gpu: add device id label (#2186)

2026-02-08 05:56:37 +00:00 · 2025-08-28 06:36:10 +02:00
parent 71cedbc4d0
commit 0b8a257b31
11 changed files with 490 additions and 193 deletions
--- a/.idea/dictionaries/project.xml
+++ b/.idea/dictionaries/project.xml
@@ -4,6 +4,7 @@
      <w>containerd</w>
      <w>endpointstats</w>
      <w>gochecknoglobals</w>
+      <w>lpwstr</w>
      <w>luid</w>
      <w>operationoptions</w>
      <w>setupapi</w>
--- a/docs/collector.gpu.md
+++ b/docs/collector.gpu.md
@@ -20,28 +20,28 @@ These metrics are available on supported versions of Windows with compatible GPU

 ### Adapter-level Metrics

-| Name                                             | Description                                                                        | Type  | Labels        |
-|--------------------------------------------------|------------------------------------------------------------------------------------|-------|---------------|
-| `windows_gpu_info`                               | A metric with a constant '1' value labeled with gpu device information.            | gauge | `luid`,`name`,`bus_number`,`phys`,`function_number` |
-| `windows_gpu_dedicated_system_memory_size_bytes` | The size, in bytes, of memory that is dedicated from system memory.                | gauge | `luid`        |
-| `windows_gpu_dedicated_video_memory_size_bytes`  | The size, in bytes, of memory that is dedicated from video memory.                 | gauge | `luid`        |
-| `windows_gpu_shared_system_memory_size_bytes`    | The size, in bytes, of memory from system memory that can be shared by many users. | gauge | `luid`        |
-| `windows_gpu_adapter_memory_committed_bytes`     | Total committed GPU memory in bytes per physical GPU                               | gauge | `luid`,`phys` |
-| `windows_gpu_adapter_memory_dedicated_bytes`     | Dedicated GPU memory usage in bytes per physical GPU                               | gauge | `luid`,`phys` |
-| `windows_gpu_adapter_memory_shared_bytes`        | Shared GPU memory usage in bytes per physical GPU                                  | gauge | `luid`,`phys` |
-| `windows_gpu_local_adapter_memory_bytes`         | Local adapter memory usage in bytes per physical GPU                               | gauge | `luid`,`phys` |
-| `windows_gpu_non_local_adapter_memory_bytes`     | Non-local adapter memory usage in bytes per physical GPU                           | gauge | `luid`,`phys` |
+| Name                                             | Description                                                                        | Type  | Labels                                                          |
+|--------------------------------------------------|------------------------------------------------------------------------------------|-------|-----------------------------------------------------------------|
+| `windows_gpu_info`                               | A metric with a constant '1' value labeled with gpu device information.            | gauge | `bus_number`,`device_id`,`function_number`,`luid`,`name`,`phys` |
+| `windows_gpu_dedicated_system_memory_size_bytes` | The size, in bytes, of memory that is dedicated from system memory.                | gauge | `device_id`,`luid`                                              |
+| `windows_gpu_dedicated_video_memory_size_bytes`  | The size, in bytes, of memory that is dedicated from video memory.                 | gauge | `device_id`,`luid`                                              |
+| `windows_gpu_shared_system_memory_size_bytes`    | The size, in bytes, of memory from system memory that can be shared by many users. | gauge | `device_id`,`luid`                                              |
+| `windows_gpu_adapter_memory_committed_bytes`     | Total committed GPU memory in bytes per physical GPU                               | gauge | `device_id`,`luid`,`phys`                                       |
+| `windows_gpu_adapter_memory_dedicated_bytes`     | Dedicated GPU memory usage in bytes per physical GPU                               | gauge | `device_id`,`luid`,`phys`                                       |
+| `windows_gpu_adapter_memory_shared_bytes`        | Shared GPU memory usage in bytes per physical GPU                                  | gauge | `device_id`,`luid`,`phys`                                       |
+| `windows_gpu_local_adapter_memory_bytes`         | Local adapter memory usage in bytes per physical GPU                               | gauge | `device_id`,`luid`,`phys`,`part`                                |
+| `windows_gpu_non_local_adapter_memory_bytes`     | Non-local adapter memory usage in bytes per physical GPU                           | gauge | `device_id`,`luid`,`phys`,`part`                                |

