Z370 Extreme4 XMP bug

I recently bought Z370 Extreme4 mobo and after upgrading the BIOS to 1.30 XMP stopped working properly. Windows would randomly freeze (almost always after stopping a CPU benchmark). Turning off XMP profile worked and I didn't experience any crashes after that. Later I downgraded BIOS to 1.22 and again turned on XMP. After the downgrade everything was fine so I'm sure there is an issue with the newest BIOS.

wojwen wrote: I recently bought Z370 Extreme4 mobo and after upgrading the BIOS to 1.30 XMP stopped working properly. Windows would randomly freeze (almost always after stopping a CPU benchmark). Turning off XMP profile worked and I didn't experience any crashes after that. Later I downgraded BIOS to 1.22 and again turned on XMP. After the downgrade everything was fine so I'm sure there is an issue with the newest BIOS. My specs:  CPU i7-8700K Motherboard ASRock Z370 Extreme4 RAM Corsair Vengeance LPX 16GB 3200MHz GPU MSI GeForce GTX 970 GAMING 4G Storage Samsung 850 Pro 256GB, WD Blue 1TB PSU Corsair RM650X Cooling Noctua NH-D15S Operating System Windows 10 Pro

This is the first report of this situation with your board and UEFI version, although it has only been available for a week.

While it might be an issue/bug, it is also possible that this occurs with your model of memory, but not others. That could still be a bug with your memory in particular.

You said that XMP stopped working, but what does enabling the XMP profile do? It applies a built in over clock of the memory speed, sets some of the memory timings as needed for that speed, and increases the DRAM voltage. When the XMP profile is not enabled, the memory is running at its stock, default speed which is easier on the memory controller.

What is really happening is your memory overclock seems to be causing Windows to freeze randomly after closing a CPU benchmark.

Yes, the previous UEFI version with XMP enabled did not seem to cause the freezing in Windows. Are you absolutely certain that the UEFI option settings you used between the previous and new UEFI versions are absolutely identical? Including manually set voltages? Any comparison between the two must be identical. Can you see any differences in the memory settings and memory related voltages between the two UEFI versions with the XMP profile enabled?

If you believe there is an issue, please file a support request with ASRock on this page, or via the support site for your region of the world:

https://event.asrock.com/tsd.asp

Hello and sorry me again,

i found out the the xmp profile on the 1.3 bios for extreme 4 board has set
vccio and vccsa way to high. was set to 1.2 and 1.25 by default xmp.

For me this was the reason why the system could not boot into windows and was unstable.

I have now set xmp profile for my ram (3200) and then manually lowered

vccio to 1.1V and vccsa to 1.15. Not the system is rock stable again ram running finally with xmp speeds corretly.

Maybe this is helping one or another of you guys.

Best regards

flode

Edited by Flode - 10 Dec 2017 at 7:37pm

Hi,

I just registered to report the same problem.
I built my new system yesterday.
i5-8600K
16GB G.Skill Flare X 14 3200 ( F4-3200C14D-16GFX )
Z370 Extreme 4
GTX 1080
Win 10

The board came with 1.10 and I flashed 1.30 immediately. Didn't know that it could be unstable, downloaded newest BIOS from official site.

After installing Win 10 and everything, I went back to the UEFI and set XMP Profile from "Auto" to "Profile 1".
After like 3-5 minutes in Windows, the system froze.
Setting it back to "Auto", which makes my expensive RAM run with 2400Mhz, and everything is stable since then.

Since my system is brand new, I didn't toy with any voltage settings in the UEFI at all.

There seems to be something wrong with 1.30 and XMP Profiles.
I didn't buy "3200-14-14-14-34" RAM to let it run at "2400-16-16-16-39".

Is is possible to roll back to an older BIOS version without any issues?
How fast will this be fixed? I have like 10 days left on my return policy.

Oh, and PS: I really like ASRock Boards. I had a P67 Pro3 and after that, for many years, a Z77 Extreme4 with which I was very satisfied. Both served my good old Sandy Bridge well for many years.
I wanted to be loyal, read some impressions of the Extreme4 and went with it.

Hopefully this can be fixed.

best regards
Doc



Edited by DocBrown - 11 Dec 2017 at 8:02am

These posts seem to confirm the issue. I say "seem" only because it is unusual for an issue to apply 100% of the time when there are hardware variables involved. The variables here are different models of memory, with differing OC speeds applied by the XMP profile.

Enabling the XMP profile causes an automatic increase in the VCCIO and VCCSA voltages. That is not caused by the XMP profile itself, but the code in the UEFI. That is a standard adjustment when over clocking memory, it normally enhances stability.

The interesting thing here is, all three users posting about this issue are using 3200 speed memory. My guess is the voltage values used for VCCIO and VCCSA when the XMP profile is enabled is simply one voltage value, regardless of the memory OC speed.

A memory OC of 3200 is not a very high memory over clock relative to Intel Z370 and Z270 boards and the Kaby Lake and Coffee Lake processor's memory controllers. Memory over clocks for Z370 and some Z270 systems of over 4000 are possible. The high VCCIO and VCCSA voltages are more appropriate for use with memory over clocks over ~3600.

The purpose of the high VCCIO and VCCSA voltages applied when the XMP profile is enabled is to insure (as much as possible) a memory OC will POST successfully. As we see with CPU over clocks, more VCore is needed, but too much is unnecessary and we reduce it to an appropriate value.

The problem is all users are not aware of the need to configure the VCCIO and VCCSA voltages with memory over clocks. The memory OC potential of the memory controller is also a factor. The compromised solution is to set them on the high side.

VCCIO and VCCSA voltages also  tend to be "sweet spot" type voltages, the optimum values need to be dialed in for the memory being used, and the OC speed. Increasing them worked with memory over clocks, but too much was not known to cause stability issues, at least with Z270 systems.

My ASRock Z270 board's UEFI also set the VCCIO and VCCSA voltages higher than was needed with my G.SKILL 3600 memory. But I did not experience any of the freezing issues described here. talos reported the same experience with his Z270 system. I am surprised that reducing those voltages caused the freezing to stop, given past experience. I'm glad that it did. It seems things have changed with the memory controllers in Coffee Lake processors.

FYI, you can check the actual applied VCCIO and VCCSA voltages in the H/W Monitoring screen in the UEFI, and in Windows with programs such as HWiNFO64. My point is the actual/resulting voltage tends to be a bit more than the value entered for either option, according to the readings. You might want to select a slightly lower value for the option, to achieve a perfect result for your memory.