AM5 System Unstable: 7950X @ ASRock X870E Taichi

Hardware Hub Build a PC
, ,
1

Hi,

I hope someone here can offer advice on what to do next. I’ve honestly had it with this whole mess and am beginning to doubt the universe itself.

Context:

I had an almost 11-year-old desktop that was my main system, and it worked fine. Then, I decided it was time for a new one in 2023.

Attempt 1:

After some research, I chose the AM5 platform. However, after countless mainboards, power supplies, and RAM kits (each component swapped at least 5 times; this isn’t an exaggeration—I counted 7 mainboards!), I couldn’t get the system stable and gave up in frustration. In hindsight, it was probably the CPU (7800X3D) that was defective.
Having been burned by AM5, I couldn’t trust it anymore. I was able to return most components and sold the rest.

Attempt 2:

Attempt 2 was with Intel 1700, specifically a 14900K. You can probably guess what happened. For a few months, I was happy. Then, the degradation issues began. After two CPU RMAs, which only improved things for 2-3 months each before degrading again, I knew this couldn’t continue. No matter how many microcode updates Intel would end up releasing, I would never be able to trust this system again.

But what to do now? I can’t keep using my old desktop in 2024.

Attempt 3:

So, back to AM5 again it is.
And, of course, the system is unstable. Sigh.

Current components:

  • PSU: Seasonic Vertex GX 1000
  • GPU: PNY RTX 4080 XLR8
  • CPU: AMD 7950X*
  • MB: ASRock X870E Taichi*
  • RAM Kit 1: (ECC) 2x Kingston KSM56E46BD8KM-48HM*
  • RAM Kit 2: (non-ECC) 2x G.SKILL F5-6400J3239G32GX2-TZ5RK

Note: The RAM kits are not installed at the same time! I’ve explained below how I’ve installed and tested them. The system is unstable even with only a single stick of RAM installed!

The PSU, GPU, and G.SKILL RAM have all been in use for over a year with the old Intel system and worked fine, apart from the CPU degradation issues that eventually showed up. So, I assume these components are fine.
The components marked with * are the newly purchased ones, and I’m still within the return window.

The Situation / Problem:

I have the latest stable non-beta firmware on the motherboard (version 3.16 from December 2024).
No matter whether I’m running Windows or Linux (Kernel 6.12.16, 6.13.1), and no matter the RAM configuration, I encounter various errors and crashes:

  • Linux: Kernel null pointer dereferences and various other oopses.
  • Linux: Sporadic USB disconnects (both USB 2 and USB 3).
  • Linux: System crashes with frozen mouse during higher USB 3 activity (reproduced with a 2.5" HDD on a USB-C adapter and NVMe USB C adapter; both work fine on other systems).
  • Windows: Sporadic BSODs of all kinds (no specific driver or module), even right after installation during the oobe setup

Previous Attempts:

  • Reset UEFI to Defaults (the G.SKILL defaults to 4800 MHz, Kingston to 5200 MHz).
  • Used only one RAM stick (Kingston or G.SKILL) in slot A2/B2; this reduced BSOD frequency in Windows, but didn’t stop them completely.
  • Used only one RAM stick in A1/B1; this gave me a POST code C5.
  • Ran Passmark MemTest v11.2 Pro for 4 hours with both G.SKILL and both Kingston sticks: no errors.
  • Ran Y-Cruncher with all tests for 2-3 hours: no errors.
  • Ran TM5 with @anta777 ABSOLUT(01102021) for almost 2 hours: no errors

Current Status:

  • UEFI is set to defaults, with the latest stable version (3.16 from December 2024).
  • Windows 11 24H2 is fully up-to-date, including chipset drivers from the ASRock website.
  • One of the G.SKILL sticks (the stable one from the Intel system) is in slot B2, running at 4800 MHz with timings 40/40/40/117 due to UEFI defaults.
  • Yesterday, I got two BSOD, which both were unique ones I hadn’t seen before (it disappeared quickly, but I believe it was “PFN List Corrupt”, then during shutdown another one).

Honestly, I’m starting to lose faith in myself and in the modern PC market. I’m about ready to return the AM5 components and give up on the idea of a modern desktop altogether. How can it be that after three build attempts across two platforms with brand-new hardware, something always goes terribly wrong?

What now?

Does anyone have an idea what might be going on here? What else can I do?

2

I don’t see anything in there investigating the boot drive(s). That would be my next stop.