 ### Per-process Metrics

-| Name                                         | Description                                     | Type    | Labels                                        |
-|----------------------------------------------|-------------------------------------------------|---------|-----------------------------------------------|
-| `windows_gpu_engine_time_seconds`            | Total running time of the GPU engine in seconds | counter | `luid`,`phys`, `eng`, `engtype`, `process_id` |
-| `windows_gpu_process_memory_committed_bytes` | Total committed GPU memory in bytes per process | gauge   | `luid`,`phys`,`process_id`                    |
-| `windows_gpu_process_memory_dedicated_bytes` | Dedicated GPU memory usage in bytes per process | gauge   | `luid`,`phys`,`process_id`                    |
-| `windows_gpu_process_memory_local_bytes`     | Local GPU memory usage in bytes per process     | gauge   | `luid`,`phys`,`process_id`                    |
-| `windows_gpu_process_memory_non_local_bytes` | Non-local GPU memory usage in bytes per process | gauge   | `luid`,`phys`,`process_id`                    |
-| `windows_gpu_process_memory_shared_bytes`    | Shared GPU memory usage in bytes per process    | gauge   | `luid`,`phys`,`process_id`                    |
+| Name                                         | Description                                     | Type    | Labels                                                    |
+|----------------------------------------------|-------------------------------------------------|---------|-----------------------------------------------------------|
+| `windows_gpu_engine_time_seconds`            | Total running time of the GPU engine in seconds | counter | `device_id`,`luid`,`phys`, `eng`, `engtype`, `process_id` |
+| `windows_gpu_process_memory_committed_bytes` | Total committed GPU memory in bytes per process | gauge   | `device_id`,`luid`,`phys`,`process_id`                    |
+| `windows_gpu_process_memory_dedicated_bytes` | Dedicated GPU memory usage in bytes per process | gauge   | `device_id`,`luid`,`phys`,`process_id`                    |
+| `windows_gpu_process_memory_local_bytes`     | Local GPU memory usage in bytes per process     | gauge   | `device_id`,`luid`,`phys`,`process_id`                    |
+| `windows_gpu_process_memory_non_local_bytes` | Non-local GPU memory usage in bytes per process | gauge   | `device_id`,`luid`,`phys`,`process_id`                    |
+| `windows_gpu_process_memory_shared_bytes`    | Shared GPU memory usage in bytes per process    | gauge   | `device_id`,`luid`,`phys`,`process_id`                    |

 ## Metric Labels

@@ -57,7 +57,7 @@ These are basic queries to help you get started with GPU monitoring on Windows u
 **Show GPU information for a specific physical GPU (0):**

 ```promql
-windows_gpu_info{description="NVIDIA GeForce GTX 1070",friendly_name="",hardware_id="PCI\\VEN_10DE&DEV_1B81&SUBSYS_61733842&REV_A1",phys="0",physical_device_object_name="\\Device\\NTPNP_PCI0027"} 1
+windows_gpu_info{bus_number="8",device_id="PCI\\VEN_10DE&DEV_1B81&SUBSYS_61733842&REV_A1",function_number="0",luid="0x00000000_0x00010F8A",name="NVIDIA GeForce GTX 1070",phys="0"} 1
 ```

 **Show total dedicated GPU memory (in bytes) usage on GPU 0:**
--- a/internal/collector/gpu/gpu.go
+++ b/internal/collector/gpu/gpu.go
@@ -21,9 +21,9 @@ import (
 	"errors"
 	"fmt"
 	"log/slog"
-	"strconv"

 	"github.com/alecthomas/kingpin/v2"
+	"github.com/prometheus-community/windows_exporter/internal/headers/cfgmgr32"
 	"github.com/prometheus-community/windows_exporter/internal/headers/gdi32"
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/pdh"
@@ -41,7 +41,7 @@ var ConfigDefaults = Config{}
 type Collector struct {
 	config Config

-	gpuDeviceCache map[string]gdi32.GPUDevice
+	gpuDeviceCache map[string]gpuDevice

 	// GPU Engine
 	gpuEnginePerfDataCollector *pdh.Collector
@@ -85,6 +85,12 @@ type Collector struct {
 	gpuProcessMemoryTotalCommitted *prometheus.Desc
 }

