The compute market has all the time been hungry for reminiscence bandwidth, notably for high-performance functions in servers and datacenters. In latest years, the explosion in core counts per socket has additional accentuated this want. Despite progress in DDR speeds, the out there bandwidth per core has sadly not seen a corresponding scaling.
The stakeholders within the business have been trying to deal with this by constructing extra know-how on prime of current widely-adopted reminiscence requirements. With DDR5, there are presently two applied sciences trying to extend the height bandwidth past the official speeds. In late 2022, SK hynix launched MCR-DIMMs meant for working with particular Intel server platforms. On the opposite hand, JEDEC – the standards-setting physique – additionally developed specs for MR-DIMMs with an analogous strategy. Both of them construct upon current DDR5 applied sciences by trying to mix a number of ranks to enhance peak bandwidth and latency.
How MR-DIMMs Work
The MR-DIMM customary is conceptually easy – there are a number of ranks of reminiscence modules working at customary DDR5 speeds with an information buffer in entrance. The buffer operates at 2x the velocity on the host interface facet, permitting for primarily double the switch charges. The challenges clearly lie in having the ability to function the logic within the host reminiscence controller on the larger velocity and conserving the facility consumption / thermals in test.
The first model of the JEDEC MR-DIMM customary specifies speeds of 8800 MT/s, with the subsequent era at 12800 MT/s. JEDEC additionally has a transparent roadmap for this know-how, conserving it in sync with the the enhancements within the DDR5 customary.
Micron MR-DIMMs – Bandwidth and Capacity Plays
Micron and Intel have been working carefully in the previous few quarters to deliver their former’s first-generation MR-DIMM lineup to the market. Intel’s Xeon 6 Family with P-Cores (Granite Rapids) is the primary platform to deliver MR-DIMM assist at 8800 MT/s on the host facet. Micron’s standard-sized MR-DIMMs (appropriate for 1U servers) and TFF (tall form-factor) MR-DIMMs (for 2U+ servers) have been certified to be used with the identical.
The advantages supplied by MR-DIMMs are evident from the JEDEC specs, permitting for elevated knowledge charges and system bandwidth, with enhancements in latency. On the capability facet, permitting for added ranks on the modules has enabled Micron to supply a 256 GB capability level. It have to be famous that some distributors are additionally utilizing TSV (through-silicon vias) know-how to to extend the per-package capability at customary DDR5 speeds, however this provides extra price and complexity which can be largely absent within the MR-DIMM manufacturing course of.
The tall form-factor (TFF) MR-DIMMs have a bigger floor space in comparison with the standard-sized ones. For the identical airflow configuration, this enables the DIMM to have a greater thermal profile. This offers advantages for vitality effectivity as properly by lowering the potential of thermal throttling.
Micron is launching a complete lineup of MR-DIMMs in each customary and tall form-factors right this moment, with a number of DRAM densities and velocity choices as famous above.
MRDIMM Benefits – Intel Granite Rapids Gets a Performance Boost
Micron and Intel hosted a media / analyst briefing lately to show the advantages of MR-DIMMs for Xeon 6 with P-Cores (Granite Rapids). Using a 2P configuration with 96-core Xeon 6 processors, benchmarks for various workloads have been processed with each 8800 MT/s MR-DIMMs and 6400 MT/s RDIMMs. The chosen workloads are notably infamous for being restricted in efficiency by reminiscence bandwidth.
OpenFOAM is a widely-used CFD workload that advantages from MR-DIMMs. For the identical reminiscence capability, the 8800 MT/s MR-DIMM reveals a 1.31x speedup based mostly…
