Tech Giants Unite OCI MSA Group Formed to Standardize Optical Interconnects for the AI Era

Beyond Copper: How the OCI MSA Alliance is Redefining AI Data Center Efficiency.

A powerhouse coalition including AMD, Broadcom, Meta, Microsoft, NVIDIA, and OpenAI has announced the formation of the Optical Compute Interconnect (OCI) Multi-Source Agreement (MSA) group. This strategic alliance aims to establish a universal standard for high-performance data connectivity, enabling interoperability across different manufacturers while significantly reducing the industry’s reliance on traditional copper cabling.

Paving the Way for Terabit Speeds

The OCI MSA will debut with a standard featuring 4 channels, providing a bandwidth of 50Gbps per channel per direction. The roadmap is aggressive, with plans to scale total bandwidth to 800Gbps per fiber, and a long-term goal of reaching a staggering 3.2Tbps.

The standard defines a shared Physical Layer (PHY) to ensure hardware compatibility regardless of the vendor. These connections can be implemented directly on circuit boards or integrated into chips via Co-Packaged Optics (CPO) technology.

The Shift from Copper to Light

High-bandwidth connectivity is the backbone of AI data centers, where massive datasets must be moved between machines instantaneously. By shifting from electrical signals over copper to optical signals, energy efficiency can be improved by as much as 3.5 times. The inclusion of NVIDIA in this group is particularly notable, signaling a future where diverse AI hardware can communicate seamlessly through a unified optical fabric.

In the past, copper cables were the primary standard, but as we entered the AI ​​era requiring terabit-level speeds, copper cables began to experience problems with "heat" and "distance" (signal decay). Switching to photonics not only increased speed but also drastically reduced heat buildup in racks, a key factor in lowering data center electricity costs.

CPO technology involves placing the photonic module directly on the same package as the GPU/CPU. This significantly reduces the distance electrical signals must travel on the board, minimizing power loss and lowering latency to near-internal chip connectivity.

The collaboration between giants like Microsoft and Meta (major buyers) and NVIDIA and AMD (major vendors) to establish a common standard means that in the future, organizations can choose GPUs from one vendor and switches from another without vendor lock-in, fostering more balanced competition in the market.

The 3.5x energy savings aren't just about cost; they're about sustainability. Since AI consumes enormous amounts of power, having an OCI standard will help ensure that the expansion of AI doesn't excessively harm the environment. 

 

 

[Rumor] Meta Secret Model Avocaco Slips to May Following Underwhelming Benchmarks. 

 

Source: BusinessWire 

💬 AI Content Assistant

Ask me anything about this article. No data is stored for your question.

Beyond Copper: How the OCI MSA Alliance is Redefining AI Data Center Efficiency.

A powerhouse coalition including AMD, Broadcom, Meta, Microsoft, NVIDIA, and OpenAI has announced the formation of the Optical Compute Interconnect (OCI) Multi-Source Agreement (MSA) group. This strategic alliance aims to establish a universal standard for high-performance data connectivity, enabling interoperability across different manufacturers while significantly reducing the industry’s reliance on traditional copper cabling.

Paving the Way for Terabit Speeds

The OCI MSA will debut with a standard featuring 4 channels, providing a bandwidth of 50Gbps per channel per direction. The roadmap is aggressive, with plans to scale total bandwidth to 800Gbps per fiber, and a long-term goal of reaching a staggering 3.2Tbps.

The standard defines a shared Physical Layer (PHY) to ensure hardware compatibility regardless of the vendor. These connections can be implemented directly on circuit boards or integrated into chips via Co-Packaged Optics (CPO) technology.

The Shift from Copper to Light

High-bandwidth connectivity is the backbone of AI data centers, where massive datasets must be moved between machines instantaneously. By shifting from electrical signals over copper to optical signals, energy efficiency can be improved by as much as 3.5 times. The inclusion of NVIDIA in this group is particularly notable, signaling a future where diverse AI hardware can communicate seamlessly through a unified optical fabric.

In the past, copper cables were the primary standard, but as we entered the AI ​​era requiring terabit-level speeds, copper cables began to experience problems with "heat" and "distance" (signal decay). Switching to photonics not only increased speed but also drastically reduced heat buildup in racks, a key factor in lowering data center electricity costs.

CPO technology involves placing the photonic module directly on the same package as the GPU/CPU. This significantly reduces the distance electrical signals must travel on the board, minimizing power loss and lowering latency to near-internal chip connectivity.

The collaboration between giants like Microsoft and Meta (major buyers) and NVIDIA and AMD (major vendors) to establish a common standard means that in the future, organizations can choose GPUs from one vendor and switches from another without vendor lock-in, fostering more balanced competition in the market.

The 3.5x energy savings aren't just about cost; they're about sustainability. Since AI consumes enormous amounts of power, having an OCI standard will help ensure that the expansion of AI doesn't excessively harm the environment. 

 

 

[Rumor] Meta Secret Model Avocaco Slips to May Following Underwhelming Benchmarks. 

 

Source: BusinessWire