Put simply: Nvidia's GPUs are the linchpin of its burgeoning AI business, but the company is struggling to keep up with industry demand. ASIC chips, which offer customized silicon designs that hardware-accelerate specific computing workloads, could soon become the next big thing in the field.
According to a job listing recently posted online, Meta is looking to hire highly skilled hardware engineers to develop a new generation of application-specific integrated circuit (ASIC) designs. ASIC chips are customized at the hardware level for specific computing applications and aren't suited for general-purpose use like traditional CPUs. But ASICs are highly efficient at what they're designed for, such as processing machine learning algorithms for AI workloads and chatbot services.
Meta's latest job openings spotted by The Register are in Bangalore, India, and Sunnyvale, Calif. In India, Facebook's parent company is hiring an ASIC engineer to build hardware accelerators for data centers. According to the job listing, the new hire will be a key member of the ASIC team working on cutting-edge chip architectures that accelerate AI/machine learning algorithms.
Meta wants to build “green” data center accelerators, despite the notorious problem that AI applications consume large amounts of power. The new ASIC engineer will need to work on both new advanced hardware architectures and the algorithms that will drive and test those new chips. The role's responsibilities include developing performance and functional models to validate new architectures, creating machine learning kernels to analyze ASIC chips, and optimizing architectures for ML workloads.
Facebook is looking for candidates with at least a bachelor's degree in computer science or engineering, and at least 10 years of experience and knowledge of computer architecture concepts such as processors, memory systems, on-chip interconnect networks, etc. Candidates should also have extensive experience with low-level, object-oriented programming languages such as C++ (and C).
Some of Meta's ASIC architecture jobs were first posted in December 2023 but were relisted two weeks ago, The Register noted. The company has previously said it wants to develop its own “inference accelerator,” which is expected to come online in 2024.
According to Nvidia CEO Jensen Huang, GPUs have played a key role in driving Meta's “recommendation engine” and the company's overall revenue. But he also said that Nvidia likely won't be able to provide enough GPU accelerators for people in the AI business in the near future. Meta also appears to be working on the elusive concept of artificial general intelligence (AGI), a task that is prohibitively difficult for modern GPU technology, but which could benefit greatly from a dedicated ASIC silicon design.
