On Thursday, Intel unveiled a new line of computer chips, among them the Gaudi3, an AI chip designed for generative AI software. When Gaudi3 launches the following year, it will take on rival Nvidia and AMD chips that power large and resource-demanding AI models.
The most well-known AI models, such as ChatGPT from OpenAI, operate on cloud-based Nvidia GPUs. For this reason, Nvidia’s stock has increased by almost 230% year to date, compared to a 68% increase in Intel shares. And for that reason, businesses like Nvidia’s dominant position in the market, in hopes of drawing in AI companies, have announced chips, as has AMD and now Intel.
As of Thursday, Intel’s stock was up 1%.
Although little information was provided by the company, Gaudi3 will take on Nvidia’s H100, which is the primary option for businesses that construct massive chip farms to power AI applications, as well as AMD’s upcoming MI300X, which will begin shipping to customers in 2024.
Following its acquisition of Habana Labs, a chip developer, in 2019, Intel began producing Gaudi chips.
At a New York launch event where he unveiled Gaudi3 and other chips aimed at AI applications, Intel CEO Pat Gelsinger remarked, “We’ve been seeing the excitement with generative AI, the star of the show for 2023.”
“We believe that in the coming year, the AI PC will be the main attraction,” Gelsinger continued. And that’s where the new Core Ultra processors from Intel, which were also revealed on Thursday, will be useful.
New Xeon chips and Intel Core Ultra
Along with new fifth-generation Xeon server chips, Intel also unveiled Core Ultra chips intended for Windows laptops and PCs. To run AI programs more quickly, both come with an NPU, a specialized AI component.
It is the most recent indication that conventional processor manufacturers, such as AMD and Qualcomm, Intel’s competitors, are realigning their product lineups and warning investors that artificial intelligence (AI) models may result in an increase in demand for their chips.
While it can manage smaller tasks, the Core Ultra won’t be able to run a chatbot like ChatGPT without an internet connection. For instance, Zoom uses Intel chips to operate its background-blurring feature. The company’s 7-nanometer process, which uses less power than previous chips, is used in their construction.
Crucially, however, the 7-nanometer chips demonstrate that Gelsinger’s plan to overtake Taiwan Semiconductor Manufacturing Co. in terms of chip manufacturing capabilities by 2026 has not stalled.
More powerful gaming features are also included in Core Ultra chips, and the increased graphics power can make applications like Adobe Premier run more than 40% faster. The collection debuted with laptops that went on sale on Thursday.
Lastly, servers used by major corporations such as cloud providers are powered by Intel’s fifth-generation Xeon processors. Pricing was not disclosed by Intel, but the previous Xeon was quite expensive. Xeon processors from Intel and GPUs from Nvidia are frequently used in systems for generative AI training and deployment. Eight GPUs and one or two Xeon CPUs are paired in certain systems.
According to Intel, inferencing—the process of deploying an AI model—will benefit greatly from the newest Xeon processor since it requires less power than training.