During its Vision event on Tuesday, April 9, Intel (INTC.O) unveiled a new version of its artificial intelligence chip that aims to challenge Nvidia’s (NVDA.O) market supremacy in semiconductors that drive AI.
Tech firms are searching for a different supplier of the rare chips required for artificial intelligence. According to Intel, the Gaudi 3 chip could train certain large language models 50% faster than Nvidia’s previous generation H100 CPU. For certain of the models Intel examined, it can also compute generative AI responses—a process known as inference—faster than the H100 CPUs.
“Our customers, first and foremost, are asking for choice in the industry,” stated Jeni Barovian, vice president of strategy and product management at Intel. They are coming to us and they are expecting that Intel, as a computing leader, will follow the wave of (generative AI) and deliver solutions that meet their needs. And they are looking for an open approach.”
Both Advanced Micro Devices (AMD.O) and Intel have had difficulty creating a package of chips and software that is appealing enough to allow developers to create AI applications that could displace Nvidia. In 2023, Nvidia dominated about 83% of the market for data center chips, with Google’s bespoke tensor processing units (TPUs), which it does not sell directly, holding the lion’s share of the remaining 17%.
To create the chips, Intel employed Taiwan Semiconductor Manufacturing Co.’s (2330.TW) new tab 5nm technique. The Gaudi 3 is more than twice as fast as its predecessor and consists of two main processing chips fused together. The chip can produce a massive amount of computing power when connected in series with thousands of additional chips.
WHAT COMES NEXT?
In the second quarter of this year, server manufactures including Supermicro (SMCI.O), opens new tab, and Hewlett Packard Enterprise will get access to the Gaudi 3 processor.
