Intel Xeon 6 Granite Rapids Memory Scaling Performance From 6 To 12 MRDIMMs
👁 Xcompact3d Incompact3d benchmark with settings of Input: Channel FLow. 8 x MRDIMM was the fastest.
It depends upon the software as well as the size of model and other factors, but for a number of the HPC workloads the performance benefits dropped off after eight MRDIMMs with this single socket Xeon 6980P test setup.
The OpenFOAM computational fluid dynamics software scaled nicely up to twelve MRDIMMs with the larger models.
For those trying to reduce costs with today's memory pricing, going for eight MRDIMMs at least initially would be one way to achieve most of the performance without as much of an initial investment.