3

I am observing instability both with OSes installed on either an NM620 or a 990 PRO, and also with Linux Live Systems booted from USB (Manjaro, Fedora 41 (no ethernet due to Kernel <6.11, Fedora rawhide (“42”)).
The two SSDs and the USB drive are all confirmed to be OK on other systems.

4

FWIW, the 7950X + 4080 XLR8 build I did with 4x48GB on X670E Pro RS has just worked with the 2DPC 2R at 4800 for over a year. It’s on 10 22H2 with drivers via AMD as I’ve only used ASRock for BIOS updates.

LGA pressure, board warp, or issues with the PSU might be possibilities. Usually in situations like this I check the cooler screw torque and, if that’s ok, pull the board to see if the standoffs are planar. If nothing else looks likely I’ll pull the CPU and inspect the lands and socket pins.

5

Given that the PSU and GPU were both working fine on the Intel system, is it safe to assume that they should be good? Could the PSU or GPU potentially be only causing issues with AMD systems but not Intel systems?

GPU is easy to test for me, I could just pull the card and check if running on the iGPU makes it stable.
However, I’d have to order and potentially return another PSU just to see if that could be the reason.

It’s an NH-D15S and the screws have a stop position. I did not use any force once I hit that point. Best I could do is loosen them by, say, a quarter of a rotation.

Would testing the board on the cardboard box constitute alternative test conditions that could rule this out?

They looked perfect if I remember correctly. After all, this is supposed to be a new board. Assuming none of the other tests get me any step further, I might unscrew the cooler and check the socket.

6

One small thing that goes outside of the common sense - is there a way to try and test with a UPS (or a portable power station)? Or maybe bring PC to work\friend scenario? I’ve heard a story once about a “We built a pc for customer. Checked it. After few days the customer called and said it crashes constantly. Brought it back. A few days of all possible scenarios - no problems. Brought it to the customer. 5 minutes and BSOD” (old fridge was on the same power line as the PC).

7

Yes, I was about to suggest the same.

Instead of a UPS, try measuring the voltage with a watt meter. It’s a cheap (~$10) device that plugs into the wall, and you plug your PSU into it. This will show real-time voltage fluctuations.

In Europe, the standard is 230V ±10%, meaning it can range from 207V to 253V. You might see 240V or 220V, depending on how close you are to a transformer. Voltage isn’t constant, so it fluctuates slightly.

Different hardware reacts differently. Newer components may have tighter tolerances and struggle with lower voltages. If your system is unstable periodically, it could be due to voltage dips.

This is a simple way to rule out power issues before investing in a UPS.

Mind that hardware is usually built to handle that fluctuation, so the fluctuation isn’t really the issue, and shouldn’t be. But for the sake of debugging, I would spent $10 before getting an UPS. As what the UPS would do is stabilize the power delivery, plus keep the system alive if it fails… but stabilization only helps if that’s an issue to begin with.

8

Although this is specific to where I live (post soviet), I had a conversation not long ago with a person, from whom I have been buying PC stuff for a while now. He said that although the PSU should handle this sort of scenario (he knows I have a beQuiet DarkPower 13, which is Platinum or something), it is better to have this kind of power hungry PCs powered through a UPS (or a power station, like in my specific scenario).

9

Yes indeed.

10

Unfortunately, I do not have an UPS and I don’t really have enough time and mental capacity to deep-dive into these and figure out whether I actually want one and which one would suit my needs. In fact, I don’t even have enough capacity to deal with the PC in the first place, it’s just so extremely frustrating being kept in unstable-pc-limbo for about 1.5 years now.

That would be quite the effort, but I will consider this as a last resort in case everything else fails.

Not really, though. These aren’t high-precision measurement devices or oscilloscopes. They mostly just show you a voltage that is hopefully at least True-RMS (but who knows, given how cheap these are).
I do have one, and it shows 236.4-236.5 V right now. That’s not near the bottom of the tolerance spectrum at least, it’s actually above the nominal voltage FWIW.
My True-RMS DMM confirms this reading, showing around 236.5 V.

See the top of my post, the reply to your other comment.

I might agree with you on a principle, but I have never had any sort of power issue with any device I’ve ever powered in my home and I feel uneasy accepting this rather expensive PC should it turn out that it is so sensitive that I am risking data corruption whenever the voltage dares to not be within 0.1 % of whatever the system deems perfect. The UPS idea feels like a hacky workaround. I’ve never seen a PC that requires an UPS to run and I don’t really want one either.

11

No one is going to disagree with you that it shouldn’t be the case, or even that it doesn’t make sense. But, the reality is there is an AWFUL lot of anecdotal evidence that a UPS really can fix problems. Probably by compensating for the real problem that’s much more difficult to identify. Unfortunately there’s really no way to know other than to try. Stability issues aside, they are pretty good things to use just in general.

