Glossary
Technology evolves at a rapid-fire pace. That’s why we’ve built an easy-to-use glossary to help you better understand the terms, technologies and trends that impact your business.Scale-up
What is scale-up?
In AI infrastructure, scale-up means expanding compute performance within a tightly coupled system by connecting more GPUs or other accelerators in the same server, rack or scale-up domain.
Scale-up architectures are designed to support very high-bandwidth, low-latency communication between processing elements that exchange data frequently.
Why is scale-up important?
AI training workloads require intensive communication and synchronization between processors. Scale-up connectivity helps these processors operate together efficiently, maximizing accelerator utilization and reducing training times.
However, scale-up capacity is ultimately constrained by the physical limits of the system and its interconnect technology.
Key considerations
- Ultra-low-latency connectivity
- High-bandwidth interconnects
- Physical-link integrity
- Cluster synchronization
- Consistent performance across accelerator links
Learn more
Explore testing solutions for high-speed, high-density networks inside the data center.