Memory and CPU settings

[Hopper64]

I think I am near the final settings on this 14900ks. Ended up with LLC4, AC .2, DC 1.02, 1700 per Braegnok as decreasing caused WHEA errors. Power is sub 300W and temp is about 74, but noticed only about 40K in CB23. I suspect that's due to calibration. 

Memory ended up with c38 and refresh I of 131K. Had errors with DIMM flex and tighter timings but I only tested a few profiles for DIMM flex. Could have spent more time with those. Thanks to Scone and Braegnok for the help!

[Hopper64]

BTW-I keep seeing references at OC about an "SA" bug. I suspect that's an issue with 14900k processors not requiring as much system agent voltage compared to 13th gen? I dialed back the SA to 1.22V manual after having a couple of memory training issues on post. It would stop at 55 and sometimes proceed to boot, other times not. Since reducing the SAV to 1.22V even cold boot seems to work ok. So is this bug too much or too little SAV?

[Braegnok]

Some folks reported issues when using above 1.23v SA, most of those SA-bug chips are stable however using lower SA voltage.

It seems having SA limit of 1.23v would effect ability to run higher frequency profiles.

My system defaults SA at 1.25v for xmp 8200.

And default runs 1.329v SA for 8600 CL40.

Adjusting, running lower SA VID, VDDQ TX, and IMC VDD is almost always the best way to remove TM5 errors when testing, fine tuning your memory profile vs running default/auto settings.

It all depends on your CPU IMC, quality of the RAM ICs and motherboard. Some systems are plug & play at any frequency with everything in auto,.. and some struggle, require adjustments with all voltages even for 8000MTs.

[Hopper64]

I think I have experienced the SA bug. Everything seemed stable in ramtest but I am still getting freezes especially when windows starts. Nothing in the event viewer. Noticed people referring to the SA bug at OC so I reduced SA voltage in 20mV intervals from 1.28V. I’m currently at 1.16V and TM5 stable for 1 hour 24 minutes with no errors. Hard to believe. I was shocked. Still testing and might reduce SAV further. Noticed one error in Ramtest at 45 minutes with SAV 1.16V so not sure what that means. Still testing and reading the forums….

[Silent_Scone]

Hi Hopper,

At this point it's fairly well documented that some CPUs aren't comfortable with more than 1.23-1.25v VCCSA - silicon variance is always a reality. Having not had a sample that has this problem I'm not able to comment on how this restricts the CPUs frequency range.

[Hopper64]

I'm gradually reducing the SAV and finding it's stable so I will continue to test this and update later. The weird freezes seem to be less frequent but I need to watch it for a few weeks.

[Braegnok]

You can test your CPU, find out quickly if it does indeed have the SA-Bug.

Simply load Default Settings, and bump the SA voltage up to 1.35v, run AIDA64 Cache & Memory Benchmark, the benchmark will instantly crash if you have the bug. 

Default settings 5600MTs, SA VID 1.35v,.. no bug. 👍

[Hopper64]

Mine was running 1.249V default SAV so when I noticed some training problems on boot, I simply reduced it to 1.22V and tested Ramtest again. It was fine with that. No errors. I suspect the 20-30mV decrease helped but I'm going to watch it a few more days. I noticed the reduced voltage you had applied in your 8400 profile. Thanks!!

[iBruceypoo]

Great work guys, way over my head but great work all the same.  😊

[Hopper64]

Spoke too soon. Getting a lot of instability with memory on 9903. Default works fine of course. XMP1 does not pass memtest in the bios. No other changes except for XMP1 only. Adding a little VDD/VDDQ at 1.42V with XMP1 does not pass. So I was thrown off by this. It’s likely something I am missing in my OC. SAV at 1.22V does not pass with XMP1 and tightened timings with VDD/VDDQ at 1.5V. I’m going to tinker with it work permitting and report back.