Nvidia aims for faster AI next year with new software stack

Nvidia has set out its AI 2023 aspirations with a new version of its AI software stack to go along with its Hopper GPUs.

AI Enterprise 3.0 software marks a shift in how Nvidia approaches its AI product stack, leveraging partners to explore the applications and pre-trained models it makes available to customers in vertical industries .

“Our approach has been the kind of full-stack approach where we’ve advanced the hardware, and we’ve advanced the software at the same time, and it’s the conjunction of hardware design and software together that makes really run the stack,” said Manuvir Das, vice president for enterprise IT, during a press and analyst briefing.

The combination of Nvidia’s GPUs and Enterprise AI work together to deliver significant speedup over time.

“We’re not just talking about a 10x or 100x improvement, we’re actually talking about a 1,000,000x improvement over a period of time, which is really mind-boggling,” Das said.

Das explained the million-plus improvement over time through a software and hardware co-design approach. Engineers work on optimizing the software stacks – which are released monthly – on the hardware.

“Algorithms run much faster the more we tune. It’s a whole thing. It’s not just GPU processing power. It’s network connectivity and how we send data over the network between GPUs, how we scale between GPUs, how we use memory associated with GPUs more intelligently,” Das said.

The new version of the AI ​​software includes new frameworks to support speech, visual recognition, natural language processing, cybersecurity and other applications.

Nvidia’s AI models in AI Enterprise 3.0 are largely proprietary and will support training models made available by organizations like OpenAI. OpenAI recently made available its ChatGPT conversational AI model to generate AI-powered answers to questions posed by users.

AI Enterprise includes workflows, which are software packages and use case models much closer to the problems customers are trying to solve, Das said.

An example includes creating a digital twin of a human being for security reasons. AI tools can create an employee profile based on fingerprints or other behavioral tendencies.

“If you observe behavior from that employee that deviates from the fingerprint…then you raise an alarm because there could be an intruder impersonating that employee,” Das said.

Nvidia has added many new workflows to AI Enterprise 3.0, as well as pre-trained models based on publicly available data.

The company provides unencrypted pre-trained models for vision, AI and object recognition applications. Das said this feature is particularly important for “explainable AI,” which DARPA has described as a type of AI with a high level of prediction accuracy, and which enables human users to understand and trust source results.

“The reason we provide these unnumbered models is that a developer can then look at what are called weights and biases inside the model so you can get a better idea of ​​how the model works and why it makes the judgments it makes,” Das said.

The software also adds customer service workflows that include virtual assistants and audio transcription.

Nvidia is also tapping Deutsche Bank to develop workflows for the financial services industry. These workflows are for fraud detection, voice, and other applications used in the highly regulated industry.

“We work closely with Deutsche Bank and we continue to innovate and add new frameworks. More SDKs, more pre-trained models, more workflows to Nvidia AI Enterprise as well as more places it can be used and deployed,” Das said.

Related