Need Help Tracking System Freeze

[Brandon Shinobu]

Hey All,

 

So I recently built a brand new system from scratch, something I've been saving a few years for. However, it's having a complete system lock-up that I've been having issues tracking down. I'll be as brief, but as complete, as I can.

 

Build

• Asus Zenith II Extreme Alpha (motherboard)
• 3960x AMD Threadripper
• IceGiant Elite Prosiphon CPU (cooler)
• 128Gb G.Skill RAM, 3600Mhz 14-15-15-35
• 1600w Leadex SuperFlower PSU 80+ Ti
• Lian-Li 24-pin power Strimer cable extension (power struggled to reach without an extension)
• CoolerMaster 90-degree 24-pin power adapter with built-in capacitor (bought to reduce strain on the 24-pin mobo connector
• 3 M.2 NVMEs. Samsung 980 Pro 2Tb x2 and 250Gb x1

Quirk

   Not sure if this is important, but going to mention it anyway. Most of the time when the PC locked up, I'd have to reboot the computer twice. First reboot to get it out of the freeze, it would typically hang on Windows load with an error ranging from "Load VGA BIOS" to "Memory Test" to "Detect HDD". After rebooting a second time, typically it would load into Windows with no issue.

 

History

• The first time the system locked up was while attempting to use ClockTuner. Trying to run the "Diagnostic" or even click the "CPU Stability" button instantly locked up the PC, but it ran fine if I didn't use the program.
• The second time was when I was running the Asus AI Suite 3 software. It actually did just fine through the CPU overclocking each time, but locked up during the "fan" phase. It always did this at 51%.
  • I eventually realized my PC case wasn't set to let the mobo control the fans. After changing that, no more lockup, and it ran through AI Suite 3's system tuning every time with no lockup.
• There were two instances the PC seemed to lock up while going into or after going into a sleep mode, but honestly I wasn't around to see what happened. Just found it locked up and screens dark.

While this sounds like a lot, they were spread out, and eventually 'stopped' when I didn't try messing with anything. Ran Cinebench multiple times without incident, even while overclocked through the Asus software. However, to be on the safe side, I uninstalled the Asus AI Suite 3, and later disabled auto-tuning from the board core boost and the precision boost, and removed OC settings.

• A while later after getting software and everything set up, I tried to run a game. Within five minutes, the system locked up.
• Tried multiple games/graphic programs, including the Heaven benchmark, locked up each time within a max of 10 minutes, typically less than 5.
• Once locked up from opening, closing, and reopening Chrome too quickly (might indicate more of something was in use from two instances of chrome being 'active'?)

Working with a friend, he suggested the GPU was bad so I swapped in my old 1070Ti. Lockups indeed stopped. It ran all the same programs without a hitch. I went and RMA'd the GPU thinking we'd found the problem.

• Next day, I decide to try some Warframe. System locked up in 5 minutes.
• This time I had to reboot the system 3 times, and it showed new errors, including "Check NVRAM" and "POST Error" which hadn't occurred before.
• I cleared the CMOS and re-flashed the BIOS to default, basically only applying the DOCP for the RAM and leaving everything else be.
• Went back, played Warframe for several hours with no problem, and another game for a few hours with no problem.

A friend suggested just to be safe I should run Prime95 as a stress test while I slept that night to see if it locked up and flush out any lurking problems. So I did, and ran it before bed. I woke up to find the PC locked up, but no clear indicator when. After rebooting the system, I ran it a second time, and it locked up my system almost immediately.

 

Final Thoughts

• Before I go run out and RMA anything else, I really need to try and determine what actually needs to be RMA'd if anything.
• RAM has been tested and didn't show any errors in memtest.
• Friend suggested that something was needing more power and not getting it. In that case, the 90-degree adapter might be a culprit, or the Strimer cable? Hard to test the strimer cable as having that extension is the only way the mobo is getting power at this point, but I'll think of something if you all believe it might be at fault.
• CPU has been running hot, even for a 3960x, but it is nowhere near max when lockups occur.

 

Hopefully this all... "helps." After dumping everything I'd saved into this PC just to have this happen I've been pretty torn up as you might imagine. I can't exactly run out and grab replacements. I'm going to try removing the 24-pin power adapter since it's the one thing that's "easy" to test, but any thoughts you all have in the meantime would be helpful.

[AbydosOne]

8 minutes ago, Brandon Shinobu said:

128Gb G.Skill RAM, 3600Mhz 14-15-15-35

Are you running the RAM at DOCP or stock? If DOCP, repeat tests at stock.

