Intel has revealed its latest computer chip, named the Gaudi3 AI Accelerator, an artificial intelligence (AI) chip tailored specifically for generative AI software.
The new chip, set to launch next year, is poised to compete directly with counterparts from Nvidia and AMD, which currently dominate the landscape when it comes to powering large and power-intensive AI models.
OpenAI’s ChatGPT predominantly uses Nvidia GPUs in the cloud, contributing to Nvidia’s stock surge of nearly 230% year to date, in contrast to Intel’s 68% increase.
In response, AMD and now Intel have introduced chips designed to challenge Nvidia’s market lead.
Last month, NVIDIA warned investors of a forthcoming decline in sales following US export controls towards China on high-end AI chips.
The news comes as the Biden administration calls for restrictions on overseas use of the most advanced US-made AI chips.
While specific details about Gaudi3 remain limited, it is positioned to rival Nvidia’s H100, a popular choice for companies building extensive chip farms to drive AI applications, as well as AMD’s upcoming MI300X, which is expected to ship to customers in 2024.
Intel initiated the development of Gaudi chips in 2019 following the acquisition of chip developer Habana Labs.
The Core Ultra chips, tailored for Windows laptops and PCs, incorporate a specialised AI component known as an NPU, enhancing the efficiency of AI programme execution.
This move aligns with the broader industry trend, whereby traditional processor manufacturers like Intel, AMD and Qualcomm are realigning their product lines to cater to rising demand for chips driven by AI models.
While the Core Ultra may lack the sheer power needed to run a chatbot like ChatGPT without an internet connection, it excels in handling smaller tasks.
The Core Ultra line-up is also tailored to gaming, with added graphics capabilities that accelerate programs like Adobe Premier by more than 40%.
In addition to Core Ultra, Intel announced new fifth-generation Xeon server chips tailored for large organisations including cloud companies.
Though pricing details were not disclosed, previous Xeon processors were high-cost, and often paired with Nvidia GPUs in systems for training and deploying generative AI.
