At the ongoing ISC High Performance 2025 conference in Hamburg, dedicated to high-performance computing, GigaIO unveiled its next-generation AI Fabric technology on June 9, 2025. This significant announcement, covered by specialized publications like HPCWire, introduces an innovative approach to designing large-scale AI cluster architectures. GigaIO's AI Fabric technology is an ultra-fast interconnect based on the PCIe standard, which enables the creation of a "composable disaggregated infrastructure." Simply put, it allows expensive AI accelerators, such as GPUs, to be physically "detached" from specific servers and pooled into a single shared resource pool. This enables any server in the rack to dynamically connect to the required number of GPUs as if they were locally installed. This solves one of the main problems in modern data centers – the inefficient use and "stranding" of expensive hardware when accelerators are tied to a single server and cannot be used by others. The new generation of AI Fabric, according to GigaIO, offers even higher bandwidth and record-low latency, which is critical for training giant AI models that require intensive data exchange between accelerators. This not only increases overall system performance but also significantly reduces the total cost of ownership (TCO) through higher resource utilization and improved energy efficiency. This technology represents a crucial step in developing the infrastructure needed for the next wave of resource-intensive AI applications.
GigaIO Unveils Next-Gen AI Fabric at ISC High Performance 2025 Conference
