Hello,
Yesterday on discord there was once again a discussion about there being no point in pushing older platforms to their limit in memory benches, though the focus was on memory frequency. Memory benches can output wildly different results on different platforms despite being ran at similar memory settings (pyprime AM4 vs LGA 1200 for example), making it not even worth it to run them on many platforms.
Thus we arrived at the following suggestion:
Change memory benchmarks from hwpoints per DDR generation + globals overall to hwpoints per CPU Socket and memory globals from overall to per DDR generation.
An alternative suggestion of per CPU generation was also thrown into the room, but that one might be much harder to implement. I've added it to the poll anyways.
Following this suggestion would also have global rankings no longer be the exact same ranking as WR ranking, similar to how it is with multicore benches. Globals per DDR generation might also motivate some people to push for some of the older DDR* records which have remained unbroken for years at this point.
- why not per memory IC? There is no required input field for it - it would be impossible to track for older scores and would require massive effort collecting the data for older scores if possible at all. Also very easy to wrongly identify some IC.
- why not per chipset? Motherboard was never required as an input field for memory benchmarks. It would require a lot of manual work sorting through all the scores missing the motherboard entry.