The world-famous chip maker, Nvidia, is using its powerful graphics cards to help its engineers design the next generation of GPUs. Nvidia is currently at the forefront of GPU manufacturing and artificial intelligence. Nvidia DLSS and OptiX are two of the popular AI technologies the company has developed recently. Additionally, the company is always in the process of developing applications that utilize its GPUs in the best possible manner.
Nvidia currently dominates the competition in terms of discreet GPU market share and gaming. According to Jon Peddie Research, Nvidia has 81 percent of the dGPU market share. Additionally, as per the Steam gaming hardware survey, 77.13 percent of PC gamers use Nvidia GPUs on the platform. This high demand for Nvidia GPUs rose through the roof during the pandemic due to a mixture of the supply chain, scalping and crypto mining issues. As a result, in 2021, Nvidia cards saw a 300 percent price hike in their street price and yet remained out of stock most of the time. Thankfully, Nvidia GPU prices have started to come down in recent months.
Related: Nvidia’s RTX 3090 Ti Costs $1999 And Is Available Now, If You Can Find One
Nvidia makes some of the best GPUs on the market. And they’re now taking their chip design capabilities to another level by incorporating machine learning features by using its own GPUs. During an Nvidia GTC conference, the company’s chief scientist Bill Dally said, “It’s natural as an expert in AI that we would want to take that AI and use it to design better chips.” So to design better GPUs, the development team at Nvidia is enhancing current-gen computer-enabled design tools with AI capabilities. It is primarily doing so in four major areas.
Nvidia GPUs Work More Efficiently Than Humans
The first area involves mapping voltage drops to determine where the power is used in an Nvidia GPU. Dally explained during the presentation that running it manually on CAD would take three hours, but with the help of an AI-powered GPU setup, the same can be accomplished in around 18 minutes. The second area involves testing parasitics to check how a circuit design would perform, which is a frequentative process that AI handles. In the third area, an AI-powered GPU tests different layouts of the chips to determine the least congested design format. And lastly, GPUs are used to create new designs as well. Nvidia’s NVCell technology uses Reinforcement Learning to work as an automatic standard cell layout generator. Dally explained that whenever technology evolves, like the transition of the fabrication process from 7nm to 5nm nodes, thousands of these cells have to be redesigned using “a very complex set of design rules.” Nvidia’s NVCell can recreate 92 percent of the cell library with seemingly no error.
For reference, it would require a group of 10 people to work for over a year to port a new technology cell library. The same can be accomplished with the help of a few powerful Nvidia GPUs in a matter of days, explained Dally. Of course, human intervention is still required in all these areas, as futuristic as it sounds. However, the AI-enabled Nvidia GPUs help the company massively save time and make a better-designed chip. In addition to GPUs, Nvidia might soon dip its toes in the CPU manufacturing business as well, but that’s a story for another time.