Specifications
Vendor
Montage Technology
MRCD Part Number
M88MR5RCD01
MDB Part Number
M88MR5DB01
Product Type
1x Multiplexed Registered Clock Driver (MRCD) + 1x Multiplexed Data Buffer (MDB)
Generation
Gen1 DDR5 MRDIMM / MCRDIMM chipset
Data Rate
8800 MT/s (Gen1)
Gen2 Platform Rate
12800 MT/s, 2x the theoretical bandwidth of DDR5-6400 RDIMM at the same DRAM speed
Multiplexing Scheme
Time-division multiplexing with double data rate — two memory arrays accessed per clock
Role of MRCD
Coordinates and multiplexes rank data streams, driving the DIMM clock and command bus
Role of MDB
Caches and multiplexes multiple DRAM rank data streams before driving the CPU data bus
Module Types
DDR5 MRDIMM and MCRDIMM
Module Capacities
64 GB, 96 GB, 128 GB and 256 GB per MRDIMM module
Companion Chips
M88SPD5118 SPD Hub, M88TS5110 temperature sensor
CXL Compatibility
Pairs with CXL memory expansion to build tiered high-bandwidth local + pooled capacity memory
Latency Effect
Reduces access latency versus standard RDIMM rank interleaving
Primary Workloads
Generative AI inference, RAG retrieval, agent memory, KV-cache residency, in-memory analytics
Server Class
AI server, general-purpose server, storage server
Availability
Contact for quote — production quantities, MOQ and lead time on request
Overview
Montage Technology's DDR5 server memory-interface family is the silicon that sits between a server CPU's memory controller and the DRAM ranks. The M88MR5RCD01 and M88MR5DB01 are the first-generation MRDIMM/MCRDIMM chipset: the MRCD (multiplexed registered clock driver) and MDB (multiplexed data buffer) that together convert a standard DDR5 module into a multiplexed module capable of serving two ranks per clock.
The mechanism is time-division multiplexing at double data rate. The MDB caches and multiplexes the data streams from several DRAM ranks while the MRCD coordinates the transfer to the CPU. At Gen1 the chipset runs at 8800 MT/s; Montage's Gen2 MRDIMM platform raises this to 12800 MT/s, which the vendor states doubles the theoretical bandwidth of a DDR5-6400 RDIMM built from the same DRAM speed grade — without needing faster DRAM particles.
For AI infrastructure this matters because generative AI inference is frequently memory-bandwidth bound rather than compute bound. Wider memory bandwidth speeds the reading of model weights, KV cache and intermediate activations between CPU and main memory, so the installed accelerators stay fed. A single MRDIMM module carries 64 GB, 96 GB, 128 GB or 256 GB, letting a node scale into the terabyte range without de-populating DIMM slots.
MRDIMM is also designed to decouple cleanly against CXL. A common server topology uses MRDIMM as the low-latency, high-bandwidth local tier holding hot weights and KV cache, and CXL memory expansion as the capacity tier for cold parameters — which is exactly how QS Compute sizes memory subsystems for edge and datacenter inference deployments.
Key Benefits
Double the DIMM bandwidth without a faster DRAM speed grade — Gen2 MRDIMM reaches 12800 MT/s. Up to 256 GB per module so a node scales to terabyte memory in fewer slots. Full suite sourcing — MRCD, MDB, SPD hub and temperature sensor from one vendor for drop-in module designs. Tiered-memory ready — designed to pair with CXL expansion and pooling.
Applications
DDR5 MRDIMM and MCRDIMM module designs · generative AI inference nodes and RAG servers · KV-cache and agent-memory residency · in-memory analytics and large language model serving · general-purpose and storage servers where memory bandwidth per core is the bottleneck
Request a Quote — MONTAGE M88MR5RCD01 / M88MR5DB01 — GEN1 DDR5 MRDIMM MRCD AND MDB CHIPSET
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →