Originally Posted by The Stilt
Higher DFICLK itself doesn't improve the performance. Increasing the DFICLK (by increasing the MEMCLK) only helps if the DFI is the bottle neck.
1600MHz DFICLK (i.e. 3200MHz MEMCLK) is sufficient for > R7 1800X and the returns from higher speeds are diminishing. At higher CPU speeds (which are unachievable on current gen. Ryzen CPUs) the required DFICLK for optimal performance will increase as well.Warning: Spoiler! (Click to show)
Latency is king >= 3200MHz.
, thank you as always
Only my opinion from your last set of results, posted here
. 3200MHz C12 @ 1.45V-1.5V may not be something all attain. The delta of 3200MHz C12 is ~1.6% vs 3520MHz C14, so user may find something else is optimal in the context of what user can attain IMO.
Any chance of info on VDIMM you've used for above and past linked dat, cheers
My own testing is echoing that the 4% MEMCLK gain on 3333MHz vs 3200MHz yields nothing except gain in AIDA64 bench and possibly MLC (yet to view data but in zip below). Which is useless "paper performance gain"
RAW data zip link
UEFI 9943 used, same timings per MEMCLK, settings and old data here
Originally Posted by noko59
I am going to have to retest BGSA to confirm if I actually got an improvement in both SD combine and Aida64 Mem & Cache. You have very accurate testing methods and recordings.
Last run Aida64 crashed at 1hr 15min. So I am running now with Cad Bus at 30 Aida64 Stress testing again (3500, BCLK 109.4, 3.9ghz PS 0).
, I try what I can with my limited knowledge/tools
. I must do SD for those tests
, be great to compare data
20ohms had errors quicker for me than 30ohms. 24ohms seems best, as 60ohms has gone iffy as well
. I may now just go back to UEFI 9943 where this profile did work for HCI Memtest and other stuff, as it maybe as The Stilt has said the UEFI 1401 PMU FW is not as good as UEFI 9943 for tightened RAM stability.