Search results for CXL 2.0
Industry
HBM supply hits limits as memory race shifts from bandwidth to utilisation
Limits on expanding high-bandwidth memory are pushing the memory industry to focus on improving resource utilisation in AI servers rather than boosting bandwidth alone. With only about 50 to 60 percent of installed server memory used for computation, companies are pursuing technologies to pool and reallocate capacity, including CXL, PIM and high-bandwidth flash. Supply constraints and rising costs reinforce the shift toward a tiered memory approach combining different memory types.
Industry
AI bottleneck shifts to servers, pushing faster eSSD testing
Investment in AI data centres is expanding from high-bandwidth memory to enterprise SSDs, with prices for 30TB to 64TB drives up more than 50 percent. As capacity and NAND layers rise, burn-in and final tests take longer and inspection becomes a bottleneck, driving demand for higher-parallelism, PCIe 5.0-capable equipment. New testing needs are also emerging around CXL 2.0 memory pooling and switching, prompting shipments of dedicated testers.
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%.
