Recently, WCCFtech.com's Internet resource has received some internal documents that confirm that Intel is preparing for the launch of the next-generation artificial intelligence training accelerator with monstrous capabilities at the end of this year. The named Platform of Habana Gaudi 2, aims to new NVIDIA products for data processing centers (released later this week) for deep learning and aims to provide a more profitable (from the point of view of the total cost of ownership) alternatives. Keep in mind that the documents were dated earlier this year, and the launch window could be shifted during this time, so we assume to this with the fraction of skepticism.

This chip is equipped with 6 memory stacks of type HBM, the capacity of which can be 48 or 96 GB, depending on whether HBM2 (E) or HBM3 is used. The processor uses the OCP-OAM form factor (Open Compute Accelerator mezzonine), which is the same type as its HL205 design of 2019.

It is unclear what technical process is used in the HL2080 chip, but assuming that the company has been acquired by Intel, it is possible that it can be an Intel's own factory. Last year, the CEO of Intel Pat Gelsinger stressed that the company has a very aggressive path to the artificial intelligence of Habana:

The HL2080 information leak appeared just a few days before presentation of the NVIDIA Hopper architecture on GTC 2022. NVIDIA architecture was often used to compare the performance of acceleration AI. The release date of the HL2000 is currently unknown.

#artificial intelligence