+type gpuDevice struct {
+	gdi32    gdi32.GPUDevice
+	cfgmgr32 cfgmgr32.Device
+	ID       string
+}
+
 func New(config *Config) *Collector {
 	if config == nil {
 		config = &ConfigDefaults
@@ -121,97 +127,97 @@ func (c *Collector) Build(_ *slog.Logger, _ *mi.Session) error {
 	c.gpuInfo = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "info"),
 		"A metric with a constant '1' value labeled with gpu device information.",
-		[]string{"luid", "name", "bus_number", "phys", "function_number"},
+		[]string{"luid", "device_id", "name", "bus_number", "phys", "function_number"},
 		nil,
 	)

 	c.gpuSharedSystemMemorySize = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "shared_system_memory_size_bytes"),
 		"The size, in bytes, of memory from system memory that can be shared by many users.",
-		[]string{"luid"},
+		[]string{"luid", "device_id"},
 		nil,
 	)
 	c.gpuDedicatedSystemMemorySize = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "dedicated_system_memory_size_bytes"),
 		"The size, in bytes, of memory that is dedicated from system memory.",
-		[]string{"luid"},
+		[]string{"luid", "device_id"},
 		nil,
 	)
 	c.gpuDedicatedVideoMemorySize = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "dedicated_video_memory_size_bytes"),
 		"The size, in bytes, of memory that is dedicated from video memory.",
-		[]string{"luid"},
+		[]string{"luid", "device_id"},
 		nil,
 	)

 	c.gpuEngineRunningTime = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "engine_time_seconds"),
 		"Total running time of the GPU in seconds.",
-		[]string{"process_id", "luid", "phys", "eng", "engtype"},
+		[]string{"process_id", "luid", "device_id", "phys", "eng", "engtype"},
 		nil,
 	)

 	c.gpuAdapterMemoryDedicatedUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "adapter_memory_dedicated_bytes"),
 		"Dedicated GPU memory usage in bytes.",
-		[]string{"luid", "phys"},
+		[]string{"luid", "device_id", "phys"},
 		nil,
 	)
 	c.gpuAdapterMemorySharedUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "adapter_memory_shared_bytes"),
 		"Shared GPU memory usage in bytes.",
-		[]string{"luid", "phys"},
+		[]string{"luid", "device_id", "phys"},
 		nil,
 	)
 	c.gpuAdapterMemoryTotalCommitted = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "adapter_memory_committed_bytes"),
 		"Total committed GPU memory in bytes.",
-		[]string{"luid", "phys"},
+		[]string{"luid", "device_id", "phys"},
 		nil,
 	)

 	c.gpuLocalAdapterMemoryUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "local_adapter_memory_bytes"),
 		"Local adapter memory usage in bytes.",
-		[]string{"luid", "phys"},
+		[]string{"luid", "device_id", "phys", "part"},
 		nil,
 	)

 	c.gpuNonLocalAdapterMemoryUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "non_local_adapter_memory_bytes"),
 		"Non-local adapter memory usage in bytes.",
-		[]string{"luid", "phys"},
+		[]string{"luid", "device_id", "phys", "part"},
 		nil,
 	)

 	c.gpuProcessMemoryDedicatedUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "process_memory_dedicated_bytes"),
 		"Dedicated process memory usage in bytes.",
-		[]string{"process_id", "luid", "phys"},
+		[]string{"process_id", "luid", "device_id", "phys"},
 		nil,
 	)
 	c.gpuProcessMemoryLocalUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "process_memory_local_bytes"),
 		"Local process memory usage in bytes.",
-		[]string{"process_id", "luid", "phys"},
+		[]string{"process_id", "luid", "device_id", "phys"},
 		nil,
 	)
 	c.gpuProcessMemoryNonLocalUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "process_memory_non_local_bytes"),
 		"Non-local process memory usage in bytes.",
-		[]string{"process_id", "luid", "phys"},
+		[]string{"process_id", "luid", "device_id", "phys"},
 		nil,
 	)
 	c.gpuProcessMemorySharedUsage = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "process_memory_shared_bytes"),
 		"Shared process memory usage in bytes.",
-		[]string{"process_id", "luid", "phys"},
+		[]string{"process_id", "luid", "device_id", "phys"},
 		nil,
 	)
 	c.gpuProcessMemoryTotalCommitted = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, Name, "process_memory_committed_bytes"),
 		"Total committed process memory in bytes.",
-		[]string{"process_id", "luid", "phys"},
+		[]string{"process_id", "luid", "device_id", "phys"},
 		nil,
 	)

