Monitoring NVENC hardware (Active or idle)

Question

we can query the available GPU with nvenc hardware like this:

     cuResult = cuInit(0);

if (cuResult != CUDA_SUCCESS)
{
    printf(">> GetNumberEncoders() - cuInit() failed error:0x%x\n", cuResult);
    exit(EXIT_FAILURE);
}

checkCudaErrors(cuDeviceGetCount(&deviceCount));

if (deviceCount == 0)
{
    printf(">> GetNumberEncoders() - reports no devices available that support CUDA\n");
    exit(EXIT_FAILURE);
}
else
{
    printf(">> GetNumberEncoders() has detected %d CUDA capable GPU device(s) <<\n", deviceCount);

    for (int currentDevice=0; currentDevice < deviceCount; currentDevice++)
    {
        checkCudaErrors(cuDeviceGet(&cuDevice, currentDevice));
        checkCudaErrors(cuDeviceGetName(gpu_name, 100, cuDevice));
        checkCudaErrors(cuDeviceComputeCapability(&SMmajor, &SMminor, currentDevice));
        printf("  [ GPU #%d - < %s > has Compute SM %d.%d, NVENC %s ]\n",
               currentDevice, gpu_name, SMmajor, SMminor,
               (((SMmajor << 4) + SMminor) >= 0x30) ? "Available" : "Not Available");

        if (((SMmajor << 4) + SMminor) >= 0x30)
        {
            encoderInfo[NVENC_devices].device = currentDevice;
            strcpy(encoderInfo[NVENC_devices].gpu_name, gpu_name);
            NVENC_devices++;
        }
    }
}

I have 8 GPU whit NVENC capability:

How can we check that specific NVENC hardware is now running or Idle. Is there any way to monitoring NVENC hardware ?

What about specific NVENC API function "OR" CUDA Driver or API functions that help me to find out which GPU or NVENC hardware is Idle?

NOTE: I know that CUDA and NVENC hardware are completely separate things but I am looking for Direct or Indirect (using Cuda API like using Cuda for specifying the available NVENC hardware) way for checking specific NVENC's status ???

score 2 · Accepted Answer · answered Aug 17 '13 at 00:57

2

you can monitor your NVENC hardware is Active or not in a indirect way using this command

nvidia-smi

to see specific NVENC hardware is busy or idle, because when you want to use specific NVENC you should create a cuda thread on that GPU first, and use its memory too, so by checking the status of the GPU we can extend it to its NVENC hardware.

answered Aug 17 '13 at 00:57

Farzad Salimi Jazi

760
10
25

this is My temporary answer to my question ,I wanted to help some people have my problem!!! why you down vote me , as I said this is indirect way , and now I am looking for using this way in my code !!!! (this is not your fault you don't know any thing about using NVENC) – Farzad Salimi Jazi Aug 17 '13 at 04:46
I used this method + script for generating test load for NVENC to encode, and by this way I successfully evaluate the method of using the 8 NVENC hardware and you down me !!!!! – Farzad Salimi Jazi Aug 17 '13 at 04:58
hope to help somebody temporarily :) – Farzad Salimi Jazi Aug 17 '13 at 21:42

Robert Crovella · Answer 2 · 2013-08-09T21:38:44.767

Regarding this question:

What about specific ... CUDA Driver or API functions that help me to find out which ... NVENC hardware is Idle?

There are none.

NVENC and CUDA are orthogonal. The APIs to access and manage CUDA are completely separate from the APIs to access and manage NVENC. The code you've shown here identifies what GPUs are in the system, and happens to use the CUDA API to do that, but your question otherwise has nothing to do with CUDA. There are no CUDA API functions that will allow you to do anything with NVENC hardware.

Ultimately, managing the NVENC hardware is the responsibility of the client application that is using the NVENC API. This client application can easily make a busy/non-busy status available upstream to other applications if desired (since it knows the individual status of the hardware it is managing.

Dear @Robert : I know that these are separate thing and I used OR between them. exactly I want to know in a indirect way like this way can I use cuda API to check that specific NVENC is busy or not because when you want to use specific NVENC you should create a cuda thread first so perhaps by checking the status of the GPU we can extend it to its NVENC hardware. and what about NVENC API , does this API have function that check the status of NVENC hardware? — Farzad Salimi Jazi, Aug 09 '13 at 21:21

Monitoring NVENC hardware (Active or idle)

2 Answers2