Skip to content

Add NVIDIA GPU monitoring via NVML #533

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wateenellende wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add NVIDIA GPU monitoring via NVML #533

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 10 additions & 4 deletions meson.build
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,12 +54,18 @@ app_dependencies = [
meson.get_compiler('c').find_library('pci'),
meson.get_compiler('vala').find_library('pci', dirs: vapidir),

# Allow building without nvidia libraries because they're not packaged or does not run
# on some distros due to incompatibility with muslc. Also they're not needed on ARM architecture
meson.get_compiler('c').find_library('XNVCtrl', required: get_option('nvidia').enabled()),
meson.get_compiler('vala').find_library('libxnvctrl', dirs: vapidir, required: get_option('nvidia').enabled()),
]

if get_option('nvidia').enabled()
if not meson.get_compiler('c').has_header('nvml.h')
error('NVIDIA support requested, but nvml.h was not found')
endif

app_dependencies += [
Copy link
Copy Markdown
Member

@stsdc stsdc Apr 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

use previous construction

meson.get_compiler('c').find_library('nvidia-ml', required: true),
]
endif

config_data = configuration_data()
config_data.set('GETTEXT_PACKAGE', meson.project_name())

Expand Down
230 changes: 112 additions & 118 deletions src/Resources/GPU/GPUNvidia.vala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,163 +3,157 @@
* SPDX-FileCopyrightText: 2025 elementary, Inc. (https://elementary.io)
*/

public class Monitor.GPUNvidia : IGPU, Object {

public Gee.HashMap<string, HwmonTemperature> hwmon_temperatures { get; set; }
[Compact]
Copy link
Copy Markdown
Member

@stsdc stsdc Apr 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

create a separate vapi file

[CCode (cname = "MonitorNvmlGpu", free_function = "monitor_nvml_gpu_free")]
private class MonitorNvmlGpu {
}

public string hwmon_module_name { get; set; }
[CCode (cname = "MonitorNvmlSample")]
private struct MonitorNvmlSample {
public uint gpu_percent;
public uint64 mem_used;
public uint64 mem_total;
public uint temperature_c;
public bool ok_util;
public bool ok_mem;
public bool ok_temp;
}

public string driver_name { get; set; }
[CCode (cname = "monitor_nvml_gpu_new_from_pci_bus_id")]
private extern MonitorNvmlGpu? monitor_nvml_gpu_new_from_pci_bus_id (string pci_bus_id);

public string name { get; set; }
[CCode (cname = "monitor_nvml_gpu_sample")]
private extern bool monitor_nvml_gpu_sample (MonitorNvmlGpu gpu, out MonitorNvmlSample sample);

public class Monitor.GPUNvidia : IGPU, Object {
public Gee.HashMap<string, HwmonTemperature> hwmon_temperatures { get; set; }
public string hwmon_module_name { get; protected set; }
public string driver_name { get; protected set; }
public string name { get; protected set; }
public int percentage { get; protected set; }

public int memory_percentage { get; protected set; }

public int fb_percentage { get; protected set; }

public double memory_vram_used { get; protected set; }

public double memory_vram_total { get; set; }

public double memory_vram_total { get; protected set; }
public double temperature { get; protected set; }
protected string sysfs_path { get; protected set; }

protected string sysfs_path { get; set; }

public int nvidia_temperature = 0;

public int nvidia_memory_vram_used = 0;

public int nvidia_memory_vram_total = 0;

public int nvidia_memory_percentage = 0;

public int nvidia_fb_percentage = 0;

public int nvidia_percentage = 0;

public char * nvidia_used = "";

public bool nvidia_resources_temperature;

public bool nvidia_resources_vram_used;

public bool nvidia_resources_vram_total;

public bool nvidia_resources_used;

public X.Display nvidia_display;
private MonitorNvmlGpu? nvml_gpu;
private MonitorNvmlSample sample_cache;

public GPUNvidia (Pci.Access pci_access, Pci.Dev pci_device) {
name = pci_parse_name (pci_access, pci_device);
name = "nVidia® " + name;