@@ -253,11 +259,39 @@ func (c *Collector) Build(_ *slog.Logger, _ *mi.Session) error {
 		}

 		if c.gpuDeviceCache == nil {
-			c.gpuDeviceCache = make(map[string]gdi32.GPUDevice)
+			c.gpuDeviceCache = make(map[string]gpuDevice)
 		}

 		luidKey := fmt.Sprintf("0x%08X_0x%08X", gpu.LUID.HighPart, gpu.LUID.LowPart)
-		c.gpuDeviceCache[luidKey] = gpu
+
+		deviceID := gpu.DeviceID
+
+		cfgmgr32Devs, err := cfgmgr32.GetDevicesInstanceIDs(gpu.DeviceID)
+		if err != nil {
+			errs = append(errs, fmt.Errorf("failed to get device instance IDs for device ID %s: %w", gpu.DeviceID, err))
+		}
+
+		var cfgmgr32Dev cfgmgr32.Device
+
+		for _, dev := range cfgmgr32Devs {
+			if dev.BusNumber == gpu.BusNumber && dev.DeviceNumber == gpu.DeviceNumber && dev.FunctionNumber == gpu.FunctionNumber {
+				cfgmgr32Dev = dev
+
+				break
+			}
+		}
+
+		if cfgmgr32Dev.InstanceID == "" {
+			errs = append(errs, fmt.Errorf("failed to find matching device for device ID %s", gpu.DeviceID))
+		} else {
+			deviceID = cfgmgr32Dev.InstanceID
+		}
+
+		c.gpuDeviceCache[luidKey] = gpuDevice{
+			gdi32:    gpu,
+			cfgmgr32: cfgmgr32Dev,
+			ID:       deviceID,
+		}
 	}

 	return errors.Join(errs...)
@@ -298,31 +332,32 @@ func (c *Collector) collectGpuInfo(ch chan<- prometheus.Metric) {
 			prometheus.GaugeValue,
 			1.0,
 			luid,
-			gpu.AdapterString,
-			strconv.FormatInt(int64(gpu.BusNumber), 10),
-			strconv.FormatInt(int64(gpu.DeviceNumber), 10),
-			strconv.FormatInt(int64(gpu.FunctionNumber), 10),
+			gpu.ID,
+			gpu.gdi32.AdapterString,
+			gpu.gdi32.BusNumber.String(),
+			gpu.gdi32.DeviceNumber.String(),
+			gpu.gdi32.FunctionNumber.String(),
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuSharedSystemMemorySize,
 			prometheus.GaugeValue,
-			float64(gpu.SharedSystemMemorySize),
-			luid,
+			float64(gpu.gdi32.SharedSystemMemorySize),
+			luid, gpu.ID,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuDedicatedSystemMemorySize,
 			prometheus.GaugeValue,
-			float64(gpu.DedicatedSystemMemorySize),
-			luid,
+			float64(gpu.gdi32.DedicatedSystemMemorySize),
+			luid, gpu.ID,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuDedicatedVideoMemorySize,
 			prometheus.GaugeValue,
-			float64(gpu.DedicatedVideoMemorySize),
-			luid,
+			float64(gpu.gdi32.DedicatedVideoMemorySize),
+			luid, gpu.ID,
 		)
 	}
 }
@@ -333,31 +368,20 @@ func (c *Collector) collectGpuEngineMetrics(ch chan<- prometheus.Metric) error {
 		return fmt.Errorf("failed to collect GPU Engine perf data: %w", err)
 	}

-	runningTimeMap := make(map[PidPhysEngEngType]float64)
 	// Iterate over the GPU Engine perf data and aggregate the values.
 	for _, data := range c.gpuEnginePerfDataObject {
 		instance := parseGPUCounterInstanceString(data.Name)

-		if _, ok := c.gpuDeviceCache[instance.Luid]; !ok {
+		device, ok := c.gpuDeviceCache[instance.Luid]
+		if !ok {
 			continue
 		}

-		key := PidPhysEngEngType{
-			Pid:     instance.Pid,
-			Phys:    instance.Phys,
-			Luid:    instance.Luid,
-			Eng:     instance.Eng,
-			Engtype: instance.Engtype,
-		}
-		runningTimeMap[key] += data.RunningTime / 10_000_000 // RunningTime is in 100ns units, convert to seconds.
-	}
-
-	for key, runningTime := range runningTimeMap {
 		ch <- prometheus.MustNewConstMetric(
 			c.gpuEngineRunningTime,
 			prometheus.CounterValue,
-			runningTime,
-			key.Pid, key.Luid, key.Phys, key.Eng, key.Engtype,
+			data.RunningTime/10_000_000,
+			instance.Pid, instance.Luid, device.ID, instance.Phys, instance.Eng, instance.Engtype,
 		)
 	}

@@ -370,49 +394,33 @@ func (c *Collector) collectGpuAdapterMemoryMetrics(ch chan<- prometheus.Metric)
 		return fmt.Errorf("failed to collect GPU Adapter Memory perf data: %w", err)
 	}

