel Developer Cloud to provide customers access to multiple LLMs in a secure, private environment that monitors for harmful inputs and outputs. Organizations can also create custom deployments. The company further takes advantage of the Hugging Face Optimum Habana library to help optimize models.

Using Intel Gaudi 2, Prediction Guard has seen costs decrease and throughput increase by 2X, said Whitenack.

“This means we can help customers generate outputs faster and more cost-effectively,” he said, adding that the company now has access to the processing power it needs.

Monday.com: Taking back ownership of the model lifecycle Another Intel customer, workflow platform Monday.com, continues to see increased demand for ML tools (not unlike many other large enterprises). But previously, there was a holdup in the process: Data scientists were heavily reliant on engineers to bring models to production, explained Ohad Hegedish, a data scientist with the company.

Developers had to first set up infrastructure, which took extra time. And, even when in production, data scientists were often siloed and disconnected and didn’t have access to important MLOps capabilities such as experiment tracking and management, said Hegedish.

“We had little control on the full model lifecycle beyond the research phase,” he said.

Now with the Tiber AI Studio, Monday teams can compare different model hyperparameter configurations and training runs, train ML and deep learning models on any compute infrastructure, track model evaluation metrics, chain algorithms and write custom code in any language (among other capabilities).

Ultimately, according to Hegedish, they were able to cut down time spent on technical configurations and reduce infrastructure setup by 80%.

Intel’s MLOps capabilities have “allowed us as data scientists to take ownership of the model lifecycle end to end without a direct dependency on engineers.”

Independent attestation, transparent supply chain Organizations of all sizes worry about cybersecurity — high-profile breaches happen every day and are becoming more and more expensive.

“Threats to data and privacy are a genuine concern,” said Clark-Zigmond. “Data is one of a company’s most valuable assets.”

The new Tiber Trust and Security, she said, helps organizations work confidently and “put zero trust within reach.”

The platform facilitates confidential computing, she explained, which provides independent attestation. An admin is given cryptographic confirmation about the state of a trusted execution environment (TEE). The confirmation identifies that the TEE instantiated is genuine, conforms to security policies and is configured as expected. Attestation is governed by company policies, occurring at launch and periodically during runtime.

This capability is complemented by Intel Trust Domain Extensions (TDX), which provides isolation at the virtual machine level.

“Customers are interested in how they can stay secure at all times,” said Clark-Zigmond — including data at compute, at rest and in use.

Also, working with “transparent supply chains” helps ensure that sensitive data is verified from the moment of manufacture to its arrival, she said.

“The transparent supply chain is a set of tools that allow you to have the insight to know for sure that what you think you have, is what you actually have,” said Clark-Zigmond.