According to HPC Wire, the gathered throngs at the SC2024 conference heard how this new high-performance chip, aimed at faster AI inference, will combine Gaudi ASICs and Xe-HPC GPUs using Intel's in-house 18A node.

Despite having a terrible couple of years, Chipzilla insists it is committed to advancing its client and server platforms. The company has been developing its next-gen Battlemage GPUs, which were recently leaked, but these are not the only GPUs on the horizon.

According to HPC Wire, Intel plans to ship Gaudi 3 in volume next year, but the latest addition to its lineup, Jaguar Shores, won't be limited to the Gaudi platform. The new chip is positioned as a successor to Falcon Shores and is expected to be released in 2025. Like Falcon Shores, Jaguar Shores will likely be a dedicated GPU rather than a processor for the Intel Gaudi Accelerator.

Falcon Shores was initially designed as a hybrid GPU-CPU solution but has since transitioned to a discrete GPU for AI inference. Jaguar Shores is expected to follow suit, possibly featuring a GPU-ASIC design for enhanced AI inferencing. Leveraging technologies such as RibbonFET and backside power delivery, Intel plans to use the next-gen 18A manufacturing node for both Falcon Shores and Jaguar Shores.

An Intel spokesperson said: "Our AI investments will complement and leverage our x86 franchise – with a focus on enterprise, cost-efficient inferencing. Our roadmap for Falcon Shores remains."

While Intel outsources some chips from TSMC, like Lunar and Arrow Lake, it aims to gain an advantage with the 18A node, which is also planned for use in the Clearwater Forest next year. By 2027, Intel expects to use its in-house 14A process node.

Jaguar Shores will compete with industry giants NVIDIA and AMD, which dominate the AI training and inference markets with their consistent roadmaps through 2027. Intel will primarily focus on the inference market with its upcoming chips.

The Intel spokesperson said: "We continuously evaluate our roadmap to ensure it aligns with the evolving needs of our customers. While we don’t have any new updates to share, we are committed to providing superior enterprise AI solutions across our CPU and accelerator/GPU portfolio."

Last year, the company cancelled the Rialto Bridge GPU and shifted Falcon Shores from a hybrid to a discrete GPU. Furthermore, it also cancelled the Ponte Vecchio GPU following the online launch of the supercomputer Aurora.