Recent advances in technology have placed the power of artificial intelligence (AI) and machine learning (ML) in the hands of everyday users. AI enables innovation in almost all industry sectors, where its ability to adapt and respond to evolving demands creates exciting new opportunities. From the cars we drive to the smart farms that form the future of the agricultural industry, it is hard to find an industry that will not be transformed by AI.
The rise of AI applications drives the demand for advanced infrastructure that will deliver increased data speeds, greater bandwidth, and higher frequencies. To respond to the increasing demand that the explosion in AI will create, these architectures will require scale and the ability to reconfigure quickly.
The Rise of Hyperscale
Data centers form a crucial component of our modern network infrastructure. The popular image of a data center is a large warehouse building populated with rack after rack of computing equipment. Many are enterprise data centers owned by individual organizations, whether commercial or governmental. These dedicated servers process information locally, taking control of their own data needs. While these hubs of communication form part of the data processing solution, they are just a small element of the overall strategy.
Hyperscale data centers will provide the computing power to drive the AI revolution. As the name suggests, hyperscale installations are far larger than traditional data centers and will cater to many users. Alongside their growth in size, the number of hyperscale data centers is expanding considerably with the increase in data-intensive applications like AI.
Scalability and Flexibility
The complexity and variety of AI applications means that every hyperscale installation will be different, each located almost anywhere in the world. This creates the need for highly flexible solutions. Customers want to benefit from the improved performance of the latest data networks and must be confident that their equipment can keep pace with their demands. Therefore, the flexibility to scale in line with data needs is vital to meet customer expectations.
Electrical connectors are key to the ability to quickly reconfigure. Connectors deliver power and signals while ensuring the quality of the data passing through them, which is where signal integrity (SI) comes into play. SI describes the quality of the electrical signals that are transmitted through connectors and over cables.
The latest connectors feature a low profile, small size, and reduced pin spacings to make them suitable for the latest compact server applications. The high density of signals passing through confined spaces has the potential to cause interference from crosstalk. Therefore, the cables, connectors, and printed circuit board (PCB) traces through which the signal travels are designed specifically to preserve SI.
With so much riding on signal integrity in the emergence of hyperscale data centers, Molex has developed a range of AI and ML solutions that help ensure data processing reliability and performance. Minimizing signal losses, particularly in data-hungry AI applications, is challenging. This is especially true when using conventional PCB-mounted connectors. Molex’s high-speed Direct Attach Cables (DACs) present an advanced solution to this problem. DACs offer transmit data rates as high as 100Gbps and are available in off-the-shelf and custom lengths.
For rack-based server applications, Molex has created board-to-board 224Gbps-PAM4 connector solutions, such as the Mirror Mezz. These connectors offer data speeds up to 56Gbps per differential pair in a stackable mating, low-profile design that is ideal for the high-density servers employed in hyperscale data centers.
Power for Hyperscale Data Centers
Power plays an enormous role in the future of our data networks. While exact predictions vary depending on the source, it has been estimated that as much as 4 percent of all electricity generated globally will be used in data centers by the end of this decade.[1] While SI is vital to the quality of communications, power distribution will be just as important in ensuring the smooth running of hyperscale data centers.
Efficient power usage is essential to minimize operational costs and maximize performance. Power distribution and modular infrastructure are also critical in allowing for expansion. A modular power supply enables data centers to add capacity as needed without disrupting existing operations.
Hardware acceleration can play a significant role in the efficient computing at the heart of data centers. Hardware acceleration allows specific computing tasks to be transferred to specialized hardware components within the system, enabling greater efficiency than is possible in software running on a general-purpose CPU alone.
Conclusion
AI systems are changing how we handle data. To reach their full potential, AI systems will depend on hyperscale data centers to deliver the flexibility and scalability needed to respond quickly to changing demands.
Connectivity is critical to the performance and modularity of the latest AI architecture. Molex is solving the complex challenge of delivering flexibility and ease of use with the latest standards needed in SI and high-speed data. Molex solutions balance the demands of smaller size and higher density to deliver the performance required for the hyperscale data centers of the future.
Author: David Pike is well known across the interconnect industry for his passion and general geekiness. His online name is Connector Geek.