-	dedicatedUsageMap := make(map[PidPhysEngEngType]float64)
-	sharedUsageMap := make(map[PidPhysEngEngType]float64)
-	totalCommittedMap := make(map[PidPhysEngEngType]float64)
-
 	for _, data := range c.gpuAdapterMemoryPerfDataObject {
 		instance := parseGPUCounterInstanceString(data.Name)

-		if _, ok := c.gpuDeviceCache[instance.Luid]; !ok {
+		device, ok := c.gpuDeviceCache[instance.Luid]
+		if !ok {
 			continue
 		}

-		key := PidPhysEngEngType{
-			Pid:     instance.Pid,
-			Luid:    instance.Luid,
-			Phys:    instance.Phys,
-			Eng:     instance.Eng,
-			Engtype: instance.Engtype,
-		}
-		dedicatedUsageMap[key] += data.DedicatedUsage
-		sharedUsageMap[key] += data.SharedUsage
-		totalCommittedMap[key] += data.TotalCommitted
-	}
-
-	for key, dedicatedUsage := range dedicatedUsageMap {
 		ch <- prometheus.MustNewConstMetric(
 			c.gpuAdapterMemoryDedicatedUsage,
 			prometheus.GaugeValue,
-			dedicatedUsage,
-			key.Luid, key.Phys,
+			data.DedicatedUsage,
+			instance.Luid, device.ID, instance.Phys,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuAdapterMemorySharedUsage,
 			prometheus.GaugeValue,
-			sharedUsageMap[key],
-			key.Luid, key.Phys,
+			data.SharedUsage,
+			instance.Luid, device.ID, instance.Phys,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuAdapterMemoryTotalCommitted,
 			prometheus.GaugeValue,
-			totalCommittedMap[key],
-			key.Luid, key.Phys,
+			data.TotalCommitted,
+			instance.Luid, device.ID, instance.Phys,
 		)
 	}

@@ -425,29 +433,19 @@ func (c *Collector) collectGpuLocalAdapterMemoryMetrics(ch chan<- prometheus.Met
 		return fmt.Errorf("failed to collect GPU Local Adapter Memory perf data: %w", err)
 	}

-	localAdapterMemoryMap := make(map[PidPhysEngEngType]float64)
-
 	for _, data := range c.gpuLocalAdapterMemoryPerfDataObject {
 		instance := parseGPUCounterInstanceString(data.Name)

-		if _, ok := c.gpuDeviceCache[instance.Luid]; !ok {
+		device, ok := c.gpuDeviceCache[instance.Luid]
+		if !ok {
 			continue
 		}

-		key := PidPhysEngEngType{
-			Luid: instance.Luid,
-			Phys: instance.Phys,
-		}
-
-		localAdapterMemoryMap[key] += data.LocalUsage
-	}
-
-	for key, localUsage := range localAdapterMemoryMap {
 		ch <- prometheus.MustNewConstMetric(
 			c.gpuLocalAdapterMemoryUsage,
 			prometheus.GaugeValue,
-			localUsage,
-			key.Luid, key.Phys,
+			data.LocalUsage,
+			instance.Luid, device.ID, instance.Phys, instance.Part,
 		)
 	}

@@ -460,28 +458,19 @@ func (c *Collector) collectGpuNonLocalAdapterMemoryMetrics(ch chan<- prometheus.
 		return fmt.Errorf("failed to collect GPU Non Local Adapter Memory perf data: %w", err)
 	}

-	nonLocalAdapterMemoryMap := make(map[PidPhysEngEngType]float64)
-
 	for _, data := range c.gpuNonLocalAdapterMemoryPerfDataObject {
 		instance := parseGPUCounterInstanceString(data.Name)

-		if _, ok := c.gpuDeviceCache[instance.Luid]; !ok {
+		device, ok := c.gpuDeviceCache[instance.Luid]
+		if !ok {
 			continue
 		}

-		key := PidPhysEngEngType{
-			Luid: instance.Luid,
-			Phys: instance.Phys,
-		}
-		nonLocalAdapterMemoryMap[key] += data.NonLocalUsage
-	}
-
-	for key, nonLocalUsage := range nonLocalAdapterMemoryMap {
 		ch <- prometheus.MustNewConstMetric(
 			c.gpuNonLocalAdapterMemoryUsage,
 			prometheus.GaugeValue,
-			nonLocalUsage,
-			key.Luid, key.Phys,
+			data.NonLocalUsage,
+			instance.Luid, device.ID, instance.Phys, instance.Part,
 		)
 	}

