Possibly faulty rtx 4090? Games constantly crashing

[mrimo]

Hi All,

I have recently bought a MSI GeForce RTX 4090 GAMING X SLIM 24G and installed it into my PC.

OS: Windows 11 Version 23H2 Build 22631.4037

Motherboard: ASRock Z690 Phantom Gaming 4, Intel Z690, S 1700, DDR5, SATA3, PCIe 5.0,
CPU: INTEL CORE I9-12900K s1700
RAM: 128GB Corsair DDR5 DOMINATOR Platinum RGB Black, PC5-44800
Disk: 3x NVMe, OS / Data / Games
PSU: Corsair HX1500i Fully Modular Ultra-Low Noise ATX Power Supply - ATX 3.0 & PCIe 5.0 Compliant

I have installed multiple versions of the Geforce driver, currently installed: 561.09

Now, every time I play a game, I've tested several from, RDR2, GTA5, Sniper 5, Stranded Deep, Kingdom Come Deliverance

The game crashes after about 1-3 minutes of game play.

--
Example crash logs:
-- Example 01 ------------
Faulting application name: sniper5_dx12.exe, version: 2023.8.10.35004, time stamp: 0x64d544bd
Faulting module name: sniper5_dx12.exe, version: 2023.8.10.35004, time stamp: 0x64d544bd
Exception code: 0xc0000005
Fault offset: 0x0000000000188f64
Faulting process id: 0x0x3170
Faulting application start time: 0x0x1DB0488854CDC24
Faulting application path: C:\Program Files\WindowsApps\Rebellion.52602B995C00E_1.2.621.0_x64__2vbwqmt31j4mr\bin\sniper5_dx12.exe
Faulting module path: C:\Program Files\WindowsApps\Rebellion.52602B995C00E_1.2.621.0_x64__2vbwqmt31j4mr\bin\sniper5_dx12.exe
Report Id: 2d4fc8ba-48ba-47c1-84ec-d4e89e480a20
Faulting package full name: Rebellion.52602B995C00E_1.2.621.0_x64__2vbwqmt31j4mr
Faulting package-relative application ID: Game

-- Example 02 ------------

Faulting application name: KingdomCome.exe, version: 1.9.6.0, time stamp: 0x63aa8b6b
Faulting module name: nvgpucomp64.dll, version: 32.0.15.6094, time stamp: 0x66bc3617
Exception code: 0xc0000005
Fault offset: 0x0000000000b9069a
Faulting process id: 0x0x35EC
Faulting application start time: 0x0x1DB04860C99B19E
Faulting application path: E:\SteamLibrary\steamapps\common\KingdomComeDeliverance\Bin\Win64\KingdomCome.exe
Faulting module path: C:\Windows\System32\DriverStore\FileRepository\nvmdi.inf_amd64_33559cc6c2fd215a\nvgpucomp64.dll
Report Id: 1719e1fe-953b-46b7-9598-56bcaf3ec2cd
Faulting package full name:
Faulting package-relative application ID:

It seems when ever the GPU is used, my PC crashes - it even crashes when I run GPU benchmarking tools.

The troubleshooting steps taken:
Replaced my PSU from the Corsair 850w to the corsair 1500w
Tried older drivers - and the most latest
re-inserting the GPU into the PCi4 slot
Placing it in a PCIe5
Replacing the power cables from my PSU
replacing the adapter from the GPU to the PSU.
The adapter that inserts into the GPU splits into 3, and each one of those have a pcie powercable directly into the PSU - None of them are piggybacked.
I have re-installed Windows
Ive tried a sfc scan - no issues found


At this point, I thought I must have a faulty GPU, so I returned it back to Amazon, and bought another.

Only to get the same thing happening again. Surely I cant be unlucky enough to get two faulty cards?

Its also worth mentioning, the prior to purchasing the GPU, the PC was running Ubuntu / KVM hosting around 10-15 virtual machines, always under high load - and never crashed once.
Its also worth mentioning, if I remove the GPU and go back to the intel integrated GPU, again my PC never crashes.