name = "NVIDIA " + pci_parse_name (pci_access, pci_device);
sysfs_path = pci_parse_sysfs_path (pci_access, pci_device);
driver_name = pci_device.get_string_property (Pci.FILL_DRIVER);
}
hwmon_module_name = driver_name;
hwmon_temperatures = new Gee.HashMap<string, HwmonTemperature> ();

construct {
nvidia_display = new X.Display ();
}

private void update_nv_resources () {
#if NVIDIA_SUPPORT
nvidia_resources_temperature = NVCtrl.XNVCTRLQueryAttribute (
nvidia_display,
0,
0,
NV_CTRL_GPU_CORE_TEMPERATURE,
&nvidia_temperature
);

if (!nvidia_resources_temperature) {
debug ("Could not query NV_CTRL_GPU_CORE_TEMPERATURE attribute!\n");
return;
string pci_bus_id = "%04x:%02x:%02x.%u".printf (
pci_device.domain_16,
pci_device.bus,
pci_device.dev,
pci_device.func
);

nvml_gpu = monitor_nvml_gpu_new_from_pci_bus_id (pci_bus_id);
if (nvml_gpu == null) {
warning ("Failed to initialize NVML handle for %s (%s)", name, pci_bus_id);
}
#endif
}

nvidia_resources_vram_used = NVCtrl.XNVCTRLQueryTargetAttribute (
nvidia_display,
NV_CTRL_TARGET_TYPE_GPU,
0,
0,
NV_CTRL_USED_DEDICATED_GPU_MEMORY,
&nvidia_memory_vram_used
);

if (!nvidia_resources_vram_used) {
debug ("Could not query NV_CTRL_USED_DEDICATED_GPU_MEMORY attribute!\n");
return;
private bool update_nv_resources () {
#if NVIDIA_SUPPORT
if (nvml_gpu == null) {
return false;
}

nvidia_resources_vram_total = NVCtrl.XNVCTRLQueryTargetAttribute (
nvidia_display,
NV_CTRL_TARGET_TYPE_GPU,
0,
0,
NV_CTRL_TOTAL_DEDICATED_GPU_MEMORY,
&nvidia_memory_vram_total
);

if (!nvidia_resources_vram_total) {
debug ("Could not query NV_CTRL_TOTAL_DEDICATED_GPU_MEMORY attribute!\n");
return;
if (!monitor_nvml_gpu_sample (nvml_gpu, out sample_cache)) {
return false;
}

nvidia_resources_used = NVCtrl.XNVCTRLQueryTargetStringAttribute (
nvidia_display,
NV_CTRL_TARGET_TYPE_GPU,
0,
0,
NV_CTRL_STRING_GPU_UTILIZATION,
&nvidia_used
);

// var str_used = (string)nvidia_used;
nvidia_percentage = int.parse (((string) nvidia_used).split_set ("=,")[1]);
nvidia_fb_percentage = int.parse (((string) nvidia_used).split_set ("=,")[3]);
debug ("USED_GRAPHICS: %d%\n", nvidia_percentage);
debug ("USED_FB_MEMORY: %d%\n", nvidia_fb_percentage);
return true;
#else
return false;
#endif

if (!nvidia_resources_used) {
debug ("Could not query NV_CTRL_STRING_GPU_UTILIZATION attribute!\n");
return;
}

}

private void update_temperature () {
temperature = nvidia_temperature;
}

private void update_memory_vram_used () {
memory_vram_used = (double) nvidia_memory_vram_used * 1000000.0;
public void update_temperature () {
#if NVIDIA_SUPPORT
if (sample_cache.ok_temp) {
temperature = (double) sample_cache.temperature_c;
} else {
temperature = 0;
}
#else
temperature = 0;
#endif
}

private void update_memory_vram_total () {
memory_vram_total = (double) nvidia_memory_vram_total * 1000000.0;
public void update_memory_vram_used () {
#if NVIDIA_SUPPORT
if (sample_cache.ok_mem) {
memory_vram_used = (double) sample_cache.mem_used;
} else {
memory_vram_used = 0;
}
#else
memory_vram_used = 0;
#endif
}

private void update_memory_percentage () {
memory_percentage = (int) (Math.round ((memory_vram_used / memory_vram_total) * 100));
public void update_memory_vram_total () {
#if NVIDIA_SUPPORT
if (sample_cache.ok_mem) {
memory_vram_total = (double) sample_cache.mem_total;
} else {
memory_vram_total = 0;
}
#else
memory_vram_total = 0;
#endif
}

