Originally Posted by chakku
Originally Posted by diaaablo
In the beginning I really hated 1501, but after discovering correct CLDO_VDDP it runs smooth and easy. It takes time, but it definitely worth it to find your own. And stop crying, I hope soon Elmor will make another "from-the-box" bios for you guys.
What's a good technique for testing this value and what value range do you test between?
Do you just set up strap/timings you know work with CLDO_VDDP at Auto and tweak it there or do you assume a certain strap/timings should work and just test different CLDO_VDDP values until it boots properly?
I may not be doing justice to it and my recall may be influenced by what I did, but Ramad's method was to lower voltages (say, VDRAM) to where the error rate by some test is noticeable, and then for a given proc_ODT vary CLDO_VDDP and observe whether the system can POST. You should find regions that POST and regions that don't POST. Those that don't can be declared memory holes for this purpose. Do this for proc_ODT values that make sense (that others are successfully using) for your particular brand of DRAM. Once settling on a proc_ODT, choose a CLDO_VDDP region that POSTs over a fairly wide zone (at least +/- 10 mV) and set CLDO_VDDP at a middle value. Then raise VDRAM to 1.35V or so and do your stability tests.
Of course, it is necessary to be operating at the desired DRAM frequency and that the DRAM timings are in the ballpark.
Find a recent message of mine restating Ramad's scheme for getting CLDO_VDDP to "stick" after each change.
There are other voltages that affect bit reading accuracy, so you will need to troll through a lot of posts here to see how they affect things. My view is that all posts made after the announcement of the BIOS you intend to use should be viewed for clues to achieving stability.
This is a vary tedious effort best started from a baseline where others with the same DRAM have been successful.
While AMD may have sold Ryzen with the strong suggestion of great overclocking potential
, they are only going to "support" default performance. What we know from this thread is what we -- the Asus C6H beta testers -- have discovered through trial and error in concert with help from Asus employees. While improving AGESA's and improving UEFI/BIOSes may help over time, maybe even resulting in DRAM XMP tables that are relevant to Ryzen (don't bate your breath), there is too much variation among components to establish a "do this and it will work at 3333 MT/s" set of directions. I think a lot of our successes here are accidental (except for Ramad's and The Stilt's, who appear to have the time and patience for methodical testing).