Intel Corp. AI chip development unit Habana Labs is laying off dozens of employees at its development center in Caesarea. As far as is known, the decision was not taken by Intel itself but local management, which has remained relatively independent of the chipmaker's headquarters in Santa Clara, California.
Habana Labs, which manufactures AI processors Gaudi and Greco, was acquired by Intel in 2019 for $2 billion. The company's founders included serial entrepreneur and chip developer Avigdor Willenz, who is not active in the company. Habana Labs is currently led by CEO David Dahan, the former COO of DSP Group and prior to that PrimeSense, which was acquired by Apple. Habana Labs VP RnD Ran Halutz was formerly HW Development at Apple and prior to that Group Manager at PrimeSense.
Intel acquired Habana Labs in order to compete with graphics processing giant Nvidia, which has leveraged the great processing power of these chips for AI applications and smart operation of data centers. Intel is in fierce competition with Nvidia's hardware performance, and its Cuda AI software environment, and even fiercer competition with the chips produced by Amazon - which itself is also an Intel customer - and this is reducing its room to maneuver.
Intel said, "Habana Labs periodically reviews and revises its technical and business focuses, in order to adapt to the current business reality and continue to improve competitiveness. As part of these processes, adjustments are made from time to time in the situation of employees and the balance between various disciplines. This is a normal process that takes place all the time, and which allows Habana to continue developing attractive and competitive products and solutions."
In a performance test conducted in July between Intel's Gaudi2 processor and its rival Nvidia’s flagship A100 GPU graphic processor, the Israeli chip was found to be relatively cheap in performance by MLPerf in areas such as identifying images and natural language processing. According to the performance trial, Gaudi2 succeeded in training an artificial neural network in 36% less time than Nvidia's chip and this translated into a 56% improvement in performance. In the field of natural language processing, the Gaudi2 processor took 7% less time than was required by the Nvidia processor.
Published by Globes, Israel business news - en.globes.co.il - on October 11, 2022.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2022.