Nvidia, Ayar Labs team on optical infrastructure for AI, HPC
Nvidia Corp. (Santa Clara, Calif.) and startup Ayar Labs Inc. (Santa Clara, Calif.), are workig together to develop artificial intelligence (AI) and high-performance computing (HPC) infrastructure.
Nvidia recently invested in Ayar Labs’ Series C round of financing (seeHP, Nvidia back Ayar Labs’ $130 Million Series C).
AI model sizes have been growing exponentially. Nvidia said that it believes models will have 100 trillion or more connections in 2023, a 600-fold increase over just two years. This will exceed the capability of copper interconnect to service and require optical interconnect.
The two companies plan to accelerate the development and adoption of optical I/O technology to support AI and machine learning (ML) applications and data volumes.
Optical benefits worth it
Optical I/O increases the bandwidth of components at lower latency and at a fraction of the power of electrical I/O. It comes at the expense of complexity and either the need to use compound semiconductors or specialized silicon photonics.
However, with the move to chiplet multi-die packaging this trade-off is starting to favor the use of optical signalling. The technology supports emerging heterogeneous compute systems, disaggregated/pooled designs, and unified memory architectures, Ayar Labs claims.
“今天最先进的人工智能/毫升training architectures are limited by current copper-based compute-to-compute interconnects to build scale-out systems for tomorrow’s requirements,” said Charles Wuischpard, CEO of Ayar Labs, in a statement. “Our work with NVIDIA to develop next-generation solutions based on optical I/O provides the foundation for the next leap in AI capabilities to address the world’s most sophisticated problems.”
Related links and articles:
News articles:
HP, Nvidia back Ayar Labs’ $130 Million Series C
First terabit optical link for chip-to-chip connections