@@ -494,65 +483,47 @@ func (c *Collector) collectGpuProcessMemoryMetrics(ch chan<- prometheus.Metric)
 		return fmt.Errorf("failed to collect GPU Process Memory perf data: %w", err)
 	}

-	processDedicatedUsageMap := make(map[PidPhys]float64)
-	processLocalUsageMap := make(map[PidPhys]float64)
-	processNonLocalUsageMap := make(map[PidPhys]float64)
-	processSharedUsageMap := make(map[PidPhys]float64)
-	processTotalCommittedMap := make(map[PidPhys]float64)
-
 	for _, data := range c.gpuProcessMemoryPerfDataObject {
 		instance := parseGPUCounterInstanceString(data.Name)

-		if _, ok := c.gpuDeviceCache[instance.Luid]; !ok {
+		device, ok := c.gpuDeviceCache[instance.Luid]
+		if !ok {
 			continue
 		}

-		key := PidPhys{
-			Pid:  instance.Pid,
-			Luid: instance.Luid,
-			Phys: instance.Phys,
-		}
-		processDedicatedUsageMap[key] += data.DedicatedUsage
-		processLocalUsageMap[key] += data.LocalUsage
-		processNonLocalUsageMap[key] += data.NonLocalUsage
-		processSharedUsageMap[key] += data.SharedUsage
-		processTotalCommittedMap[key] += data.TotalCommitted
-	}
-
-	for key, dedicatedUsage := range processDedicatedUsageMap {
 		ch <- prometheus.MustNewConstMetric(
 			c.gpuProcessMemoryDedicatedUsage,
 			prometheus.GaugeValue,
-			dedicatedUsage,
-			key.Pid, key.Luid, key.Phys,
+			data.DedicatedUsage,
+			instance.Pid, instance.Luid, device.ID, instance.Phys,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuProcessMemoryLocalUsage,
 			prometheus.GaugeValue,
-			processLocalUsageMap[key],
-			key.Pid, key.Luid, key.Phys,
+			data.LocalUsage,
+			instance.Pid, instance.Luid, device.ID, instance.Phys,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuProcessMemoryNonLocalUsage,
 			prometheus.GaugeValue,
-			processNonLocalUsageMap[key],
-			key.Pid, key.Luid, key.Phys,
+			data.NonLocalUsage,
+			instance.Pid, instance.Luid, device.ID, instance.Phys,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuProcessMemorySharedUsage,
 			prometheus.GaugeValue,
-			processSharedUsageMap[key],
-			key.Pid, key.Luid, key.Phys,
+			data.SharedUsage,
+			instance.Pid, instance.Luid, device.ID, instance.Phys,
 		)

 		ch <- prometheus.MustNewConstMetric(
 			c.gpuProcessMemoryTotalCommitted,
 			prometheus.GaugeValue,
-			processTotalCommittedMap[key],
-			key.Pid, key.Luid, key.Phys,
+			data.TotalCommitted,
+			instance.Pid, instance.Luid, device.ID, instance.Phys,
 		)
 	}

--- a/internal/collector/gpu/utils.go
+++ b/internal/collector/gpu/utils.go
@@ -23,26 +23,29 @@ import (
 )

 type Instance struct {
-	Pid     string
-	Luid    string
-	Phys    string
-	Eng     string
-	Engtype string
-	Part    string
+	Pid      string
+	Luid     string
+	DeviceID string
+	Phys     string
+	Eng      string
+	Engtype  string
+	Part     string
 }

 type PidPhys struct {
-	Pid  string
-	Luid string
-	Phys string
+	Pid      string
+	Luid     string
+	DeviceID string
+	Phys     string
 }

 type PidPhysEngEngType struct {
-	Pid     string
-	Luid    string
-	Phys    string
-	Eng     string
-	Engtype string
+	Pid      string
+	Luid     string
+	DeviceID string
+	Phys     string
+	Eng      string
+	Engtype  string
 }

