So after a bit of playing around, 3200 @ tightened timings is hard. But it seems easy enough to tighten timings by upping the clockspeed on the RAM and letting the motherboard adjust the timings automatically.
To achieve the above (unsure if stable just yet, will let you know when i’ve had time to check) I set the below manually and the left the rest.
CPU LLC: lvl2
Current capability: 120%
VDDSOC LLC: lvl2
Current capability: 120%
yeah that certainly seems to be the case, I wonder if the benefits of rank interleaving make going with a 2x16gb kit a decent choice if you want 32gb now? I’ve seen some people struggle with the 4x8gb single rank kits…
It’s a non QVL kit, i haven’t got any other RAM to test stuff with.
I haven’t tried too hard (well, not really at all with this bios, literally just changing the clock so far). Stock voltage, etc.
My timings are something like 15-15-15-35. I’m not too fussed with chasing extreme speeds. anything above 3000 is diminishing returns on Ryzen and i’m getting close now.
I was surprised at the difference in CPU performance just going from 3200 average timing to 3400 and tight timings to be honest, I tested the difference in mining lyra2z - got around 880Mh/s @ default 3200 timings. Consistent 1000mh/s @ 3400. Same power draw as well.
I didn’t have much luck with the DRAM calculator. Much better off just setting the main timings and pushing the clocks a bit, then when you hit a wall start tightening up the subtimings.
Did a standard firestrike test. Almost an identical physics score for me.
i just migrated my linux workstation to
asrock taichi ultimate x470
with gskill 4x16GB ram - i picked this sticks, since one swiss shop had them some weeks ago in special sale (got 4 for the price of 2,5 as i originally anyway wanted to get 2x16, but got lucky)
in bios i just took the XMP profile, which showed as option and set the speed to 3200 as suggested - works out of the box
here lshw -C memory output:
EDIT:
zu früh gefreut somehow the 3200 15 15 15 35 that the sticks XMP profile shows causes segfaults while using all 16 cores on some heavy computing. i reduced now to 2933 16 16 16 36 and trying stress tests to see if this is better… hmm
EDIT2:
yep, that did the trick: no segfaults for 15 min under heavy computing and compiling at the same time. since here in europe, we are having a heatwave since weeks and will be another week for sure, i will stop heating my office more doing stress tests - if someone wants me to test something, let me know (linux; windows and macos only in virtualboxes)