[SOLVED] 5800x keeps crashing my game

[KiruseiNagisa]

Recently been following an undervolting guide by Optimum Tech and applying it to my 5800x. Enabling PBO, disabling PBO limits, turning Curve Optimizer to "All Cores" negative, and setting the core curve magnitude to 30.
I ran a 10 minute Cinebench session with no crashes(scored 15719 on multi-core, hitting around 80 degrees) and thought I was ok. But once I decided to run a game afterwards, specifically Evil Dead the Game, my PC crashed a couple of minutes after starting a match with giving me an error or BSOD. It immediately restarted my computer. Haven't ran a test yet seeing if it works properly without PBO enabled

Any idea why it crashed?

Specs:
R7 5800x
MSI DUKE 1070 Ti
2x16GB 3600MHz
B550 MAG Tomahawk
Noctua NH-D15S
Samsung 970 Evo Plus
Straight Power 11 750W

 

[drivinfast247]

Try running default settings. Are you on latest BIOS?

 

[KiruseiNagisa]

Yeah, just updated my BIOS yesterday as well as the chipset drivers.
Default settings as in PBO disabled?

 

[drivinfast247]

Yeah, just updated my BIOS yesterday as well as the chipset drivers.
Default settings as in PBO disabled?

Yes, set ALL to defaults. Try running game. If successful, try enabling XMP/D.O.C.P

RAM installed in slots a2-b2?

Latest chipset driver from AMD?

 

[KiruseiNagisa]

Reset all BIOS settings and ran a game without crashing. Afterwards, I enabled XMP and ran a game without crashing.

Both ram sticks are installed correctly according to the mobo manual, and I installed the latest chipset drivers from AMD's site.

 

[Karadjgne]

Are you Following the Guide? Or are you Copying the Guide. Just because it's a 5800x does not imply it's exactly the same and will respond exactly the same as OT's. You may need to drop the curve to 20/25 as your particular cpu may not entirely like 30.

YouTube, or any review whether amateur or professional, can only show that something is Possible, they can't even claim their results are Probable. Use the guide as a tool to figure out exactly what is good for your particular cpu, not as an absolute that will work just because it worked on his.

Anything and Everything that has Anything to do with Any kind of OC is Theory and should be treated as such, until proven or disproven on Your pc.

There's only One Guarantee when OC is involved, that there are simply No guarantees at all.

 

[drivinfast247]

Reset all BIOS settings and ran a game without crashing. Afterwards, I enabled XMP and ran a game without crashing.

Both ram sticks are installed correctly according to the mobo manual, and I installed the latest chipset drivers from AMD's site.

So you're good then?

 

[KiruseiNagisa]

Yup, thanks!

 

[zx128k]

Before you overclock/undervolt, ask the qestion if the system works at stock. If the system does, then think about the changes you will make and make them one at a time. Cinebench is not a stablity test and shouldn't be used as one. I can get a massive overclock and pass Cinebench but BSOD like crazy under a different load. Handbrake for example or prime95.

With undervolting and boosting higher in games is that the lighter loads require their own higher vcore. Basically each point on the v/f curve has to be stable. Thus you can be too low, thus crash under one load and fine in another. With AMD good cooling will allow a decent overclock in games without any big changes. Temperature is taken into account when boosting.

Most of the AMD overclocking is with RAM tuning, unless you have the 5800x3d. This gives the best boost in games and takes the most skill to pull off. Reduce the amount you are undervolting until real stablity tests complete correctly. A test is something like prime 95 aida64, occt etc. Everyone has their own test they respect.

Memory overclocking is best covered here. This is the github guide and covers both Intel and AMD.

 

[Karadjgne]

Cinebench is ok for stability testing, but not nearly as good as Aida64 or Asus RealBench as those programs use a variety of instruction sets at multiple levels of intensity with multiple levels of ram and gpu input.

Handbrake and Prime95 less so, they concentrate on a few instruction sets, but don't punish a large enough variety. Back when cpus were simpler, less cores, no scheduling, no infinity fabric, Prime95 26.6 was very decent for stability, because games were also simpler and followed along similar instruction sets. After BF4, that all changed and modern games use a much larger variety, in varying levels of intensity, there's even a Skyrim mod using AVX and AVX2.

That's why you can run 8 hours of Prime95, get Zero errors, open up a web browser and bsod.

Stability testing is all about trying to make the cpu fail, bsod, not testing to see if it will.

 

[zx128k]

I have found that vccio being too low will cause my 10900k to be stable in prime95 one moment and then unstable in another (normal after boot again in the morning). Also Prime 95 shouldn't be the only test. I have always been stable after running prime 95 for a full small ffts and 1 hour of large ffts. This is because I am almost sure that the system is stable before I run prime 95. If there is an error its normally after 6 hours, its one error and can be fixed by a bump in vccio. I used TestMem5 and a custom config called Extreme1@ anta777.cfg. Everyone has their own way. I am not saying TestMem5 will catch all your problems as well.

Aida64 and occt would pass for me but prime 95 small ffts would fail. Even prime 95 large fft would pass. Games would appear stable. Getting to know the funny way my system reacts to overclocking has allowed me to get the overclock I wanted. The issues I have had were strange. Once the system was unstable I would have to turn it off or even stable setting would BSOD in games. Getting VCCIO right in the end got rid of all these strange issues.

Normally if at the end a full pass of small ffts and one hour of large ffts was enough. The 6 hours of large ffts was enough to cover all the test with the settings I used. This is after 2 hours of RAM tests. The RAM tests won't find all the problems with the RAM but will get rid of almost everything. The rest one hour of large ffts will catch. No issue currently but yeah every overclock has issues with one stablity test or another.

Cinebench is useless for stablity testing because it has no error checking. Basically you can run it for hours and get loads of issues with data corruption afterwards. That was the first program I would try. Everyone has a system they swear by and a common goal of not hammering the cpu with prime95. linpack is another test people like to use. Also intel extreme tuning utility.

I used aida64, intel extreme tuning utility, occt, TestMem5 , memtest86 and prime95. I found TestMem5, and prime 95 to be the best way for me. I did prime 95 small ffts for both SSE and FM3. I did large ffts for FM3 for 6 hours with custom settings. After all that no issues. Cinebench and aida64 are the weakest. aida64 fpu on its own is better. Only prime 95 will pull 300+ watts and crash my system with small ffts. Rant over.