RTX Pro 6000 TCC Mode

We just received a new Windows Workstation with a RTX Pro 6000 Blackwell GPU. It’s set to WDDM mode, and then it works fine. As soon as we set it to TCC mode, ater a reboot in the device manager the message This device cannot start (Code 10). Insufficient system resources exist to complete the API appears and the device can’t be used anymore, unless we reinstall the driver (which sets it back to WDDM mode). What could we be doing wrong?

Hi and thanks for posting, however this is the forum for Omniverse and GPU hardware. This is more of a generic GPU hardware question. Having said that, since you are here, if you are trying to run NVIDIA Omniverse you need the GPU to be in WDDM mode. Why are you trying to force the GPU to TCC mode unless you want data center compute “only”? The default should be “hybrid” mode, which does both. Keep it like that. You need WDDM mode to do anything with RTX and graphics.

I thought GPU hardware made sense, I missed the Omniverse part. I do want it to be on compute only, and don’t want to display anything (and don’t want to run the risk of any software trying to use it for graphics)

Ok so this is not the right forum then. You need to contact the Vendor you bought the system or GPU from and work through the “nvidia-smi” workflow to achieve the results you want. There are some videos on the internet about this. But I cannot help further here. I can tell you one big tip that is mentioned below. You really NEED at least one gpu, your GPU 0 set to Display mode / WDDM, for system reasons. The rest of the cards can be TCC for compute only, but not the primary. So if you have multiple cards in your system, no problem. If you have just one, then it will not be possible. It is a “workstation”, for work. It is not a data center. It NEEDS to be a workstation with a display. And what does this mean “run the risk of any software trying to use it for graphics”? Risk?

Bottom line is if you want pure “compute” you need a data center level server with a L40 /L40s. You are not really going to get that from a workstation with a single RTX enabled card like a PRO 6000. That particular card comes standard in “hybrid mode” anyway, meaning it designed for both at the same time and will auto detect the workflow. But it needs at least some GUI or monitor or virtual monitor. Again, it is a “workstation” not a data server.

Here is some advice from our AI. Please follow this.

On supported NVIDIA GPUs you switch from WDDM to TCC using nvidia-smi from an elevated command prompt, then reboot; you cannot do this on consumer GeForce cards or on the GPU that is driving your Windows display. docs.nvidia

1. Check if your GPU supports TCC

  1. Open an elevated Command Prompt or PowerShell (Run as administrator).

  2. Run:

    "C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe"
    

    In the output, look at the Disp.A and TCC/WDDM columns. Cards that support TCC will show a mode column like WDDM or TCC. 2022.help.altair

    • Many Tesla, some RTX A‑series, and some data‑center GPUs support TCC. docs.nvidia
    • GeForce and many mobile/workstation GPUs only support WDDM; if you try to switch them, nvidia-smi prints Unable to set driver model ... Not Supported. forums.developer.nvidia

2. Switch a GPU from WDDM to TCC

Assuming GPU 0 supports TCC and is not the one driving your monitor:

  1. In elevated cmd / PowerShell:

    "C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe" -i 0 -dm 1
    

    or equivalently:

    nvidia-smi -g 0 -dm 1
    
    • -i / -g selects the GPU index (as shown in nvidia-smi).
    • -dm 1 sets driver model to TCC (0 would be WDDM). leadergpu
  2. You should see a message like:

    Set driver model to TCC for GPU 00000000:03:00.0.
    Reboot required.
    ```  [leadergpu](https://www.leadergpu.com/articles/505-switch-gpu-modes-in-windows)
    
    
  3. Reboot Windows. After reboot, run nvidia-smi again to confirm the mode shows TCC for that GPU. leadergpu

3. Important caveats

  • You cannot put the primary display GPU into TCC; it must be headless or used only for compute. agisoft
  • Some datacenter GPUs require specific GRID / vGPU drivers or licensing to expose WDDM vs TCC; in those cases, you may need the appropriate driver stack instead of just -dm. nvidia
  • If nvidia-smi -dm 1 returns “Not Supported,” there is no supported way to force TCC on that card with the current driver. iwgh

On current RTX PRO 6000 Blackwell workstations, switching that card from WDDM to TCC on Windows and then getting a Code 10 “Insufficient system resources exist to complete the API” almost certainly indicates a driver / firmware or platform support problem, not a mistake in your procedure, and the practical answer is: you should keep it in WDDM (hybrid) mode for now, especially if you use Omniverse or any RTX graphics. nvidia

Why this is happening

  • RTX PRO 6000 Blackwell is marketed as a professional workstation GPU that is expected to run in WDDM mode on Windows for graphics, AI, and visualization workloads. nvidia
  • TCC mode on Windows is primarily intended for datacenter / compute‑only boards (Tesla, data‑center RTX, etc.); even when a GPU technically exposes TCC, it must be fully supported by the driver, firmware, and system BIOS. If anything in that stack is not ready, putting the card into TCC can cause the device to fail initialization and show Code 10. 2022.help.altair
  • The specific Code 10 “Insufficient system resources exist to complete the API” on high‑end GPUs often appears when there is a mismatch between the driver’s expectations (BAR size, memory layout, PCIe resources, or supported driver model) and what the OS or firmware actually provides for that device. forums.developer.nvidia