GPU strange behaviour and crashes.

[mvpetev]

Hello everyone, making my first post here. I've been part of this community for some time, mostly as an observer. Originally, I planned to discuss this on a different forum, but I stumbled upon a post here that sounds quite like my GPU issue.

 

 

I'm referring to the thread GPU buzzing noise - Troubleshooting - Linus Tech Tips

 

 

Recently, I upgraded to a new system after using a 4th gen Intel for almost a decade. My current setup comprises:

CPU: AMD Ryzen 5 7600
Motherboard: ASRock X670E PG Lightning AMD X670E
RAM: 32GB G.Skill Trident DDR5-6400 DIMM CL32-39-39-102 Dual Kit
PSU: 850 watts be quiet! Pure Power 12M Modular 80+ Gold
CPU Cooler: Arctic Liquid Freezer II 280 A-RGB All-in-One
GPU: 16GB ASRock Radeon RX 6800 Challenger Pro DDR6 256bit Retail

While everything else works flawlessly, I'm facing issues with the GPU.

 

Randomly, the screen goes black, but the PC remains functional. Any ongoing tasks continue, but the display vanishes. The driver cites a crash, mainly because the GPU seems to disappear as a device until a power cycle. For emergencies, I've connected a backup monitor to the onboard GPU.

 

Problem Symptoms:

After booting, the screen turns black after an unpredictable duration without the PC crashing.
This occurs regardless of whether the GPU is under load or idle.
It happens with fans both active or at rest.
I've experienced it under various clock settings: underclocked, stock, and overclocked.
The issue persists across different PCI slots.
The GPU's thermals have been excellent, even under continuous loads (max 80 deg on junction). Initially suspecting overheating, I invested in extensive cooling solutions.

 

Steps I've Undertaken for Troubleshooting:

Rolled back to a previous motherboard BIOS version.
Examined both default 4800 MHz RAM and the 6400 XMP settings.
Tested stock, underclocked, and overclocked GPU settings across both RAM configurations.
Manually switched PCI to GEN4 and retested.
Installed a fresh version of Windows with both the current and older chipset & GPU drivers.
Updated the BIOS for both the GPU and motherboard and retried the above tests.
Re-seated the GPU and checked its power connections multiple times.

 

Despite these measures, the problem remains unresolved. Intriguingly, this can't be merely a driver issue since sometimes the GPU vanishes from the boot sequence as well.

When a crash happens, it seems the GPU retains power, evident from the fans still spinning. However, on restarting, if the GPU doesn't reappear immediately (often requiring multiple restarts), its fans exhibit a peculiar twitching behavior. Giving them a slight push causes them to spin at the speed I've set with the push.

 

Additionally, the GPU produces a fluctuating sound. While some might attribute this to coil whine, I'm skeptical (please refer to the attached audio and note the subtle background buzz). My reasons for doubting it's coil whine include: firstly, this sound only manifests when the GPU isn't under load; even with fans at maximum speed and no active tasks, it's present. Yet, the moment there's a load, the noise ceases. Secondly, the sound maintains a consistent pitch, unlike typical coil whine variations. To my ears, it resembles capacitor arcing more than anything.

 

I'm contemplating an RMA, but it's worth noting this would be my second time returning this specific GPU model. Either I'm tremendously unlucky, or the RX6800 series just dont work.

 

The previous time when i returned it the old GPU actually had fan issues and it was overheating but never crashing like this one. In addition the problem doesnt occur with my old GPU however it is GTX1080 so im not sure if its comparable. At least i can exclude PCI slot problems i guess

You can see the whole process in this case starting minecraft. But for the type of crash look around 2:50min

 

Anyhow I would be curious to hear some suggestions wether i missed something or if someone has similar experience. 

 

230919_0050.flac

[191x7]

Are you using any custom cables for the PSU, riser cables, extension cables, or cables from a different (or older) PSU?

 

This is something the PSU tier list says about Pure Power 12M units:

 

Have you considered balancing the load on the 12V rails? 

 

Since your MBO (CPU) is on V1, have you connected the GPU to V2?

 

Maybe 1 cable on the GPU to V2 and the other V1?

[mvpetev]

Hey, thanks for your reply,

I had considered the possibility of the PSU being the culprit. However, a couple of things make me think otherwise. Firstly, the power levels we're discussing don't seem high enough to cause issues. Secondly, I'd expect to hear the internal relay of the PSU triggering or experience more significant system failures if that were the case. To put it in perspective, the GPU at its peak consumes around 230W, and the CPU uses about 80W. Even considering sudden power spikes, it's unlikely they'd max out the PSU. Moreover, these crashes occur just as frequently during idle times as they do under load.

 

I'm not using any supplementary cables. The GPU is directly inserted into the motherboard's PCI 16x Gen5 slot, and I've stuck with the original PSU cables.

 

Here's a rundown of my current connections:

• The primary motherboard connector is attached directly to the motherboard.
• The GPU is connected via two distinct cables linked to the 12v1 and 12v2 rails.
• The additional 4-pin and 8-pin CPU power connectors at the top of the motherboard are each connected to the P8 PSU 12v2 connectors using separate cables.

