Search results for Compute Express Link
AI & Enterprise
Meta develops \'Vistara\' chip technology to reuse memory from older servers
Meta has developed a technology called Vistara that removes DDR4 memory from older servers and installs it in new servers, using its own CXL chip to share memory between applications without latency problems, The Register reported. A Meta paper said about 40% of its servers, or millions of machines, cannot expand memory. Meta said the approach cut incidents of work stopping due to memory shortages by 33%.
AI & Enterprise
IITP to boost full-stack linking AI semiconductors and physical AI to respond to AX 2.0
South Korea\'s Institute for Information & Communications Technology Planning & Evaluation (IITP) is strengthening full-stack competitiveness by linking AI semiconductors, physical AI and next-generation networks, it said. It aims to respond to the spread of agentic AI and physical AI in industrial sites in the AX 2.0 era. IITP outlined R&D and talent development plans, including support for AI semiconductor firms, 6G and satellite communications, cybersecurity initiatives, defence AX programmes and expanded AI training.
AI & Enterprise
CXL technology gains traction amid memory supply crunch as Korean firms step up
Memory chip supply continues to lag demand as AI spreads, pushing big tech to look for alternatives. Compute Express Link, or CXL, is drawing attention by enabling servers in the same data centre to share memory. The Information reported Google has begun deploying CXL and installing controllers to manage traffic to external memory pools. Adoption remains limited, but Nvidia plans CXL 3.1 support and South Korean firms are advancing initiatives.