Neither my CPU or GPU have been over or under clocked - these are base installs without any custom configs.

I'd really appreciate any help! - before I return a second GPU

Cheers!

[Meatflaps]

Grab a copy of GPU-Z (Free), open it and leave it open, go play a game until it crashes then when it's crashed go back to GPU-Z that you left open, click the sensors tab, click the \/ down arror for Hot Spot and select "Show Highest Reading" then tell us what it is. Maybe your video cards are getting too hot.

EDIT: Also double check the power cable for the video card. Make sure the power connector is completely seated all the way into the card until the little plastic clip snaps in place and make sure there's no excessive bends in the cable putting any extra stress on the connector in some direction.

[raystevens73]

Found this post from someone with almost the same issue, it may help.

I've fixed the problem I had to go into my pc bios and enable my dynamic thermal tuning option and disable adaptive boost technology settings and change the performance core ratio for the cpu. After that I haven't got a nvgpucomp64.dll crash or a blue screen sense. Thank you anyways!

[Julien, cut it out.]

Post a CPU-Z validation link.

I don't think the issue stems from the GPU. You could try and test the GPU by stressing it at different power limits. The testing can be done at sock configuration with Unigine Superposition and MSI Afterburner f.e..

Some suggestions at this point in time:

> 2 DPC with 2R overclocked at 5600 MT/s might cause the data corruption.
I suggest to try the JEDEC standard of 4800 MT/s, which bandwidth corresponds to the IMC, or even lower.
ASRock does not mention that 4 DIMMs will run at that bandwidth.

> You may want to adjust the values of PL1 and PL2. At least look up the values for it. They might be overwritten by ASRock and differ from the Intel standard.
In combination with liquid cooling, the VRM might get too hot.

> Do some data monitoring with HwInfo64.

The motherboard is imo very underwhelming.

[Phénomènes Mystiques]

Use NVIDIA Debug Mode within the NVIDIA Control Panel "Help" tab and see if the games crash while using stock clock speeds because if it runs fine, the factory set the GPU clock too high or the GPU software application is set improperly, such as Zotac Firestorm application had done to mine - two other people found this helpful in the past.

[ˢᵈˣ FatCat]

Just lil bit input , if your GPU was faulty , the problem most case are : crash blackscreen on every game, artifact, or straight up no display at all

This more like Driver issue
More like the game or the driver cant pulled something from your game directory, kinda weird tho if you reinstalled windows suppoedly will be okay
Try turn off any security like antivirus Norton, kaspersky , usuyally they are overtight security

Gesus why ppl always put a suspect to PSU

[mrimo]

Hi all, thanks for your responses.

Unfortunately the issue has gone from bad to worse.

Now my PC won't boot at all. I get a permemant dram warning light on the motherboard and the cpu warning light flickers initially.

I've tried replacing the psu, reseating the cpu, changed ram configuration, placed each of the four sticks in slot 2, the dram light just stays and the pc won't post.

I've tried clearing cmos, removing the battery and using the jumpers to clear.

I'm at the point where I think either the CPU or motherboard is dead.

The pc has worked fine for years running virtual machines. I really regret the decision to get a 4090 lol

I've started with a replacement motherboard, but I've no idea at this point.

I think potentially my pc has over heated with the 4090, which may have killed the motherboard or cpu

I've no idea at this point.

Thanks again for your replies.

Il build it again over the next few days with the new motherboard and see what happens...

[Phénomènes Mystiques]

One time when my PC was brand new, on day 3 it would not boot at all, no lights nothing, so I am not sure if this will even help at all but what I did was pull out the MotherBoard main ATX cable, that big ol' sucker which is scary, and a PITA to get off, but after removing, putting some die-electric grease on it so it would not stick when it gets heated and yes these get hot on some systems and stick, but after plugging it back in (mind you this is done when the PSU back-switch is off or the PC is unplugged, never on stand-by), the thing started right up but I tell ya, when trying to get the plug out at first, it felt like the board was going to snap on the corner so careful if attempting this - but *poof* that was my last ditch attempt at fixing the new PC - hope this helps.

