XMP profile still unstable on Ryzen even after successful stress tests

[MarkR17]

I am currently tring to get the 3600Mhz RAM stable which is the XMP settings. But I am still getting the bluescreen (UNEXPECTED_KERNEL_MODE_TRAP), which indicated some kind of RAM failure?

 

The steps so far:

 

- I enabled the XMP profile, after many, MANY problems of getting the computer to run at all. I can link the thread that involves that if required

- The CPU itself is completely untouched, given the advice that overclocking any Ryzen 3000 CPU is not worth the effort and heat.

- Prime95 [Blend] fails within a minute with XMP enabled

- Increased the RAM voltage to 1.4V, which is supposed to be the max safe voltage for DDR4 at 24/7 operation. This may have fixed *most* of the problems at first glance.

- Prime95 [Blend] (latest version with AVX on) runs without errors for 12 hours

- HCI Memtest also runs without errors for 8 hours.

- During normal use, my computer still bluescreened. It happened while browsing the web.

- CPU was never above 80C during these stress tests.

 

I still got a bluescreen followed by a lockup by my computer which required a full CMOS reset to fix.

 

Also, when I tried to manually enter the 3600Mhz XMP settings in the BIOS the computer crashed afterwards and needed a CMOS reset.

For now I am rather unwilling to mess with these settings, at least for now

 

One major caveat is that if I do really have to stress test the computer for very long, do I really need to run that for 24 hours?

I do need to use the computer for other things.

Another thing I noticed is that the stress tests fully uses the CPU which means they do not take light CPU usage, like web browsing into account(?). I heard that Prime95 not a good stress test.

 

I am aware that my current combination of motherboard and RAM is not exactly supposed to run like that but yet it seems so close to fully working.

The RAM was marketed as Intel-compatible, not AMD-compatible but I dismissed that.

This RAM is also not on the QVL list for that motherboard which officially only runs up to 3200Mhz RAM, but that figure is for 1st Gen Ryzen.

 

More questions:

- What can I do to improve memory stability further, in context with the Ryzen 3000 series CPU?

- I heard about increasing the SOC voltage by a bit and the RAM voltage. What is the safe limit for daily use?

- Can RAM sticks get too hot? There are no temperature sensors for RAM.

- I may think of upgrading to a new motherboard, is it worth it? I am not planning to upgrade to a 5000 series CPU.

- Are there any other good tools for memory stress testing?

 

I am willing to lower the RAM speeds if getting that XMP profile stable is not an option.

 

CPU: AMD Ryzen 7 3800XT (completely at stock ignoring XMP. 5600X is unavailable)

MoBo: Asus ROG Strix X370-F

RAM: GSkill RipJaws V DDR4 3600 16-19-19  32GB (at 1.4V from 1.35V, XMP)

GPU: Radeon RX 570

SSD: Samsung EVO 850 500GB

PSU: beQuiet Straight Power E10 600W

 

Other notes:

- Latest drivers are installed, including chipset drivers

- Currently using Ryzen Balanced Power Profile

- Windows 10 was not freshly installed, same installation as previous Ryzen build

- Upgraded from Ryzen 5 1600 and cheap 16GB RAM.

 

EDIT: Someone who can move that thread to Troubleshooting please, I screwed up again. Turns it is a troubleshooting issue.

 

[Guest]

Enable XMP, manually set 3200mhz and test. If stable, leave it at 3200mhz. 

 

(IMC supports up to 3200mhz, so you start here and then OC up._)

[MarkR17]

After playing a bit arround, some even weirder findings:

 

- Decreasing the SOC voltage by a bit, but not by too much actually improves day-to-day stability by a bit but not enough, it still bluescreens rarely.

- Setting the LLC levels to 3 (out of 5, equivalent to Medium) also made it less stable for day-to-day use.

- Setting RAM to 3200 Mhz did NOT fix the bluescreens

- Windows power plans also affect stability by a bit, the recommended AMD Ryzen profile caused  trouble.  Win10 High Perf. profile seems to be the most stable, relatively

- The unofficial 1usmus power plan actually made it even less stable, even after applying recommended BIOS settings, these are undone.

 

- Blue screens only occur when idle or during very light use, like browsing the web. They never occur during stress tests or gaming.

I pretty much used every stress test that is freely available and they never failed, even when used overnight. I am aware however that these may not represent real-life usage.

However, blue screens do not occur during gaming.

 

Some options:

- Cleaning Windows 10 or maybe even completely reinstalling it. I already did not have good experiences with Windows 10 but at least it did not bluescreen until now.

-- FYI the Windows 10 installation is 3 years old at that point. Isn't it recommended to reinstall Windows every now and then?

- Disable the Ryzen 3000 boosting behavior, I suspect that is the cause of the bluescreens, since the clocks and voltages are jumping all over the place.

-- I assume using a fixed multiplier (lets say 39x) should work?

- Is there any way to disable power saving options for that CPU?

 

of course i could just disable XMP altogether but then I would leave a lot of performance on the table.