The authors propose a high-capacity cache architecture that leverages emerging high-bandwidth memory modules. High-capacity caches capture the secondary data working sets of scale-out workloads while uncovering significant spatiotemporal locality across data objects. Unlike state-of-theart dram caches employing in-memory block-level metadata, the proposed cache is organized in pages, enabling a practical tag array, which can be implemented in the logic die of the high-bandwidth memory modules.