 func parseGPUCounterInstanceString(s string) Instance {
--- a/internal/headers/cfgmgr32/cfgmgr32.go
+++ b/internal/headers/cfgmgr32/cfgmgr32.go
@@ -0,0 +1,92 @@
+// SPDX-License-Identifier: Apache-2.0
+//
+// Copyright The Prometheus Authors
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package cfgmgr32
+
+import (
+	"fmt"
+	"unsafe"
+
+	"github.com/prometheus-community/windows_exporter/internal/headers/win32"
+	"golang.org/x/sys/windows"
+)
+
+func GetDevicesInstanceIDs(deviceID string) ([]Device, error) {
+	var (
+		err      error
+		listSize uint32
+	)
+
+	deviceIDLWStr := win32.NewLPWSTR(deviceID)
+
+	err = CMGetDeviceIDListSize(deviceIDLWStr, &listSize)
+	if err != nil {
+		return nil, err
+	}
+
+	listBuffer := make([]uint16, listSize)
+
+	err = CMGetDeviceIDList(deviceIDLWStr, listBuffer)
+	if err != nil {
+		return nil, err
+	}
+
+	deviceInstanceIDs := win32.ParseMultiSz(listBuffer)
+	devices := make([]Device, 0, len(deviceInstanceIDs))
+
+	for _, deviceInstanceID := range deviceInstanceIDs {
+		var devNode *windows.Handle
+
+		err = CMLocateDevNode(&devNode, deviceInstanceID)
+		if err != nil {
+			return nil, err
+		}
+
+		var (
+			busNumber     uint32
+			deviceAddress uint32
+			propType      uint32
+		)
+
+		propLen := uint32(4)
+
+		err = CMGetDevNodeProperty(devNode, DEVPKEYDeviceBusNumber, &propType, unsafe.Pointer(&busNumber), &propLen)
+		if err != nil {
+			return nil, err
+		}
+
+		if propType != DEVPROP_TYPE_UINT32 {
+			return nil, fmt.Errorf("unexpected property type: 0x%08X", propType)
+		}
+
+		err = CMGetDevNodeProperty(devNode, DEVPKEYDeviceAddress, &propType, unsafe.Pointer(&deviceAddress), &propLen)
+		if err != nil {
+			return nil, err
+		}
+
+		if propType != DEVPROP_TYPE_UINT32 {
+			return nil, fmt.Errorf("unexpected property type: 0x%08X", propType)
+		}
+
+		devices = append(devices, Device{
+			InstanceID:     windows.UTF16ToString(deviceInstanceID),
+			BusNumber:      win32.UINT(busNumber),
+			DeviceNumber:   win32.UINT(deviceAddress >> 16),
+			FunctionNumber: win32.UINT(deviceAddress & 0xFFFF),
+		})
+	}
+
+	return devices, nil
+}
--- a/internal/headers/cfgmgr32/syscall.go
+++ b/internal/headers/cfgmgr32/syscall.go
@@ -0,0 +1,94 @@
+// SPDX-License-Identifier: Apache-2.0
+//
+// Copyright The Prometheus Authors
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package cfgmgr32
+
+import (
+	"fmt"
+	"unsafe"
+
+	"github.com/prometheus-community/windows_exporter/internal/headers/win32"
+	"golang.org/x/sys/windows"
+)
+
+//nolint:gochecknoglobals
+var (
+	cfgmgr32 = windows.NewLazySystemDLL("cfgmgr32.dll")
+
+	procCMGetDeviceIDListW    = cfgmgr32.NewProc("CM_Get_Device_ID_ListW")
+	procCMGetDeviceIDListSize = cfgmgr32.NewProc("CM_Get_Device_ID_List_SizeW")
+	procCMGetDevNodePropertyW = cfgmgr32.NewProc("CM_Get_DevNode_PropertyW")
+	procCMLocateDevNodeW      = cfgmgr32.NewProc("CM_Locate_DevNodeW")
+)
+
+func CMGetDeviceIDListSize(filter *win32.LPWSTR, size *uint32) error {
+	ret, _, _ := procCMGetDeviceIDListSize.Call(
+		uintptr(unsafe.Pointer(size)),
+		filter.Pointer(),
+		uintptr(CM_GETIDLIST_FILTER_PRESENT|CM_GETIDLIST_FILTER_ENUMERATOR),
+	)
+
+	if ret != CR_SUCCESS {
+		return fmt.Errorf("CMGetDeviceIDListSize failed: 0x%02X", ret)
+	}
+
+	return nil
+}
+
+func CMGetDeviceIDList(filter *win32.LPWSTR, buf []uint16) error {
+	ret, _, _ := procCMGetDeviceIDListW.Call(