private void update_fb_percentage () {
fb_percentage = nvidia_fb_percentage;
public void update_memory_percentage () {
if (memory_vram_total > 0) {
memory_percentage = (int) Math.round ((memory_vram_used / memory_vram_total) * 100.0);
fb_percentage = memory_percentage;
} else {
memory_percentage = 0;
fb_percentage = 0;
}
}

private void update_percentage () {
percentage = nvidia_percentage;
public void update_percentage () {
#if NVIDIA_SUPPORT
if (sample_cache.ok_util) {
percentage = (int) sample_cache.gpu_percent;
} else {
percentage = 0;
}
#else
percentage = 0;
#endif
}

public void update () {
update_nv_resources ();
bool ok = update_nv_resources ();

if (!ok) {
percentage = 0;
memory_percentage = 0;
fb_percentage = 0;
memory_vram_used = 0;
memory_vram_total = 0;
temperature = 0;
return;
}

update_temperature ();
update_memory_vram_used ();
update_memory_vram_total ();
update_memory_percentage ();
update_percentage ();
}

}
}
95 changes: 95 additions & 0 deletions src/Resources/GPU/nvml-bridge.c
View file
Open in desktop
Copy link
Copy Markdown
Member

@stsdc stsdc Apr 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why there's a c code?

Copy link
Copy Markdown
Author

@wateenellende wateenellende Apr 7, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To expose a minimum amount of C library to Vala, reducing risk.

Copy link
Copy Markdown
Member

@stsdc stsdc Apr 7, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What risk are you reducing?

Original file line number Diff line number Diff line change
@@ -0,0 +1,95 @@
#include "nvml-bridge.h"

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

struct MonitorNvmlGpu {
nvmlDevice_t handle;
};

static bool monitor_nvml_initialized = false;

static bool
monitor_nvml_init_once(void) {
if (monitor_nvml_initialized) {
return true;
}

nvmlReturn_t ret = nvmlInit_v2();
if (ret != NVML_SUCCESS) {
fprintf(stderr, "NVML init failed: %s\n", nvmlErrorString(ret));
return false;
}

monitor_nvml_initialized = true;
return true;
}

MonitorNvmlGpu*
monitor_nvml_gpu_new_from_pci_bus_id(const char* pci_bus_id) {
if (pci_bus_id == NULL || pci_bus_id[0] == '\0') {
return NULL;
}

if (!monitor_nvml_init_once()) {
return NULL;
}

MonitorNvmlGpu* gpu = calloc(1, sizeof(*gpu));
if (gpu == NULL) {
return NULL;
}

nvmlReturn_t ret = nvmlDeviceGetHandleByPciBusId_v2(pci_bus_id, &gpu->handle);
if (ret != NVML_SUCCESS) {
fprintf(stderr, "NVML PCI lookup failed for %s: %s\n", pci_bus_id, nvmlErrorString(ret));
free(gpu);
return NULL;
}

return gpu;
}

void
monitor_nvml_gpu_destroy(MonitorNvmlGpu* gpu) {
free(gpu);
}

bool
monitor_nvml_gpu_sample(MonitorNvmlGpu* gpu, MonitorNvmlSample* out) {
bool any_success = false;

if (gpu == NULL || out == NULL) {
return false;
}

memset(out, 0, sizeof(*out));

nvmlUtilization_t util = {0};
if (nvmlDeviceGetUtilizationRates(gpu->handle, &util) == NVML_SUCCESS) {
out->gpu_percent = util.gpu;
out->ok_util = true;
any_success = true;
}

nvmlMemory_t memory = {0};
if (nvmlDeviceGetMemoryInfo(gpu->handle, &memory) == NVML_SUCCESS) {
out->mem_used = memory.used;
out->mem_total = memory.total;
out->ok_mem = true;
any_success = true;
}

unsigned int temp = 0;
if (nvmlDeviceGetTemperature(gpu->handle, NVML_TEMPERATURE_GPU, &temp) == NVML_SUCCESS) {
out->temperature_c = temp;
out->ok_temp = true;
any_success = true;
}

return any_success;
}
Loading
Loading