My question, OP, is were you using any GPU cable extenders, risers, or third-party cables of any kind/type for the GPU/PSU? - if so, this can end up bad or even be dangerous.

If you wanted to get really technical, it is possible to test the MotherBoard electric current using a micrometer similar to testing a PSU with a micrometer.

[mrimo]

Hey,

So I used the default cable that came with the gpu. No riser or cable extensions.

I have a multimeter, but truth be told, I wouldn't know where to start to test the motherboard .

I do wonder if when I rebuild, if I decide to go down the 4090 again, whether to water cool both the gpu and cpu, instead of air cool.

I'm hoping my cpu and ram are not dead.

I will test over the next few days.

[Phénomènes Mystiques]

Idk if this helps but when I got this GTX 1080Ti I had no idea the crashing reason and blamed it on the games themselves, later it turned out by going into the actual GPU application and reviewing their code, I found out that the GPU was forcing a much higher clock than it should of and had a defunct entry that was causing game instability due to higher than specified GPU clock timings, so after running NVIDIA Debug, it pointed me to look into that OC application and sure enough it was the issue, so after manually clocking the GPU to where it should have been + a few MHz's above the specified rating, the GPU ran better than new - I hope this can help.

If you are unsure about the micrometer thing, it is best not to go poking and prodding things you are unsure about, it was simply a suggestion if you were familiar with the testing, 'tis wiser to pull at cables when not electrified.

[Philco7a]

Your GPU is only running at 8x speed in the PCIE socket since you put a M.2 in the top slot which takes half the GPU's bandwidth to communicate directly with the CPU and bypass the chipset. You have too many M.2's for a gaming rig.

[Shaggin'Wagon]

Philco7a lähetti viestin:

Your GPU is only running at 8x speed in the PCIE socket since you put a M.2 in the top slot which takes half the GPU's bandwidth to communicate directly with the CPU and bypass the chipset. You have too many M.2's for a gaming rig.

That's actually not true. Yes the video card is running at x8 instead of x16, but that is not "half of it's speed". Reducing a card from 16x -> 8x is only reducing it's performance by about -5% to -10% max, certainly not -50% (half). It's trivial and not even a big enough change to really impact anything.

[Philco7a]

Shaggin'Wagon is correct. No GPU has been made that can fully use 16x Gen 4 pcie.
The Gen 5 cards get released in a couple months. Which makes no sense because they still can not fully utilize the Gen 4 socket.

But he's still only running it at 8x.

[r.linder]

Even 3.0 x16 is barely maxed out, the reason why 4.0 GPUs can lose performance on 3.0 is when they have fewer lanes to work with, 4.0 x8 cards lose a very minor amount of performance, a few frames here and there, while 4.0 x4 can lose a noticeable bit of performance.

But running a 4.0 x16 card at 3.0 x16 will have basically no difference because 3.0 x16 has around the same bandwidth as 4.0 x8, which is enough. Any performance loss that could be incurred from that will typically be nothing compared to the bottleneck from the CPU itself. Worry more about getting an R7 7800X3D to pair with a 4090 rather than just any 4.0 supported CPU.

[Julien, cut it out.]

Philco7a lähetti viestin:

Your GPU is only running at 8x speed in the PCIE socket since you put a M.2 in the top slot which takes half the GPU's bandwidth to communicate directly with the CPU and bypass the chipset. You have too many M.2's for a gaming rig.

You're wrong.
The PCIe lanes of the CPU are not shared on LGA1700 (Z690[i.pcmag.com]).

https://pg.asrock.com/mb/Intel/Z690%20Phantom%20Gaming%204/index.de.asp#Specification

[Phénomènes Mystiques]

Now I wonder if OP is using a stupid iGPU, man I hate these stupid things so much.

Why can't consumers just buy a proper non-iGPU CPU is beyond me.