Astera Labs' Leo X-Series memory controllers boost long-context inference TTFT by up to 62%

BenBajarin · x · 2026-09-16

Astera Labs expanded its Leo Smart Memory Controller family for agentic AI and cloud workloads. The new Leo X-Series, paired with Scorpio X-Series fabric switches, enables fabric-attached memory for long-context, KV-cache-intensive inference — up to 62% faster time-to-first-token and 22% more tokens per second. Leo 2 E/P series add CPU-attached DDR4/DDR5 expansion with PCIe 6 and CXL 3.2, doubling bandwidth and capacity, plus reliable reuse of legacy DDR4 to cut TCO. Analyst Ben Bajarin highlights the X-series letting GPUs reach more memory without routing KV-cache through the CPU memory subsystem.

Original post →

More from Infra

Infra channel →