New Ultrafast Memory Boosts Intel Data Center Chips

[erek]

"After several years of development with industry partners, Intel engineers have found a path to open that bottleneck, crafting a novel solution that has created the fastest system memory ever and is set to become a new open industry standard. The recently introduced Intel Xeon 6 data center processors are the first to benefit from this new memory, called MRDIMMs, for higher performance—in the most plug-and-play manner imaginable."

Source: https://www.techpowerup.com/328898/new-ultrafast-memory-boosts-intel-data-center-chips

 

[DanNeely]

From the article, while this is an Intel invention, they standardized it through JEDEC in 2022 a year after making their first internal prototypes.

This probably means that Intel's advantage here is at most probably going to be relatively short lived before AMD and various ARM vendors also implement it. It might be non-existent. It's unclear from me if being compatible at the physical socket level also means it's drop in compatible with all existing server platforms (save perhaps for their memory controllers not being able to clock high enough to take full advantage of it); or if it requires modifications to the memory controller to use.

I'm also curious how feasible it would be to extend down to desktop platforms. The core count growth over the last few generations of CPU has significantly outpaced increases in ram bandwidth; leading to memory bottlenecks being more common than they were previously. RDIMMs already have a buffer chip (they use is to only have 1 chip/dimm connecting to the memory bus vs consumer ones having every ram chip doing so) as the key feature enabling them to connect more dimms to a single memory channel before doing so causes signal degradation problems. That means it was probably easier to add this feature to them in the first place than it would be on dimms that don't have any sort of buffer chip at all currently.