[191x7]

4 minutes ago, mvpetev said:

Hey, thanks for your reply,

I had considered the possibility of the PSU being the culprit. However, a couple of things make me think otherwise. Firstly, the power levels we're discussing don't seem high enough to cause issues. Secondly, I'd expect to hear the internal relay of the PSU triggering or experience more significant system failures if that were the case. To put it in perspective, the GPU at its peak consumes around 230W, and the CPU uses about 80W. Even considering sudden power spikes, it's unlikely they'd max out the PSU. Moreover, these crashes occur just as frequently during idle times as they do under load.

 

I'm not using any supplementary cables. The GPU is directly inserted into the motherboard's PCI 16x Gen5 slot, and I've stuck with the original PSU cables.

 

Here's a rundown of my current connections:

• The primary motherboard connector is attached directly to the motherboard.
• The GPU is connected via two distinct cables linked to the 12v1 and 12v2 rails.
• The additional 4-pin and 8-pin CPU power connectors at the top of the motherboard are each connected to the P8 PSU 12v2 connectors using separate cables.

Try not having the GPU on two separate rails. Some GPU-s have issues with that.

[mvpetev]

Right good idea and not something I have tested. Will check and write back. 

 

The only thing I'm not happy about with this is the fact I'll have to use the pigtail type connection but the cable should be able to handle it I hope 

[191x7]

3 minutes ago, mvpetev said:

Right good idea and not something I have tested. Will check and write back. 

 

The only thing I'm not happy about with this is the fact I'll have to use the pigtail type connection but the cable should be able to handle it I hope 

It's more than capable, don't worry about the cable. If it has 2x 8-pin it can do 300W over 12V.

[mvpetev]

5 hours ago, 191x7 said:

Try not having the GPU on two separate rails. Some GPU-s have issues with that.

Nope, just tested its the same behaviour on both the 12v1 and 12v2 rails 

[191x7]

1 hour ago, mvpetev said:

Nope, just tested its the same behaviour on both the 12v1 and 12v2 rails 

Can you test the GPU in a different system?

Or try with a different psu?

[mvpetev]

Well I guess I'll have to find a way if I want to be sure. I can test it on the old pc which is i7-4790 and will probably be a massive bottleneck but then if it still happens it should be the GPU

[191x7]

On 9/20/2023 at 9:01 PM, mvpetev said:

Well I guess I'll have to find a way if I want to be sure. I can test it on the old pc which is i7-4790 and will probably be a massive bottleneck but then if it still happens it should be the GPU

Any news?

[mvpetev]

11 hours ago, 191x7 said:

Any news?

I finally got around to testing; the past week was quite busy for me.

Here's what I observed:

When testing on a different computer, the direct GPU test didn't appear to throttle or bottleneck the performance. It reached 100% in all tasks. However, I did face some peculiar issues. For instance, the graphics card wouldn't initiate until after I restarted the computer. It didn't crash, though.
 

This is the sequence of events:

• Cold boot: The GPU isn't detected and doesn't even initiate the fans, which typically turn on during a boot.
• Upon restart: Everything is in order. The GPU functions properly, and both stress testing and gaming don't cause it to shut down at least for the short testing time.

Now, here's where it gets odd. Once I reinstalled it into my primary computer, the performance seems to be somewhat stable.

There was one crash during a highly intensive task, but there haven't been any random shutdowns.
 

Is it possible that GPUs have a 'burn-in' period similar to audio amplifiers that I might not be aware of?

P.S. As I mentioned earlier, the older PC uses an i7 4750. The power supply unit is a single rail Be Quiet! 600W, which should adequately support the demand. The RAM is DDR3; I haven't checked its exact specifications, but I deem it irrelevant for this context.

 

P.P.S. It started randomly crashing again on both machines. I am gonna call it its some capacitor somewhere going wild and taking some time to charge. I guess I`ll have to RMA it again.

[191x7]

1 hour ago, mvpetev said:

I finally got around to testing; the past week was quite busy for me.

Here's what I observed:

When testing on a different computer, the direct GPU test didn't appear to throttle or bottleneck the performance. It reached 100% in all tasks. However, I did face some peculiar issues. For instance, the graphics card wouldn't initiate until after I restarted the computer. It didn't crash, though.
 

This is the sequence of events:

• Cold boot: The GPU isn't detected and doesn't even initiate the fans, which typically turn on during a boot.
• Upon restart: Everything is in order. The GPU functions properly, and both stress testing and gaming don't cause it to shut down at least for the short testing time.

Now, here's where it gets odd. Once I reinstalled it into my primary computer, the performance seems to be somewhat stable.

There was one crash during a highly intensive task, but there haven't been any random shutdowns.
 

Is it possible that GPUs have a 'burn-in' period similar to audio amplifiers that I might not be aware of?

P.S. As I mentioned earlier, the older PC uses an i7 4750. The power supply unit is a single rail Be Quiet! 600W, which should adequately support the demand. The RAM is DDR3; I haven't checked its exact specifications, but I deem it irrelevant for this context.

 

P.P.S. It started randomly crashing again on both machines. I am gonna call it its some capacitor somewhere going wild and taking some time to charge. I guess I`ll have to RMA it again.

GPU-s don't have a burn-in. Neither do amplifiers, that is a myth.

 

RMA the card, it seems to be faulty.

[mvpetev]

Just as a quick update - the new new card came back and after a week of continuous running with and without load there has been zero crashes so i guess I was just unlucky and had 2 consecutive defective GPU`s.