Meta Seeking ASIC Designers for ML Accelerators and Data Center SoCs • The Register

AI and ML Jobs


Meta wants to build accelerators and SoCs that run in data centers for uses like machine learning, but it seems to be having trouble finding the talent to design them.

Registry We found numerous job ads in Bangalore, India and Sunnyvale, California for ASIC engineers with expertise in architecture, design and testing. Some of the jobs listed Meta as the employer, while others listed Facebook as the job.

The architecture job details an opportunity to “help design cutting-edge machine learning accelerators.”

The design job seeks candidates with “deep experience across one or more key disciplines required to successfully build world-class, complex SoCs and IP for data center applications,” and will be doing so in Facebook's infrastructure organization in Bangalore, India.

The design validation job requires “experience in design validation of data center applications including video, AI/ML, and network design.”

Interestingly, some of the jobs were first posted on LinkedIn in late December 2023, but were reposted two weeks ago. Some of the Sunnyvale jobs have only received a handful of applicants, despite the salary approaching $200,000.

While it hasn't said what Meta wants to build, it's not hard to guess what the social network is up to: The company has previously spoken about its homegrown “Meta Training Inference Accelerator” and its intention to bring it online in 2024.

With AI exploding and Nvidia admitting it will struggle to meet demand for its products, it's no surprise that Meta has decided to double down on its efforts to avoid having to bid on its products in a very tight market. Indeed, Nvidia CEO Jensen Huang claimed on Wednesday that the company's hardware has helped improve Meta's recommendation engine, which has boosted its bottom line. So Meta clearly needs the power of ML, and may have decided to flex its cash flow to generate it, rather than relying on struggling suppliers.

Meta is also developing artificial general intelligence, which may require specialized silicon.

Whatever Meta wants to build, the Indian government would be happy to see it advertise in Bangalore, where it sees chip design talent as an attempt to become a bigger player in the global semiconductor industry. New Delhi would be thrilled if it had an Indian-designed ML accelerator.

We've asked Meta what they plan to build, but we're not going to wait around hoping for a meaningful answer.®

P.S.: Rumors circulated this week that Microsoft is developing server network cards to improve performance for machine learning workloads and reduce reliance on Nvidia and the networking technology the GPU giant absorbed with its acquisition of Mellanox. This is a plausible story.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *