Problems with Setting a Titan RTX to TCC mode

Hi,

I’m trying to set up CUDA with Nsight debugging using a Titan RTX (in TCC mode) along side a Quadro K4200 which would run the graphics. I am using VS2010 and Win 7.

when I eventually got both cards to co-exist with the Quadro running the graphics - I then rebooted after setting :-
nvidia-smi.exe -g 0 -dm 1

At this point I got blank screens on reboot and ultimately had to remove the Titan card in order to get the display back.

My question is:-

Would upgrading to Win 10 help or is there something else I can do?

Cheers
Graham

motherboards can have their own heuristic for selecting which device will be the boot display device. nvidia-smi has no control over this. It’s not obvious to me that this is the problem, however.

also, -g 0 might not be the correct index for the Titan RTX

Hi RObert

At the mommenbt the Titan was/is in PCIe 0 and the Quadro in the PCI4 slot. Arre you suggesting that PCI 0 is the ‘natural’ home of the graphics card and hence I should swap them?

That might be the case. However the first thing I would check is the output from nvidia-smi to see which GPU is enumerated first. That will determine the proper index to use.

this was the output before I used the -g 0 switch

±----------------------------------------------------------------------------+
| NVIDIA-SMI 416.78 Driver Version: 416.78 CUDA Version: 10.0 |
|-------------------------------±---------------------±---------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Graphics Device WDDM | 00000000:01:00.0 On | N/A |
| N/A 38C P8 N/A / N/A | 360MiB / 24576MiB | 0% Default |
±------------------------------±---------------------±---------------------+
| 1 Quadro K4200 WDDM | 00000000:02:00.0 Off | N/A |
| 30% 36C P8 14W / 110W | 79MiB / 4096MiB | 0% Default |
±------------------------------±---------------------±---------------------+

So I Think the Titan is device zero in this set up…
I’ll try swapping

It’s not good that your driver says “Graphics Device” instead of naming the Titan RTX.

I would install the driver that comes with CUDA 10.1 Update 1 (425.25), before doing anything else.

https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64&target_version=7&target_type=exelocal

Hi Robert,

I reinstalled CUDA using your link above and I then swapped the cards and reinstalled the Titan RTX drivers (eventually)
Now when I run nvidia-smi.exe I get the following output.
±----------------------------------------------------------------------------+
| NVIDIA-SMI 430.86 Driver Version: 430.86 CUDA Version: 10.2 |
|-------------------------------±---------------------±---------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Quadro K4200 WDDM | 00000000:01:00.0 On | N/A |
| 32% 50C P8 23W / 110W | 301MiB / 4096MiB | 0% Default |
±------------------------------±---------------------±---------------------+
| 1 TITAN RTX WDDM | 00000000:02:00.0 Off | N/A |
| 41% 42C P8 15W / 280W | 385MiB / 24576MiB | 0% Default |
±------------------------------±---------------------±---------------------+

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 2364 C+G C:\windows\system32\Dwm.exe N/A |
±----------------------------------------------------------------------------+
Which looks better to me.

To ask a newbie question, If I want to switch the Titan to TCC what is the best way of doing it?
command line with which switches? or to go in via the Nvidia control panel somehow?

Thanks for your time

Cheers
Graham

I would do it just the way you indicated:

https://docs.nvidia.com/gameworks/content/developertools/desktop/nsight/tesla_compute_cluster.htm

nvidia-smi -g 1 -dm 1

(taking into account the new device mapping after you swapped GPU slots)

it might be that you can do it from the windows display control panel, but that’s not the way I’ve done it in the past. It will require a reboot in any event.

Hi Robert,

Success!
I had to run the command prompt as with admin privleges, but after the reboot, this is what I see:-

C:\Users\BLG5>“C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe”
Thu Jun 06 11:14:34 2019
±----------------------------------------------------------------------------+
| NVIDIA-SMI 430.86 Driver Version: 430.86 CUDA Version: 10.2 |
|-------------------------------±---------------------±---------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Quadro K4200 WDDM | 00000000:01:00.0 On | N/A |
| 31% 45C P8 23W / 110W | 300MiB / 4096MiB | 1% Default |
±------------------------------±---------------------±---------------------+
| 1 TITAN RTX TCC | 00000000:02:00.0 Off | N/A |
| 41% 31C P8 15W / 280W | 0MiB / 24281MiB | 0% Default |
±------------------------------±---------------------±---------------------+

So in recap, the problem was that the motherboard was not happy with the TCC GPU card in PCI slot zero - the WDDM card needs to be in PCI slot zero instead.

I also reloaded the Titan driver ( I already was trying to use the driver you suggested - but I reinstalled it anyway.

Thank you so much for your time - you have been incredibly helpful

Cheers
Graham