Dell Precision M6700, NVIDIA Quadro P5000 GPU upgrade

No... The BSOD has happened only when the system is in a pretty "office work" type workload with GPU temperatures in the 40's. (Even under full load, with the fan going it stays in the 70's...)
As an update, I have not had a BSOD since I disabled LSPM, and that's been almost three weeks ago. (Still using DVI for the external display, so it is attached directly to the P5000.) I'm waiting a few more weeks before I call it "stable" but things are looking good.

[Edit]
Spoke too soon, just got the BSOD again.
Back to the VGA plan.

 

TDR has plagued Widows 10 and Nvidia since win10's beginning. My Solution (This was a long time ago) was to turn off TDR entirely .........problem was Windows kept undoing the registry changes ...........I finally had to use Task scheduler to implement running a little reg file I wrote (Which disabled tdr) at startup.

 

Pretty sure in this case there is some other issue that is actually freezing up the card, disabling TDR would not fix it. I set the TDR timeout to a ridiculously high value and all that happened was the system locked up for a while, and then once the timeout was met it BSOD'ed.
I'm trying to find out if it is related to power in some way. P5000 draws more power than Maxwell or Kepler cards, and there seems to be power contention between the CPU and GPU (and possibly other components?). I'm limiting the clock speed while the system is docked to keep the power use down. (Details in the 1070 thread, I will follow up in this thread as well if it is successful.)

 

Isn't more power available when it's docked or the is bus a bottleneck?

 

When I am playing undocked it is still on AC power. The power is the same docked or not if a 240W adapter is connected. (I don't game while on battery power, typically the system is only on battery power when I am moving it between locations, I have four 240W adapters scattered around.)

 

Got another BSOD today. This time, there were no external displays connected. The system was actually sitting idle with the lid shut, so the NVIDIA GPU should not have been very busy (actually with Optimus it should have been totally idle).

This time, the bluescreen blames dxgkrnl.sys (DirectX graphics kernel) but nvlddmkm.sys is also present in the stack trace, so it is related to something that the NVIDIA driver was doing or the NVIDIA GPU itself messing up.

I'm going to take another tact and try this thing which prevents the GPU from going into turbo mode, sticking with my theory that it is something power related that is causing this, and see if that makes any difference... Otherwise, though, if it continues to BSOD then I will be selling the card and buying another M5000M, I guess... Stability is more important than speed.

 

I managed to compile SetStablePowerState.exe using NVIDIA's source code. But, this thing doesn't cap the CPU clock speed, it basically forces it to the maxiumum non-turbo speed, so it isn't something that I can leave running all of the time. And I sort of would have to (if it even solves the issue, which I'm not going to find out). It increases the idle power draw from 4.5W to around 30W and it increases the idle temperature (no fan running) from the low 40's to the low 60's.

I haven't been able to find any other ways to cap the GPU's speed or power usage to something other than the P8/idle P-state, which I'm not really interested in doing.

I think that I am going to have to call this a failed experiment. I'll hang onto the card for a short time in case anyone has any ideas but I'm making plans to switch back to the M5000M.

 

I've had no issues with P5000m with windows 7 and driver 388 just saying. It has been both sitting idle/ monero mining for weeks in late last year and no BSOD. Maybe it's the issue with your card? I mean there are potato chips that can't even run on default clocks

 

Hi, I have the P4000 in my M6700 with the LVDS screen so optimus is enabled,
My P4000 show in device manager this id :
PCI\VEN_10DE&DEV_1BB7&SUBSYS_153F1028
So I tried to change a string in the nvdmwi.inf file like that :
PCI\VEN_10DE&DEV_1BB7&SUBSYS_07B11028 => PCI\VEN_10DE&DEV_1BB7&SUBSYS_153F1028
So the same thing as your P5000 except that the P4000 is "1BB7", I enable test signing mode and I boot windows with disable driver signature enforcement but the driver keep saying something like that (translated) :
"this driver is not compatible with this windows version" and "this driver does not find compatible graphics card"
I'm using this driver : https://www.nvidia.fr/Download/driverResults.aspx/147918/fr
If want to have a look on the nvdmwi.inf for this version, there is it (untouched by me) : https://www.mediafire.com/file/efwcvn18scm9oy7/nvdmwi.inf/file

My windows 10 is the "Entreprise LTSC" 1809.

Thanks a lot if you can figure out what's going on here or what I could do wrong..

 

It sounds like you have the right string to replace. Make sure that you changed all occurrences of "PCI\VEN_10DE&DEV_1BB7&SUBSYS_07B11028" in your INF file. There will probably be three different instances. They are for different versions of Windows 10 so you have to either pay attention to which group you are in or just go ahead and change all of them for it to work.

Just curious, how did you get the system to boot with the P4000? Did you have to change the vBIOS? What vBIOS version is loaded?

If you get it working, please follow up with how stable it is after a few weeks. I had to give up on the P5000 :-\
The person that I sold it to installed it in an older Alienware system and it has been working fine (or at least I haven't heard of any trouble and I think that I would hear if there was a problem), so again, I don't think that the card was actually bad. Just somehow it did not agree with my system.
Maybe I should try again with P4000