According to China’s Taiwan Business Times, Intel’s next-generation Battlemage and Celestial GPU product lines will be OEM’d by TSMC, based on 4nm and 3nm processes respectively.
The source said that although the AXG department has been reorganized and is now divided into two departments, CCG and DCAI, Intel remains optimistic about the performance of Arc graphics cards in the gaming field and is optimistic that gaming and AI-related applications will drive strong demand for GPUs, and is currently working with TSMC to plan two GPUs.
According to the report, the first GPU will be the second generation Battlemage GPU based on Xe2 architecture, which is expected to arrive in the second half of 2024 and will use TSMC’s 4nm process node. Then in the second half of 2026, the third generation of Celestial GPUs based on the Xe3 architecture will use TSMC’s N3X node.
In addition, Intel’s Ponte Vecchio for AI and HPC will also be produced using TSMC’s 7nm (I/O connectivity chip) and 5nm (compute chip), and the second generation Rialto Bridge has been canceled and resources will be focused on the new Falcon Shores, which is expected to enter production in 2024 and hopefully ship to customers in 2025. It is hoped that 2025 can be shipped to customers.
A recent product plan leaked by Intel engineers shows that the Battlemage Xe2 and Celestial Xe3 GPUs will be available in both HPG and LPG versions.
The Battlemage LPG will be used with Lunar Lake CPUs, while the Celestial LPG will be used with Panther Lake CPUs; and the HPG Arc graphics card will be launched earlier than the LPG products, focusing on the 75-250W segment of the market.