Astera Labs unveils next-gen CXL controllers to accelerate agentic AI workloads
Article excerpt
Highlighted: the sentence this signal was extracted from
X-Series to aid offloading KV cache for AI inference Astera Labs unveiled the next generation of its Leo Compute Express Link (CXL) smart memory controllers. The vendor added a trio of devices, with the high-end X-Series touted for offloading key-value (KV) cache for high-intensity AI workloads. Paired with Astera's Scorpio network switch silicon, it's designed to provide larger cache capacity with low-latency access for agentic AI inference applications. Astera also unveiled the second generation of its E-and P-series controllers to boost both central processing units (CPUs) and rack-scale memory. The E-series provides direct CPU-attached memory expansion, adding extra capacity while reducing the need for an additional socket. The second-generation P-series pools memory across rack-scale platforms, reducing server over-provisioning to maximize available capacity. Smart memory controllers are essentially infrastructure add-ons, providing CPUs and graphics processing units (GPUs) with the means to connect to additional memory capacity for an extra oomph. They keep accelerators from idling by increasing the memory bandwidth available to processing cores. "Agentic AI is where the economics of AI infrastructure are being decided and those economics depend on putting every usable gigabyte of memory to work," Thad Omura, SVP of Astera's compute connectivity group, noted. "The...
Keep reading with a free account
The rest of this article, and every signal for Astera Labs, is in your free account.
