Nvidia driver crash only when using internal gpu, not AGA gpu.

I am having an issue with nvidia driver crashes. This only happens when using the internal gpu (970m on AW15 R1). It does not happen when I am using the AGA gpu (980 ti).

The crashes occur when using most 3d applications in windowed mode. This could be games, cad or 3d rendering software. The crashes don't happen right away, but within a few minutes of using the program. It varies. The type of crash is dependent on the nvidia driver version.

If the driver version is 4xx or newer, the driver will stop responding and driver service is restarted as reported by windows event viewer. All 3D applications using the gpu stall for a second and automatically terminate. There is a screen flicker when the driver service restarts.

If the driver version is 398.82 to 399.x, then I get a bluescreen usually about nvidia driver error.

If the driver is 398.36 or earlier, then it works perfectly fine and there are no issues.

I have looked at the release notes of the drivers and the only thing I see worth note is this mentioned in all driver releases from 397.64 to 399.24.
"Added new libraries (nvdlist.dll and nvdlistx.dll) to support Optimus and MSHybrid notebooks."

I am using Win10 x64 and always use DDU to uninstall drivers. I also have removed physx, 3dvision, gf experience to rule those out. Is anyone else experiencing these issues?

 

Nuke all nvidia drivers using DDU. Also, update Intel GPU drivers to latest available from Intel site. Now, download dell's nvidia driver and install it. Shutdown the PC and connect AGA and install the latest 4xx drivers on top of existing nvidia driver. Now, 980Ti will use 4xx version while disconnecting AGA, internal GPU will use dell driver and reduce glitches. @judal57 tested this method.

 

I will try that.

So this is a known issue with newer drivers and the internal gpu?

 

Well that's just an assumption though. Even we don't know if its known issue or not? There are only few cases.

 

After installing the 4xx driver with the AGA connected and switching over back to the 970m, the 970m will always load with the newer 4xx driver. Just cannot get the 970m to load with the dell installed nvidia driver after installing newer driver with the AGA gpu. Don't know what I am doing wrong. I am following the process exactly. I found @judal57 discussion in the AGA thread and used the same drivers as he posted on the youtube video, but just cant get it to work in my setup. Maybe because my AGA gpu is same generation as 970m, while Judyal's AGA gpu is pascal.

 

have you tried with DDU ?? and dont forget to disable the windows auto update for drivers

 

Ya I always use DDU in safe mode to remove nvidia drivers. I have tried installing drivers through the package installer in normal windows, safe mode windows, and also tried installing drivers via inf through device manager. But none of those methods work. Always overrides the older driver with the newer driver on the 970m.

I do have windows auto update drivers disabled, but it still always installs 389.12 automatically during the next normal windows session when there is no driver installed. So what I tried also doing is DDU remove drivers in safe mode, and then immediately install dell nvidia driver for the installer pkg for the 970m within same safe mode session. Then I turn off system, plug in AGA, boot straight into safe mode and install nvidia 4xx driver pkg for 980 ti. But when I boot back with 970m, its running on 4xx.

 

Then I think you have to install 4xx drivers on AGA 980Ti and disconnect and use manual updating of drivers through device manager using INF for internal dGPU. Have you tried installing 391.35 on internal dGPU and AGA?

 

I claimed this before, but Nvidia installs no matter what only one set of control DLL's and settings next to its driver suite. So you arent really installing 2 sets of drivers. You might see multiple file versions loaded into the Windows device manager, but those only load in the separated installed files that arent located in the usual driver folders but directly into system32 etc. Those arent dynamically loaded in based on which Nvidia GPU you are using. So basicall yin the end you just create a frankenstein driver.

 

Driver infs were always stored in C:\Windows\system32\drivers or drvstore folders.
You mean a ghost? DDU takes care of the ghost and so does Ghostbuster and Rapr/Driver store explorer. Those two tools are dangerous in the wrong hands that's why I didn't suggest the OP to use those tools.

 

They are stored in their respective folders but not in the root of system 32, and those also link to files stored in program files etc. Those do not change and cannot be changed on the fly by windows driver management.

 

Where are you living? With W10 you can switch from one driver to another on the fly w/o needing a reboot. Just reconnecting the external monitors is fine.
Intel GPU drivers are the only ones which needs reboot after each driver install/uninstall process.

