Z790 Hero - DDR5 G.Skill 6000 64GB QVL - unstable XMP and low tREFI

[woopsie]

I'm running Asus Maximus Z790 Hero + i9-13900K + G.Skill Ripjaws S5 64GB (F5-6000J3040G32GX2-RS5K). BIOS version is 1501. This RAM is on motherboard QVL list

After enabling XMP I, XMP II or XMP Twaked with DDR5-6000, the system is sneakily unstable. MemTest86 10.6 PRO - 4 passes(13 tests, 24 cpus parallel) without errors. TestMem5 with Extreme1@Anta777 no errors after 3h. Also tried OCCT test, Prime95, y-cruncher with VST and VT3 - no errors.

But sometimes I got errors (last one was 0xC0000005) when starting Red Dead Redemption 2. Also in Event Viewer I got warnings i8042prt id 19, id 20 and id 17. Sometimes(rarely) there are also 1-2 second mouse freezes when starting TestMem5 or starting MSI Afterburner.

System Windows 10 22H2. I also worked without XMP for a while (2 weeks) and did not get these errors.

CPU scores: SP 96; P-SP 107; E-SP 76; MC SP 64.

Currently I'm running with DDR5-5600 and manually applied CL 28, RCD 39, RP 39, RAS 91, RFC2 449, RFCsb 365, DRAM VDD 1.3V, DRAM VDDQ 1.3V, MC Voltage: 1.3V, PMIC Voltages: Auto. No error so far. See screenshot.

However, I am annoyed that the system does not work stable with such a small OC as 6000.

Is the processor bad? Is the motherboard malfunctioning? What could I do to get the 6000 running stable? Also why the tREFI [Auto] value(5461) is so low?

[woopsie]

Anyone?

[Silent_Scone]

Hi @woopsie

Have you checked the MC SP test under AI Features? The score should give us a rough indication of where you need to aim.

Tuning of the associated memory rails can sometimes be needed. Start by increasing the Memory Controller Voltage. Values can range between 1.2 - 1.5v

You can inspect the auto rules within the UEFI and then work up and down incrementally to see what aids stability.

Extreme Tweaker > CPU System Agent
Extreme Tweaker > Advanced Memory Voltages > Memory Controller Voltage
Extreme Tweaker > Advanced Memory Voltages > IVR Transmitter VDDQ

[woopsie]

Have you checked the MC SP test under AI Features? The score should give us a rough indication of where you need to aim.

---

Yes, I have checked MC SP which is 64.

[woopsie]

This is some kind of nightmare.

I do some research. Warnings i8042prt - only shows when Legacy Game Compatibility Mode is Enabled in BIOS. BIOS creating some virtual PS/2 keyboard probably for handling ScrollLock. Therefore, I think that these warnings cannot be taken into account in stability testing.

Moreover, it turned out that the memory at these settings from the first post was unstable.

Currently I would like to stabilize DDR5-5600 if possible.

So I have turned off computer and cutted power. Next I turned on the power strip and pressed Reset button for Safe Boot. In the BIOS, I have set:

• XMP II (30-40-40-96)
• DDR5-5600 frequency
• I left the settings DRAM voltages (VDD, VDDQ) on Auto (which is 1.4V),
• I left the Memory Controller Voltage on Auto (which is 1.35V)

And guess what? Unstable in Prime95 (after some time: 5h - internet browsing(light tasks) on Linux; then: shutdown without power cutoff and turn on computer with Windows and Prime95). I ran the Torture test - a large FFT(see screenshot), but one or two Prime95 workers did not allocate memory, and one CPU core had almost zero use. Then I executed "Stop" but these two workers did not stop and when I pressed "Exit" nothing happened - I could not exit Prime95.

I have now increased the Memory Controller Voltage from [Auto] to 1.37V. And I am testing and waiting when some instability appears (I suspect it will appear).

If instability occurs what should I do next? Increase MC Voltage more or something else? Please help.

[Silent_Scone]

Why are you running Prime? Use an actual memory stressor such as Karhu RAM Test or TM5, and not a power virus. 

64 is on the low side, I would consider not hitting the CPU continually with small FFT as you’re bound to degrade the CPU faster. Instead, focus on using the machine and stabilise a test such as Karhu.

MCVDD, Increasing voltage is not always the answer with signal related voltages. 

[woopsie]

I'am also using TM5 not only Prime95.

I currently use Large FFTs in Prime95 not small FFTs. I ran the Small FFTs test very rarely and for a short time.

---

@Silent_Scone wrote:

MCVDD, Increasing voltage is not always the answer with signal related voltages. 

---

What do you mean by 'signal related voltages'?

[Silent_Scone]

Any voltage that has an impact on the memory IO, including MCVDD is susceptible to change which is why we refer to “tuning” and not simply increasing.
The VCCSA is the voltage for the System Agent which is responsible for the memory controller and other IO interfaces, and as such is sensitive to voltage adjustment.

I would focus on stabilising memory orientated tests that load and isolate the memory subsystem sufficiently such as MT5 and Karhu, rather than Prime.

[PaulyFromMichia]

Im not an overclocker nor understand how to play with the memory setting like you guys are but I was reading this and is there any chance your using a contact frame? I had unstable memory problems with the frame. I removed it and no more problems...

[Silent_Scone]

An improper torque can cause this behaviour, yes, as can overtightening the CPU cooler. The contact frame just improves upon the stock mechanism in some cases to improve contact, but it's not always guaranteed to help