Dell Precision M6700, NVIDIA Quadro P5000 GPU upgrade

Agreed, I went with the Quadro instead in this case because it is confirmed working in the M6700/M6800 with the ES vBIOS. The price that I got was barely any higher than the going rate for the 1070. AFAIK we still don't have a confirmed case of the 1070 working cleanly with Windows in the M6700/M6800 (though it seems like there are alternate vBIOS out there that might work?). I also like the ability to be on the ODE driver branch which isn't available for GeForce anymore (though probably some additional INF modding could address that, I guess).

 

Agreed but I also like the part where the P5000m has 16GB of vRAM instead of the 1070's 8GB. Seems like the P5000m is still bloody expensive and the 1070 is about half the P5000m's price!

@RMSMajestic I even saw your fleabay auction for your P5000m!

 

I saw that too

 

1. P5000m's 16GB is running @6000Mhz and I believe GTX 1070 is at 8000Mhz. Given that GM204 GTX 980m already have 256bit @6000Mhz I'd have no doubt that GP104 will be severely bottlenecked by that vRAM speed. That's my reason for opt for 8GB on GTX 1070 over 16GB on P5000m. Also even if the bandwidth is not the bottleneck, the core will still be long before the 16GB vRAM is filled. Only part useful is probably cryptonight heavy mining, which doesn't make any sense with a $1k P5000m instaed of Vega FE. Only if you have use for that massive 16GB vRAM,

2. Feebay, not fleabay. If you have noticed my price here is $100 cheaper than on feebay

 

New problem... One week after installing the P5000, got my first BSOD. I had my system locked, I walked away for a couple of minutes, and came back to this...



The system was docked with an external display being driven directly by the P5000. (Today is my first day since installing the P5000 running in this configuration for an extended period of time.)
System is normally totally stable, and this one totally blames the NVIDIA driver.

Researching...

 

TDR = Timeout detection and recovery

If Windows decides that the GPU is not responding, it will reset the graphics driver. (You might have observed this before... The desktop blanks out and then recovers with a notice in the corner saying that the graphics driver was reset.) If this operation fails or doesn't complete within a specified time, I guess it BSODs?

I found this unintelligible video and this Reddit post pointing to the registry value:
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers\TdrDelay (DWORD)

I dug around some more. Here is some Microsoft documentation on these registry values.
https://docs.microsoft.com/en-us/windows-hardware/drivers/display/tdr-registry-keys

The TdrDdiDelay one is interesting, it seems to indicate that if a thread gets stuck in the display driver for more than 5 seconds then it will BSOD with the error listed above, VIDEO_TDR_FAILURE. I'm going to try increasing this value and see if I get any more BSOD. It may well be a driver bug and not a hardware issue, there are reports from other users having this repeatedly and solving it by downgrading the driver...

Granted it has only happened once so far so it may be a fluke, but with the obvious recent hardware change, I don't believe so. I did upgrade to driver version 418 only after installing the new card, though...

 

I just saw this on the VFIO Reddit:

https://www.reddit.com/r/VFIO/comments/aquajw/started_getting_w10_guest_video_tdr_failure_in

Wonder if it's a bug in the newest drivers

 

Could be a bug in the newest drivers. I'm on an older driver and don't have this issue (fingers crossed)...

 

BSOD update.
I got another one 8 days later on 2/22.
Thinking that it might be related to the latest drivers, I reverted to 411.63 (from Dell, for P5000 in Precision 7720). I was clear for about two weeks but got the same BSOD again today. So, it's not related to the newest drivers. It just seems to be something with the P5000 and this system, unfortunately...

So...
It happens less than once a week and every single occurrence has been while I had the machine docked with an external monitor connected via DVI. I don't think that's a coincidence, I use the machine undocked a lot. So, as a workaround attempt I am now running with my external monitor connected via VGA, so it is attached to the Intel GPU / Optimus flow instead of directly to the NVIDIA GPU (just like the internal display). Maybe this will avoid the BSOD.

[Edit]
Read a suggestion to disable PCI Express link state power management and that might help... I'll switch back to DVI and give that a go.

 

I think you have problems with overheating