“Intel’s Falcon Shores GPUs Set to Revolutionize AI Training for Next-Generation LLMs”

Intel says its Falcon Shores GPUs will be ready to support the next generation of LLMs despite a previously announced delay.

Intel wants to play a bigger role in training AI models. The Intel Xeon Scalable CPUs are an important part of the servers on which Large Language Models (LLMs) are built, but in fact they mainly fulfill a supporting role. The heavy lifting is for the GPU accelerators and Nvidia is lord and master there.

Intel therefore wants to throw a new weapon into the fight against the next generation of LLMs: Falcon Shores. The Falcon Shores GPUs must leave for customers in 2025, the manufacturer confirms. That timing is not that spectacular, since Intel was previously forced to postpone Falcon Shores.

Worth the wait?

What exactly we are waiting for is also not so clear. Intel does share some details at the ISC conference in Hamburg. For example, the manufacturer plans to equip the chips with a solid 288 GB of HBM3 memory. For comparison: the most powerful accelerator at the moment is the Nvidia H100 Hopper with ‘only’ 80 GB RAM. Furthermore, Intel says that the chip will support 8-bit floating point workloads, which is relevant for efficient training of new models.

However, when Falcon Shores launches, the landscape of HPC accelerators will look very different. Nvidia is not sitting still and wants to further convince the market with the Grace Hopper super chip, which combines ARM CPUs with GPU power. AMD has a similar concept ready with the Instinct MI300. A Falcon Shores chip with a CPU component was planned, but has since been dropped.