 

I am not going to search my screenshots again. But large chunks of the Nvidia drivers live outside of the normal driver framework. In that case it always uses the last isntalled driver verions. What I am saying, Windows cannot replace the whole driver set based on changing the Nvidia GPU in use. Thats why I asked Judal at the time to share screenshots of the files being actually loaded in isntead of what is appearing in his uninstall folder. I never received a reply on that (And poor attempts of trying to discredit someone)

 

I think judal57's RAPR screenshot was in my phone. I don't know where its in camera roll or my cloud storage.

 

Yeah but i showed screenshots where you could actually see per dll which version was being loaded. But i deleted those.

 

I faintly remember it but I forgot which dlls were listed in devmgr.msc you showed on 1070 and 1070Ti on AGA.

 

Yeah, non-ti versions for both though. But yeah in my tests it could never full load older drivers for another GPU. I think this is also the case with the current user running against driver issues. The AGA software did it in the past though with some kind of wrapper, but got discontinued after Nvidia drivers became unified.

 

I think you mean nv_displ.inf and now its nvdmi.inf?

 

Not only, it loads about 40 dll's, and a lot of them live in program files/nvidia. When Windows loads in other drivers, it doesnt replace those files which causes the load of some sort of frankenstein driver.

 

You mean older drivers leftovers?

 

No mor elike. if you would install version 1 for the internal Geforce GPU and version 2 for the GPU in the AGA. It will use the core files of version 2 because that was installed last. If you disconnect the AGA it might revert to the version 1 drivers installed automatically, but the majority of the driver is living outside of the driver folders and that part isnt replaced by the other driver version. So it becomes a mix of driver version 1 and 2 at the same time. To see which driver versions are being loaded, you can check that in device manager and highly likely (in my testing) you see that it still loads the latest installed version no matter what, even if the driver panel itself says it is version 1.

 

Ok I didn't check that.

 

I just tried installing 391.25 via dell installer on the 970m, with 4xx drivers already installed and overwriting the original 391.25 drivers. Now both the 970m and the AGA 980 ti load with 391.25. Just can't get them to load separate drivers.

Interestingly, when I install 391.25 first and then 4xx, RAPR shows both driver infs. If I install 391.25 last, RAPR only shows 391.25 inf.

 

Now I'm mad, usually both resides in System32 folder as separate entities and only invoked/installed on matching IDs.
Maybe you need manually method on 980 Ti to use 4xx drivers or stick with 391.35 driver if you don't have any issues!

 

I just tried about everything, but can't get two drivers to load for the separate gpus. Last one installed always takes place for both gpus.

Looking more into the driver issue. Crashes happen when I alt tab or minimize to other windows, whether the game is in windowed mode or exclusive fullscreen. Doesn't happen in all games, but I noticed more frequently on dx11+ games. Also getting crashes in 3d applications like blender. The driver restarts itself every time this happens. Never happens on the AGA gpu, but for drivers after 398.11, happens on the internal 970m. Confirmed it happens even on a clean install win10 partition. Its not overclocking or overheating. I am checking temps and anyways it works fine on 398.11 and earlier drivers. Also tried downgrading from A09 to A06 bios, and went back to earlier AGA software that doesn't support hot removal.

On a side note I never knew there was throttling issue due to the PCH temps. Ive been using the AGA so much that I never noticed, since the PCH temps are way higher when using the internal gpu. Did a full maintenance on the laptop, cleaned all the dust, repasted cpu/gpu, and did a heatsink mod on the PCH. Used thermal pad + copper shim, and also a thermal pad bridge from the shim to the gpu heatplate. Taped everything down with electrical tape and used some spare thermal pad to press against the lower chassis to make it more secure. PCH temps are down by 20C. Now I don't get any throttling and cpu is full turbo boost as long as its under power limit. Thanks to @judal57 for his discovery on that.

 

I think Nvidia blocked it somehow. I hope you're using non DCH drivers.
I switched back from 418.xx to 391.25 from Dell and so far so good. Try 391.25 driver from Dell or nvidia and they should work fine with AGA.
You can use TS to switch off BD PROCHOT and bring down multipliers to 35-38x on all cores.

 

Not using DCH drivers. On an another note, I noticed that the 389.11 drivers work correctly with AGA hot removal. As in work correctly, I mean the AGA device actually is removed and disappears from device manager when I click "Restart Later" or yank the cable. With 4xx drivers, it only works correctly after a restart from internal gpu. But if I cold boot with AGA plugged in, this doesnt happen. The AGA gpu still appears in device manager and any program that tries to read the AGA gpu leads to a BSOD.

 

I think nvidia broke something in 4xx. Even in new Win 10 builds my Fn+ Combinations for AGA never prompts anything.

 

Same for me. I have to run graphicsamplifier.exe to get the prompts to work.