Manual tuning as a remedy?
In addition to the operating voltages of the RAM, VDD and VDDQ, several others are also relevant for DDR5 for a successful memory training and stable operation. VDD2, CPU VDDQ and the SA voltage must also be in a sweet spot, which even often varies depending on the combination of mainboard, RAM kit and CPU. While the former is part of the XMP profile for DDR5, the other two are automatically set by the motherboard to a value that the manufacturer deems suitable.
Now you could start here and, for example, start from the Auto values in 25 mV steps until you have found a better working combination. But exactly here follows a second pitfall of the mainboard, because it simply can’t measure or read VDD2 and CPU VDDQ. Neither in the BIOS nor in monitoring software like HWInfo are current values of these voltages found, which means that there is no starting point and control mechanism for manual tuning.
New values can be set, but whether they are actually applied is not traceable. The values in the right column are just the default values that the board overwrites itself when loading the XMP profile. Thus, operating DDR5 RAM beyond the JEDEC specification is ultimately a complete blind operation on this motherboard. And exactly because of that, it would be even more important that buyers of a rather cost-optimized motherboard can really rely on their QVL.
For the sake of completeness, I will also mention that BIOS versions other than the latest F7b have of course been tried, but unfortunately without any improvement in behavior. What is left for a customer who is told by support that his CPU is to blame, but who doesn’t want to run his expensive DDR5 kit in JEDEC mode only? Well, loading the XMP profile and clocking down. I did exactly that and lo and behold: DDR5-5866 boots without a hitch, with all voltages on auto.
However, booting does not mean that the system is stable. Again and again single errors in various RAM stress tests occurred with this setting. A user who is happy about a finally booting system with relatively fast DDR5 probably lulls himself into a false sense of security right here, with possible file corruption as a consequence. By the way, if you suspect that heating up of the RAM might be to blame here, I can answer to that right away: There was a fan on the modules, which reached a maximum of 43 °C, far away from the 85 °C actually specified for DDR5 ICs. Only with DDR5-5600 manually set, the system finally also worked stably in dual-channel, at least with this RAM kit.
To narrow down the cause further, I now tried to run slots and modules individually in the XMP profile. In fact, this is not a problem, no matter which module and no matter whether in Channel A or Channel B, in single-channel operation the board already boots successfully after a few training attempts. Also, Testmem5 does not immediately throw errors as one would expect at the supposed limit of the hardware. This leads to the conclusion that neither the DIMM slots nor the RAM modules are the cause for the incompatibility. Instead, the BIOS seems to be overwhelmed with the training when both channels are occupied at the same time.
28 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Urgestein
1
Urgestein
Urgestein
Mitglied
Veteran
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Veteran
Mitglied
Urgestein
Mitglied
Alle Kommentare lesen unter igor´sLAB Community →