I use this https://www.newegg.com/CyberPower-CP1500PFCLCD-NEMA-5-15R/p/N82E16842102134. If you’re just going to run that one PC on it and are willing to fork out the price, I give you my TrustMeBro*TM guarantee it’s among the best options to alleviate the headache of shopping for one.

1 Like
12

I hear you. And completely understand the pain. I had a PC for somewhere a one-two years of constantly randomly freezing, with everything working fine the moment I try to force the issue. Turned out to be a corrupted filesystem on the SSD (and nobody will speak to you of it, until you actually manage to discover the issue… or do a NULL format on the darn thing… and this problem is common. Everything can work, but once you pull a complex load, it starts to break under pressure).

On the other hand. I’ve built a 13900k half a year ago (or even more). MSI board. Applied the 253w PL1/2 and 307a, found on reddit. And have never had an issue. I don’t do rendering though, nor do I do UE5 stuff, so maybe I will discover this problem later on. But for now, I am quite happy with it.

On the other other hand. Next week I plan on building a separate PC for work, and think of going 9900x. And this is the second topic, which brings me worry.

I understand the frustration. But the industry seems to go for common user having common problems, which nobody solves, instead of making the product more reliable.

I can suggest what I bought for myself (but I do have power outages, so its worth the money) - Anker Power House 767 ｜ Power Ready for Anything - Anker CA

13

I doubt it’s the power, or you’d have likely had similar issues on Intel.
If you can, return the CPU and get a new copy. Depending on your return/exchange window on the motherboard, plan to do it after trying the new CPU or do them in parallel to be on the safe side if there’s not enough time left.
Lots of people have been having random issues with AMD CPUs that were resolved by replacing them.

14

Decent chance that’s an issue as it’s necessary to adhere to the 0.6 Nm max torque specification called out in the manual. In my experience when problems like this turn up it’s not uncommon bottoming the screws is closer to 0.5-1 turn too far.

Yeah, same.

An anti-static surface’d be wiser but, either way, inspecting the case for things like standoffs that aren’t fully seated isn’t going to hurt and might turn up (part of) the problem.

15

Sorry, but how much “common” is AMD CPUs having random issues? Asking, since I plan to make a first time in many years an 9900x build for work.

16

You can build that system with confidence, it’ll most likely be fine.

I only mentioned this in the sense that CPUs these days tend to have higher failure rates than some of us might assume. A surprising number of people have been posting about troubleshooting issues on various forums, including this one, and ultimately replacing the CPU fixed things. So I thought it would be good for the OP to not waste time and replace while they can.
But of course, the vast majority of people are busy enjoying their new CPUs and not posting on forums about the their lack of issues… :slight_smile:
In terms of numbers, Puget mentions something like a 5% failure rate for the CPUs in the systems they sold in 2024, with more on the Intel side than AMD, though they won’t share the split.

1 Like
17

Yeah. I understand the dilemma of “Most of the people won’t write something when all is working, while those, who have problems, most likely will”.

I miss the days of having aquientances in some PC building\repair company, to whom you could call and say “Hey. I want to buy a system. Would you go for Intel or AMD” (and actually get a “Yeah, no, Intel/AMD has had quite a big return rate, better stick with the other one”).

18

There are so many potential issues listed and also many assumptions. Bleeding edge on anything but Windows will very likely be unstable specially video.

Go back to basics:

Rip out anything that isn’t necessary, including the video card.

That should leave you with a motherboard, one stick of RAM, CPU + Cooler (disconnect any RGB cables if available),one SSD and a PSU.
Install one stick of the Kingston RAM
Update to latest BIOS, reset BIOS to defaults and don’t touch anything except possibly enable ECC for the Kingston memory.
Run memtest86+ for ~8-12 hours,
If passes, install Windows 11 (latest), let it run Windows Update and install drivers etc)
Run 7-zip’s benchmark in a loop for ~6-8 hours
If that works, run someting like Prime95 with AVX512/256 instructions enabled for 6-8h.
Get back to us once all these tests are done with minimal hardware.

19

I would change that to a hdd(if not usb alone). My pain in the bottox (I basically replaced every living part of my pc, except the ssd… and still could reproduce my issue) was because of a corrupted file system of a ssd (and I faced that problem with other ssds later on).

20

If you have th 850 wat you should be around 91 % load with the 1000 wat it would be around 77%, its best to have that between 50 and 80 precent and it might be on the edge. maybe a 1300 wat supply would help