Rog Strix JVM issues

[bwasko]

Hi,

I have bought G814JZR over a year ago mostly for development purposes (java). And since then have been struggling with one all the time occuring issue - jvm crashes. Each and every time this occurs, it is not jvm internal failure but OS fatal-error signal which kills the jvm. It occurs only for JVMs (other applications work just fine even under heavy load, games, visual processing etc - no problems). But JVM - sometimes gets killed in runtime, sometimes just during compilations, sometimes instantly, sometimes after an hour. I have run processor and memory stress tests  (the one intel provides but also e.g prime95) - no crashes. But when I start jvm , my app, compilation or external apps like e.g Intellij Idea, sooner or later it crashes. JVM version seems irrelevant. There is only one way to mitigate these failures - disable processor boosting/turbo mode. When I do this, this suddenly stopps happening, When I switch to processor defaults again - failures starting to occur again. Using the registry switch for this.  Any ideas what can be the problem? Anyone had something like this? 

I use Windows 11 and have all the released updates installed

thanks
bart

[Zalevskiy]

del

[ElectroStingz]

Hello,

Does the issue get more frequent if you if you run a -0.05v (or max -0.08v) offset from within the BIOS?

I assume intel XTU software is not an option due to your VM?

[bwasko]

XTU requires disabling Intel Virtualisation or whatever it is called in bios. I tried this just for a test some time ago and in fact iirc the problem was much LESS frequent. I cannot however leave it disabled as I use also docker and it requires virtualisation to be enabled. So for now - I added -20 as offset in bios and will test tomorrow when I work again :). Thanks for hint

[bwasko]

@ElectroStingz , when I set -20 as voltage offset in bios, it was more or less the same, but when I then set -50, apps started getting terminated almost instantly, just a second after opening. 

[ElectroStingz · Accepted solution]

Could be that there is an issue with the CPU voltage, an offset of -0.08 should be doable at default settings so -0.05 would suggest a problem. The CPU can degrade over time so it's possible that now it is just not stable at default (whether this is related to the intel flaw with CPU's spiking to high is yet to be determined).

Does ThrottleStop work for you? I have tried this and it seems to work with all my options in the BIOS still default (VT-D enabled).

I used hardware monitor to view the CPU voltages and it appears to work, ThrottleStop version 9.7.3 (beta) from techpowerup.

Press on the FIVR to bring this screen

Tick unlock adjustable voltages, choose a scale like 250mv and you will then be able to move the offset slider to add voltage.

Here I try 250mv to see it working, my VID is now 1.360v.

The reason I show these is because if the CPU is struggling with the voltages you will need to add more too see what happens but you must be extremely careful on how you go about testing this as you must not try this with the boost ratios on default setting, see I have all my P-cores at x40 in the Turbo Groups.

So the safest way to try this, first find the max boost you can run stable. Try all x52 and test, if crashes lower gradually until you find the stable point. Once at stable state get back to the point on instability, then increase the voltage slowly.

But you must not go over 1.55V.

As a general guide, x52 default will see voltage peak around 1.45V

To check your on CPU spam the benchmark button in CPUz as it will do a simple multicore and single load so its a quick way to see the voltages peak, see in this screenshot 1.46V.

If your CPU is unstable at a very low boost frequency, it's better for you as it gives more room to increase the voltage but see how it goes first before playing with the voltages.

What's the point of all of this? let's assume the CPU is actually stable with more voltage it means it has likely degraded so replacement is the only option. If throttlestop provides you with a decent compromise and the ability to run say x52 all cores without issues and you are ok with that, at least its working with a solution.

But see how it goes and remember do not exceed 1.55V.

[bwasko]

I get the idea but before I even started playing with this voltage section, I just set lower multipliers for the first two groups, just to check what would happen. So initially/by default it was 58,58,52,52....52, now it is 54,54,52, 52.....,52 and it seems that problem vanished. This may be analogical to the XTU setup I tested some time ago, where I also descreased PCores multipliers (but could not keep it due to docker issues). But to be sure I need to observe this tomorrow, and as fallback plan - play with voltages as You described. If this is really gone with this throttlestop change, then it would be good enough for me, at least for now. Thanks for hints 🙂

[bwasko]

@ElectroStingz ok, I can fully confirm that all JVM stability issues disappear when I lower the multipliers from 58 to 56 for the two first groups in ThrottleStop (other groups have the defaults i.e 52). 

[ElectroStingz]

That's something at least, place throttlestop in the "shell:startup" folder. (start menu >Run "shell:startup") This way it will auto run when you get into windows if needed.