Ideal Memory Utilization for Intel HPCG Benchmark

[Harsha_U]

Hi,

I am currently running the Intel-optimized HPCG benchmark and would like to understand the ideal memory utilization behavior for a given problem size.

Additionally, please provide a formula to estimate the memory footprint for any given problem size in HPCG.

[noffermans]

Hi Harsha_U,

Thank you for reaching out. In the below answer, I assume that ideal memory utilization means that the run would count as valid. Please let me know if the question was about something else.

As per the official HPCG guidelines (see https://github.com/hpcg-benchmark/hpcg?tab=readme-ov-file#valid-runs

A valid run must also execute a problem size that is large enough so that data arrays accessed in the CG iteration loop do not fit in the cache of the device in a way that would be unrealistic in a real application setting. Presently this restriction means that the problem size should be large enough to occupy a significant fraction of main memory, at least 1/4 of the total.

---

When running the benchmark there will be a "Memory Use summary" in the log file that will look like this:

...
########## Memory Use Summary ##########=
Memory Use Information=
Memory Use Information::Total memory used for data (Gbytes)=XX.XXXX
Memory Use Information::Memory used for OptimizeProblem data (Gbytes)=0
Memory Use Information::Bytes per equation (Total memory / Number of Equations)=XXX.XXX
Memory Use Information::Memory used for linear system and CG (Gbytes)=XX.XXXX
...

Unfortunately, the memory used for the OptimizeProblem data is currently not reported (it will always be 0, even though some optimization was performed). However, it is fine to use the "Memory used for linear system and CG" value to assess the 25% memory usage rule. I would therefore recommend to experiment and find a problem size that satisfies the minimum memory usage on a given machine.

Best,
Nicolas