Detect if GPU is being used for graphical purposes

Recent results show that noise could be increased up to 50% due to X-Server running on the device. To warn users about the noisy environment, we could check if GPU is being used for graphical purposes. It's possible to get this kind of information with NVML:

#include <iostream>
#include <nvml.h>

void chck(nvmlReturn_t status) {
    if (status != NVML_SUCCESS) {
        throw std::runtime_error(std::string("Failed to initialize NVML: ") + nvmlErrorString(status));
    }
}

struct NVML_RAII {
    bool initialized = false;

    NVML_RAII() {
        chck(nvmlInit());
        initialized = true;
    }

    ~NVML_RAII() {
        if (initialized) {
            chck(nvmlShutdown());
        }
    }
};

int main() {
    NVML_RAII nvml_raii;
    unsigned int device_count;

    chck(nvmlDeviceGetCount(&device_count));

    for (unsigned int i = 0; i < device_count; i++) {
        nvmlDevice_t device;
        char name[NVML_DEVICE_NAME_BUFFER_SIZE];
        chck(nvmlDeviceGetHandleByIndex(i, &device));
        chck(nvmlDeviceGetName(device, name, NVML_DEVICE_NAME_BUFFER_SIZE));

        std::cout << "\t" << name << std::endl;

        unsigned int graphics_processes = 0;
        nvmlDeviceGetGraphicsRunningProcesses(device, &graphics_processes, nullptr);

        if (graphics_processes) {
            std::cerr << "X-Server is running, please turn it off (" << graphics_processes << ")" << std::endl;
        }

        unsigned int compute_processes = 0;
        nvmlDeviceGetComputeRunningProcesses(device, &compute_processes, nullptr);

        if (compute_processes) {
            std::cerr << "Target device is in use (" << compute_processes << ")" << std::endl;
        }
    }

    return 0;
}

The header could be easily overlooked, so we could also bring one's attention to the noisy environment by adding a column to the nvbench output.

Start by tracing how nvbench currently produces its output and where NVML-related initialization could fit. Use the issue's NVML graphics-process query as the reference, then determine whether the result belongs in a new output column or a warning. Done means graphical GPU use is detected and clearly reported without obscuring compute-process warnings.