Nvidia’s LLM is trained on chip design

We’re seeing the term LLM (large language model) being tossed around a lot nowadays, and it’s because of the AI explosion. Several companies like Google, Meta, OpenAI, Anthropic, etc. have one or multiple LLMs. Well, GPU maker Nvidia is one of them, and it trains its LLM on chip design.

Nvidia’s LLM isn’t particularly new, and the company has been hard at work training it on trillions of words. It has to learn just like all other models, but this one sits apart from other models on the market. Several other companies train their models to eventually be used by the general public to learn general information. This is what we’re seeing with chatbots like ChatGPT and Bard.

But, Nvidia is training its LLM on chip design

Nvidia, on the other hand, has a more specialized vision for its LLM. Its primary foundation model is named NeMo, and the company has been training it on internal information for a while.

The company’s Bill Dally introduced the latest evolution of NeMo called ChipNeMo, and it’s a version of Nvidia’s LLM trained specifically on chip design. So, while other models are designed to tell you more about the world, ChipNeMo is made to tell you more about chip design.

If you’re waiting for Nvidia to release this as a public tool for you to learn about chip design, you’re going to be disappointed. Nvidia built this LLM for internal use. One way that AI can positively impact business is that it can automate and expedite tedious or unnecessary tasks.

As per the report, Dally stated that this could help junior designers get assistance. Usually, senior designers need to help the junior designers with their questions. That costs the senior designers time that they could use for work. Well, ChipNeMo could provide information to the junior designers.

Another area where ChipNeMo has the potential to help is in the bug department. Documentation explaining bugs tends to be extremely long. People have to read these documents just to know the inner workings of the bug, which takes time. Well, ChipNeMo could summarize the bug in just a few paragraphs and tell the designers how to fix the issue.

This is a business tool at the end of the day, and it could Nvidis speed up productivity. As Dally puts it, “If we even got a couple percent improvement in productivity, this would be worth it. And our goals are actually to do quite a bit better than that.”