Intel says upcoming AI chip beats Nvidia, AMD on price and heat
▼ Summary
– Intel plans to ship its “Crescent Island” AI chip by end of this year, using cheaper air-cooling and LPDDR5 memory instead of expensive high-bandwidth memory and liquid cooling used by rivals Nvidia and AMD.
– The chip is designed for “inference” tasks, where a user makes a request, rather than the training of AI models, where Nvidia’s processors dominate.
– Intel’s earlier AI training GPU called “Gaudi” had poor sales, and its planned successor was cancelled last year.
– Kevork Kechichian, Intel’s data center chief, said the company is “starting with the basics” and is not aiming for the training market based on past experience.
– The effort is Intel’s first major push into AI infrastructure under new CEO Lip-Bu Tan, who replaced Pat Gelsinger amid concerns over the turnaround strategy.
Intel plans to deliver a new AI chip by late this year, one that relies on more affordable memory and simpler cooling than competing products from Nvidia and AMD. The move signals the company’s renewed push to carve out a foothold in the fast-growing market for semiconductors powering artificial intelligence.
Kevork Kechichian, who leads Intel’s data center group, told the Financial Times that the company is “starting with the basics” as it works to challenge its larger rivals. The new processor, code-named “Crescent Island,” is a graphics processing unit designed specifically to accelerate inference tasks , the stage when a user submits a request to an AI model , rather than the training phase, where Nvidia’s chips currently dominate.
Intel’s earlier attempt to build a GPU for AI training, called “Gaudi,” struggled with poor sales, and its planned successor was canceled last year. “We decided to start rebuilding our muscles in AI,” Kechichian said. “But we are not particularly aiming for the training market based on past experience.” He joined Intel last year from chip designer Arm.
The new chip will begin shipping in limited quantities to customers by the end of this year, following an 18-month development process. Intel is also betting on two key pain points for Nvidia and AMD: the high cost of high-bandwidth memory (HBM) and the need for liquid cooling infrastructure.
Crescent Island is an air-cooled chip that uses LPDDR5 memory, a significantly cheaper alternative to the HBM found in Nvidia’s Blackwell processors. This approach could give Intel a pricing and thermal advantage in data centers where cooling costs are a major concern.
This launch marks Intel’s first major push into the lucrative AI infrastructure market under CEO Lip-Bu Tan, who took over last year after Pat Gelsinger was ousted amid concerns that his turnaround strategy was failing. With Crescent Island, Intel is aiming to prove it can compete where it matters most: cost and practicality.
(Source: Ars Technica)