[SupaKomputa]

10 minutes ago, Brandon Shinobu said:

128Gb G.Skill RAM, 3600Mhz 14-15-15-35

This timing is very tight, try reducing the number of modules and retest.

[Brandon Shinobu]

9 minutes ago, AbydosOne said:

Are you running the RAM at DOCP or stock? If DOCP, repeat tests at stock.

 

6 minutes ago, SupaKomputa said:

This timing is very tight, try reducing the number of modules and retest.

 

Just to clarify, the RAM itself was rated for those timings and speed at stock. I only activated the DOCP so the RAM would operate at the 'stock' speed instead of the mobo's default 2133Mhz. However, I have cleared the DOCP settings and I'll run Prime95 again here in just a moment.

[Brandon Shinobu]

16 minutes ago, SupaKomputa said:

This timing is very tight, try reducing the number of modules and retest.

 

19 minutes ago, AbydosOne said:

Are you running the RAM at DOCP or stock? If DOCP, repeat tests at stock.

 

Okay, Prime95 is running without the DOCP, and thusfar the system hasn't locked up yet. I'll have to let it run for a while before I feel I'm "safe", and I'll update this thread later with the results.

 

That being said, I had specifically purchased the higher rated RAM for those speeds. Does this mean that the RAM itself is bad, or the motherboard? If not, is there any way to get the speed that was listed on the packaging? Ultimately, anything/everything will be an upgrade from the 666.5Mhz RAM on my old machine here, but i'll be disappointed if I spent extra money for something I won't be able to utilize.

[SupaKomputa]

2 minutes ago, Brandon Shinobu said:

RAM itself was rated for those timings and speed at stock

Doesn't matter, if the mobo doesn't like it, you need manual tuning, especially for running more than 4 modules on high speed at the same time (or full slot).

One way is to increase the voltage, the other is reducing the speed.

Zen 2 officially support up to 3200mhz.

[Brandon Shinobu]

1 minute ago, SupaKomputa said:

Doesn't matter, if the mobo doesn't like it, you need manual tuning, especially for running more than 4 modules on high speed at the same time (or full slot).

One way is to increase the voltage, the other is reducing the speed.

Zen 2 officially support up to 3200mhz.

Understood, and I think I may see where I went wrong then. When I was originally planning my build, I was going to be using a 5950x, but I later decided I needed the extra PCIe lanes for my work and switched over to a threadripper for rendering and load times. I remembered several places mentioning that 3600 was the "sweet spot", but that was probably for the 5950x, not the threadripper, and I never thought to check that.

 

And now I just need for Asus to get back to me and tell me "we didn't find anything wrong with your GPU, here, have it back." I suppose there are harder lessons than $60 RMA shipping costs. Like overpaying for RAM, heh.

 

I really appreciate the help. And thusfar the system is still running without locking up, whereas it was locking up within 5 seconds of Prime95 before. I'll let it run a couple hours to be sure.

 

I honestly don't know how to manual clock my RAM, and for lack of knowledge I'm a bit afraid to, but I do want to learn and also get at least some of the speed I paid for. Do you have any recommendations for speeds/timings that would be safe to do for a 3960x? Or perhaps a tool that would make it easier to do safely?

[SupaKomputa]

Yeah 3600 is the sweet spot, but most done it with only 2 modules, 4 if your lucky.

Some motherboard can't run reliably with fully populated slots without having to tweak the settings (primarily the voltage).
AMD only guarantee up to 3200mhz, even on 5th gen.

 

To manually underclock the ram, it's generally go to the bios, in memory settings choose manual instead of XMP.

Select 3200mhz in the option, save and test.

No worries, changing memory settings won't break anything.

If failed, just reset the bios settings, it will go back to default.

[Brandon Shinobu]

9 minutes ago, SupaKomputa said:

Yeah 3600 is the sweet spot, but most done it with only 2 modules, 4 if your lucky.

Some motherboard can't run reliably with fully populated slots without having to tweak the settings (primarily the voltage).
AMD only guarantee up to 3200mhz, even on 5th gen.

 

To manually underclock the ram, it's generally go to the bios, in memory settings choose manual instead of XMP.

Select 3200mhz in the option, save and test.

No worries, changing memory settings won't break anything.

If failed, just reset the bios settings, it will go back to default.

Thanks! Should I keep the same timings of 14-15-15-35, or would I need to change those also?

[SupaKomputa]

Well you can try, but if you have a problem, just set it in auto.

If that module chip what i think it is (samsung b-die), it can run at that timings.

The motherboard usually will automatically run it at the safest timings (not the fastest though).

For your applications i suggest stability over speed.