One of the exciting elements of Intel’s Xeon 6 Granite Rapids launch last year was introducing support for MRDIMMs alongside DDR5-6400 memory support. After the Xeon 6900P series debut I posted some of the first independent DDR5-6400 vs. MRDIMM-8800 benchmarks. One year later, today is a fresh look at the DDR5-6400 vs. MRDIMM-8800 performance for Granite Rapids with new/updated benchmarks, the latest Linux software improvements, and also looking at the impact on power and thermals of MRDIMM memory.
Now that I am back up and running with a working Intel Xeon 6900P Granite Rapids server thanks to Giga Computing, one of the follow-up articles I had been wanting to do is revisiting the DDR5-6400 vs. MRDIMM-8800 performance with the flagship Xeon 6980P processor. The Giga Computing R284-A92-AAL1 supports 12 channel DDR5-RDIMMs as well as MRDIMMs. While this is a dual socket platform, for this round of testing only one of the Xeon 6980P processors was populated and the respective memory. A full review on this Giga Computing server will be published on Phoronix in the weeks ahead.
With a single Intel Xeon 6980P processor I ran the same set of benchmarks using 12 x DDR5-6400 64GB DIMMs (Micron 64GB 2RX4 PC5-6400B-RA1-1211-XT) and then 12 x MRDIMM-8800 modules (MTC40F2046S1HC88XDY / PC5-8800X-HA0-1110-XT). Thanks to Intel for having supplied the memory modules as part of my original Intel Xeon 6980P Linux testing. Both sets of modules worked great and without any troubles on the R284-A92-AAL1 server.
CPU vendors typically ask to refrain from providing any system power “wall power” measurements for their reference servers given they aren’t properly tuned like production OEM/ODM servers and often have rather excessive fans with aggressive cooling profiles, which is why there were not any power numbers in my original MRDIMM vs. DDR5-6400 article last year. Now that there is the Giga Computing GNR-AP server in the lab it’s fair game for power monitoring. So now I can provide measurements of the overall system power consumption to see the difference MRDIMMs have on the overall server power consumption.
The memory temperature was also reported using the DIMM_G temperature sensor as exposed via the BMC on this Giga Computing server.
Beyond thermals and power consumption data included, a number of new/updated benchmarks were also included as part of this fresh memory comparison on Granite Rapids. In addition, an up-to-date Linux software stack of the Ubuntu Server 25.10 daily state with the Linux 6.17 kernel and GCC 15.2 compiler. So a very fresh look at DDR5-6400 RDIMM versus MRDIMM-8800 performance now nearly one year after Granite Rapids initially debuted.
With the time that passed, we also now have a better idea for MRDIMM pricing… As of writing the Micron 64GB MTC40F2046S1HC88XD1 DDR5-8800 ECC MRDIMMs can be found for around $450 USD per DIMM in-stock at different online retailers. In comparison, the Micron 64GB MTC40F2046S1RC64BDY DDR5-6400 memory modules can be found for around $350 USD per DIMM. So around $100 USD per DIMM more for going from DDR5-6400 to MRDIMM-8800, or +$1200 USD if populating all twelve available memory channels and obviously +$2400 for a dual socket server at list prices.
