Home > Memory Interface Chips > DDR5 DIMM Chipset > DDR5 SPD Hub with Internal Temperature Sensor
The Rambus SPD Hub (SPD1605Gxx) enables DDR5 Registered DIMMs (RDIMMs), MRDIMMs, Unbuffered DIMMs (UDIMMs and CUDIMMs), Small-Outline DIMMs (SODIMMs and CSODIMMs), LPDDR5 SOCAMM2, and LPCAMM2 memory modules.
The Rambus DDR5 Serial Presence Detect (SPD5) Hub device contains 1024 bytes of non-volatile memory arranged as 16 blocks of 64 bytes per block. Each block may optionally be write-protected via software command. Write protection for each block may be overridden in an offline programmer environment while overrides are prevented in normal use. The SPD5 Hub device operates from 1.8V nominal power supply input. The SPD5 Hub device is intended to operate up to 12.5 MHz on an I3C Basic bus or up to 1 MHz on an I2C bus. The SPD5 Hub device is intended to interface to I2C/I3C Basic buses which have multiple devices on a shared bus and must be uniquely addressed with fixed addressing on the same bus. The SPD5 Hub device responds to specific pre-defined I2C/ I3C Basic device select codes on a host interface bus. The SPD5 Hub device also incorporates a second local I2C/I3C Basic bus and pass-through of commands from the host bus onto the local bus for addressing of I2C/I3C Basic devices on the local bus (Hub function). The SPD5118-Gxx device incorporates thermal sensing capability which is controlled and read over I2C/I3C Basic bus.
Agentic AI is reshaping data center architectures, introducing persistent, memory-intensive workloads that demand unprecedented bandwidth, capacity, and efficiency from system memory. Unlike traditional AI/ML pipelines, agentic systems operate in continuous, context-rich loops that amplify the importance of main memory as a critical performance enabler. This white paper explores how DDR5 DIMMs address these emerging requirements, delivering significant advancements in data rate, density, power efficiency, and channel architecture over DDR4. It examines the role of DDR5 as the foundational memory layer in modern AI infrastructure, complementing heterogeneous compute architectures and supporting highly concurrent, data-driven workloads, while also outlining the key design considerations required to achieve reliable, high-performance operation in next-generation servers.
