MRDIMM 8800MT/s vs. DDR5-6400 Memory Performance With Intel Xeon 6

For code compilation workloads the MRDIMM 8800 performance made only a very small difference. Not that surprising given that only when compiling very large codebases are all of the cores being fully utilized in parallel and code compilation doesn't tend to be as memory intensive as some other areas.

When turning to the OpenMP-threaded John The Ripper crypto program, the MRDIMMs began flexing their muscle and showing some uplift compared to DDR5-6400 DIMMs. Even with the DDR5-6400 memory, the dual Xeon 6980P processors were outperforming the current AMD EPYC Bergamo / Genoa(X) processors.

In other areas there were minor uplift out of the MRDIMM 8800 modules compared to DDR5-6400 memory. It will be interesting to see the pricing difference between DDR5-6400 and MRDIMM 8800 memory for justifying the performance gains.

When hitting the server workloads, okay, the MRDIMM performance is indeed hitting expectations... First HPC benchmark up was AMG and immediately the MRDIMM 8800 uplift came in at 30% over the DDR5-6400 memory. Same CPUs, swapping out to MRDIMMs gained 30% better performance in this memory intensive workload.

With the WRF weather research and forecasting model software, going from DDR5-6400 to MRDIMMs with Granite Rapids shaved off 6 minutes of the run time... Similar to Intel's reported claims of around 9% better performance with MRDIMMs.