PC Graphics

[giroro]

I'm having GPU driver crashes and system stability problems, wondering if anybody knows what I should try next. 

The primary symptom which led me to troubleshooting is that the GPU will crash to a black screen after editing video in Adobe Premiere pro for around 20 minutes. In the instances where the system recovers or soon after reset, it will start crashing much faster (~5 minutes) when I relaunch the software.  This leads me to think the problem may be related to something in the system overheating, and not software. When I can get the system to resume, there is usually an error from AMD utility indicating a driver timeout.  

When Premiere Pro Crashes, the GPU die and memory temperatures are both under 65C. CPU temps are always under 75C when there is a crash, usually closer to 55C. I'm not finding anything overheating in my hardware monitor, but PSU and memory temperatures are not reported there. 

The system is clean, not dusty.

I became aware of the problem after a recent attempt to upgrade RAM, but it has also been ~4 months since I last ran Premiere Pro. I also had to recently recover the C:/ drive to a ~6 month old backup in that timeframe. 

An additional problem that I believe to be related, is that the system has problems running MemTest64. It will occasionally get a BSOD, the BSOD itself usually has graphical corruption. It also sometimes causes the AMD driver to report a different error than when premiere crashes. Something about detecting a black screen. Running MemTest64 many times, it always has "0 zero errors detected" when these crashes happen.  

System Specs:

• Seasonic PRIME GX-750 (from 2019)
• MSI MPG X570 Gaming Edge Wifi Motherboard
• Ryzen9 5900X with Noctua nhu-14s cooler
• AMD RX 6900X (AMD reference card purchased from AMD)
• Corsair Trident Z Neo 2x16GB memory and Team T-Force Dark Za 2x16GB - Both kits DDR4 3600 18-22-22-42. (Tested in many different configurations and timings)
• Case fans: 2x140mm intake, 2x120mm out
• Case: Sharkoon S-25W
• Creative Soundblaster Z
• DVD burner/multi card reader
• Audient ID14 (USB Audio interface)
• 2 monitors (1080p60 and 1440p144)
• Windows 10

Things I've tried:

• Updated Motherboard/chipset drivers
• Updated Windows
• Updated GPU drivers
• Rolled back GPU drivers
• Updated any other drivers I can think of
• Updated BIOS
• Rolled back BIOS
• Various Troubleshooting of Premiere Pro, including reinstall
• Tried Pro drivers for GPU & switched back to Adrenalin (clean installs)
• Tried many configurations of memory (individual sticks in different slots, pairs, all 4 sticks) 
• A lot of RAM troubleshooting and stability testing - Including configurations that used to be stable, but are no longer stable. Also tried looser timings and higher voltages. 
• Reseated GPU in its slot

'This system used to be stable with memory in XMP. I'm not sure of the best way to test if the PSU is overheating or underpowered, but I think that is the next logical thing to try. I reached out to MSI with the previous information, and they blamed the GPU and told me to put either put the GPU in a new system, or to try swapping out for Nvidia to see if that works. That's not really practical; I can't really put out for an entirely new system or a system-priced GPU just to start troubleshooting one component. 

I'm skeptical that it's the GPU itself as the only time I've found these problems is during heavy memory usage like memtest64/premiere. I've been looking at Memory -> MoBo ->PSU.  The sketchiest quality part in my computer is the MoBo, and the PSU seems to be good quality. I thought I was close to narrowing the problem down a single stick of Gskill ram running when I first got a BSOD in memtest64, but I later got the same corrupted BSOD with only both Team T-Force sticks installed. XMP was on both times, but premiere still crashes at significantly lower memory clocks. 

Does anybody have any idea what to try next, or a free way to test the GPU? I'm running out of ways to try different drivers. Or does AMD at least have, like, official customer support for this kind of thing?

[ThreeDee]

With your setup .. even though Seasonic makes great PSU's .. 750wtt, I'd say is on the light side .. 850wtt would be minimum, but I'd even go higher so you have more than enough to handle transient power spikes when under 100% load with CPU, GPU and I/O .. and keep with at least 80+ Gold rating

Make sure to run separate power cables from your PSU to each power input on your GPU

Check your system's Event Viewer for logged errors that might help troubleshoot  your issue

My wife's old configuration:

3700x,2x16GB 3200 ECC UDIMM's,RX 550 2GB, B550m Phantom Gaming 5, 500wtt RGB Thermaltake 80 Plus (White), Win11

..She was having issues with some driver timeouts and other oddities that "clearly" pointed to a faulty GPU

I replaced her PSU with a Segotep 600wtt 80+ Gold and all her issues went away. 

