Print a list of free CUDA devices on screen. This is useful to select a free GPU in a job submission script.

cuda_idle.c

/*
 * cuda_idle.c — print the index of each GPU with no compute process running.
 *
 * Build:  gcc -O2 -Wall -o gpu_idle gpu_idle.c -I/usr/local/cuda/include -lnvidia-ml
 *
 * MIT License
 *
 * Copyright (c) 2013-2026 Lars Pastewka
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in all
 * copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <stdio.h>
#include <stdlib.h>
#include <stdbool.h>
#include <nvml.h>

#define NVML_TRY(call)                                                  \
    do {                                                                \
        nvmlReturn_t rc_ = (call);                                      \
        if (rc_ != NVML_SUCCESS) {                                      \
            fprintf(stderr, "%s:%d: %s failed: %s\n",                   \
                    __FILE__, __LINE__, #call, nvmlErrorString(rc_));   \
            nvmlShutdown();                                             \
            exit(EXIT_FAILURE);                                         \
        }                                                               \
    } while (0)

/* A device is "idle" when no compute process is running on it. We query with
 * a zero-length buffer: NVML then reports the count without us allocating
 * anything or depending on the (version-dependent) layout of
 * nvmlProcessInfo_t. */
static bool device_is_idle(nvmlDevice_t dev)
{
    unsigned int count = 0;
    nvmlReturn_t rc = nvmlDeviceGetComputeRunningProcesses(dev, &count, NULL);

    if (rc == NVML_SUCCESS)
        return count == 0;              /* fit in the empty buffer -> none running */
    if (rc == NVML_ERROR_INSUFFICIENT_SIZE)
        return false;                   /* buffer too small -> at least one running */

    fprintf(stderr, "nvmlDeviceGetComputeRunningProcesses failed: %s\n",
            nvmlErrorString(rc));
    nvmlShutdown();
    exit(EXIT_FAILURE);
}

int main(void)
{
    NVML_TRY(nvmlInit());

    unsigned int count = 0;
    NVML_TRY(nvmlDeviceGetCount(&count));

    for (unsigned int i = 0; i < count; i++) {
        nvmlDevice_t dev;
        NVML_TRY(nvmlDeviceGetHandleByIndex(i, &dev));
        if (device_is_idle(dev))
            printf("%u\n", i);
    }

    NVML_TRY(nvmlShutdown());
    return 0;
}