That "might" be what is going on with your rig

[DimkaTsv]

What error does BSOD throw out in bottom of screen? 
I would expect something like PAGE_FAULT_IN_NONPAGED_AREA, or MEMORY_MANAGEMENT
There are few other other BSOD's that can trigger, but most of them are possible only during boot sequence.

[giroro]

The lower third of the screen wasn't readable. It was a mess of mostly green and black glitches. Somewhat similar looking to what happens when a GPU overheats and or dies entirely.

[goodplay]

Running mixed ram kits is a possible cause. 

[elstaci]

I agree with Goodplay concerning running different manufacturer's RAM kits together.

Even though they may have the same or similar specs the actual Memory chips might be completely different.

Are both Ram Kits on your Motherboard's QVL list for the 5000 series processors?

I would just run with one type of RAM kit, either G-Skill or T-Force RAM kit and see if your PC runs more stable to rule incompatible RAM sticks.

You can always run the PC with just one RAM stick and see if it is stable.

Did you run MEMTEST64 with each stick separately or with the entire kit installed?

If you are getting BSOD's while running MEMTEST64 does indicate a hardware issue. Either Incompatibility or defetive RAM or Motherboard DIMM Slots.

NOTE: Recently another User did the same thing by purchasing two different manufacturer's kits but with the same specs. He later found out that the Memory chips from each kit was made from different manufacturers. One Kit Memory chips was Micron and the other Ram kit Memory chips was Hynix.

EDIT: Ryzen processors are fairly sensitive to the type of RAM that is installed. As for compatibility issues, I once had a 4 RAM Stick kit of 4GB each (16 GB kit). One of the RAM sticks went bad using MEMTEST64.

Opened a Corsair Warranty ticket. They wanted me to send in the entire 4 stick kit to be replaced which would have left me without a computer for an extended time. I asked why they just couldn't send me one Stick of the same type. Corsair said due to incompatibility issues which they couldn't guarantee.

[giroro]

For the problem of mismatched RAM, I spent around 8 hours trying most socket and RAM combinations. Ended up cutting myself pretty bad on one of those corsair heatsinks.

I had tried running each individual stick in each individual slot. Using pairs from either manufacturer in different slot combinations also didn't work

[elstaci]

Thanks for the update.

I solved my Corsair RAM problem by just purchasing the same RAM stick as in the kit except as a single RAM stick. Worked without any issues.

[giroro · Accepted solution]

I found a solution. My 2 monitors were running at at different framerates. I set both to 60 Hz and my system has been stable all week, passing memtest dozens of times and around 12 hours of video editing in premiere pro. Playback performance in Premiere also improved. Such a dumb and frustrating problem.

Why can't this system handle the mismatched framerates when there's a lot of ram and/or CPU usage?  No idea, but I'm leaning toward an actual driver issue. It can't be known without a way to contact AMD or see if it's a known bug.  

FWIW, seagate support didn't think it was the PSU, MSI told me to try swapping in an nvidia GPU (because people just have those lying around? ). I found the problem before getting in deep with either. I wish I had a 3rd party GPU so there was a company I could have contacted about it before wasting all this pain.

[elstaci]

I find it really strange that running 2 Monitors at different Resolutions would cause all those problems with RAM and GPU and  BSOD errors.

Your GPU shouldn't have any issues supporting two different Resolutions at the same time in my opinion.

By the way, Good troubleshooting for finding a fix for such a rare problem.

[giroro]

I agree, it was hard to find the problem because I never would have expected it. 

Maybe a resolution mismatch causes the system to hit the memory a lot more?  Maybe the PCI-e bus starves when the memory bus is overloaded?  Don't know.

The mismatched system seemed fine, outside of memtest and premiere.

I wish I had 2x 144Hz monitors to try seeing if the problem is the mismatch or just from running above 60Hz, in general. 

The problem still happens with the 1080p monitor at 60Hz and the 1440p monitor at 120 Hz.