+		filter.Pointer(),
+		uintptr(unsafe.Pointer(&buf[0])),
+		uintptr(len(buf)),
+		uintptr(CM_GETIDLIST_FILTER_PRESENT|CM_GETIDLIST_FILTER_ENUMERATOR),
+	)
+
+	if ret != CR_SUCCESS {
+		return fmt.Errorf("CMGetDeviceIDList failed: 0x%02X", ret)
+	}
+
+	return nil
+}
+
+func CMLocateDevNode(devInst **windows.Handle, deviceID []uint16) error {
+	ret, _, _ := procCMLocateDevNodeW.Call(
+		uintptr(unsafe.Pointer(devInst)),
+		uintptr(unsafe.Pointer(&deviceID[0])),
+		0,
+	)
+
+	if ret != CR_SUCCESS {
+		return fmt.Errorf("CMLocateDevNode failed: 0x%02X", ret)
+	}
+
+	return nil
+}
+
+func CMGetDevNodeProperty(devInst *windows.Handle, propKey *DEVPROPKEY, propType *uint32, buf unsafe.Pointer, bufLen *uint32) error {
+	ret, _, _ := procCMGetDevNodePropertyW.Call(
+		uintptr(unsafe.Pointer(devInst)),
+		uintptr(unsafe.Pointer(propKey)),
+		uintptr(unsafe.Pointer(propType)),
+		uintptr(buf),
+		uintptr(unsafe.Pointer(bufLen)),
+		0,
+	)
+
+	if ret != CR_SUCCESS {
+		return fmt.Errorf("CMGetDevNodeProperty failed: 0x%02X", ret)
+	}
+
+	return nil
+}
--- a/internal/headers/cfgmgr32/types.go
+++ b/internal/headers/cfgmgr32/types.go
@@ -0,0 +1,70 @@
+// SPDX-License-Identifier: Apache-2.0
+//
+// Copyright The Prometheus Authors
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package cfgmgr32
+
+import (
+	"github.com/go-ole/go-ole"
+	"github.com/prometheus-community/windows_exporter/internal/headers/win32"
+)
+
+const (
+	// Configuration Manager return codes
+	CR_SUCCESS = 0x00
+
+	// Filter flags
+	CM_GETIDLIST_FILTER_ENUMERATOR = 0x00000001
+	CM_GETIDLIST_FILTER_PRESENT    = 0x00000100
+
+	DEVPROP_TYPE_UINT32 uint32 = 0x00000007
+)
+
+// DEVPROPKEY represents a device property key (GUID + pid)
+type DEVPROPKEY struct {
+	FmtID ole.GUID
+	PID   uint32
+}
+
+type Device struct {
+	InstanceID     string
+	BusNumber      win32.UINT
+	DeviceNumber   win32.UINT
+	FunctionNumber win32.UINT
+}
+
+//nolint:gochecknoglobals
+var (
+	// https://github.com/Infinidat/infi.devicemanager/blob/8be9ead6b04ff45c63d9e3bc70d82cceafb75c47/src/infi/devicemanager/setupapi/properties.py#L138C1-L143C34
+	DEVPKEYDeviceBusNumber = &DEVPROPKEY{
+		FmtID: ole.GUID{
+			Data1: 0xa45c254e,
+			Data2: 0xdf1c,
+			Data3: 0x4efd,
+			Data4: [8]byte{0x80, 0x20, 0x67, 0xd1, 0x46, 0xa8, 0x50, 0xe0},
+		},
+		PID: 23, // DEVPROP_TYPE_UINT32
+	}
+
+	// https://github.com/Infinidat/infi.devicemanager/blob/8be9ead6b04ff45c63d9e3bc70d82cceafb75c47/src/infi/devicemanager/setupapi/properties.py#L187-L192
+	DEVPKEYDeviceAddress = &DEVPROPKEY{
+		FmtID: ole.GUID{
+			Data1: 0xa45c254e,
+			Data2: 0xdf1c,
+			Data3: 0x4efd,
+			Data4: [8]byte{0x80, 0x20, 0x67, 0xd1, 0x46, 0xa8, 0x50, 0xe0},
+		},
+		PID: 30, // DEVPROP_TYPE_UINT32
+	}
+)
--- a/internal/headers/gdi32/gdi32.go
+++ b/internal/headers/gdi32/gdi32.go
@@ -34,41 +34,12 @@ const (
 	KMTQAITYPE_ADAPTERADDRESS = 6
 	// KMTQAITYPE_ADAPTERREGISTRYINFO pPrivateDriverData points to a D3DKMT_ADAPTERREGISTRYINFO structure that contains registry information about the graphics adapter.
 	KMTQAITYPE_ADAPTERREGISTRYINFO = 8
+	// KMTQAITYPE_PHYSICALADAPTERDEVICEIDS pPrivateDriverData points to a D3DKMT_QUERY_DEVICE_IDS structure that specifies the device ID(s) of the physical adapters. Supported starting with Windows 10 (WDDM 2.0).
+	KMTQAITYPE_PHYSICALADAPTERDEVICEIDS = 31
 )

 var ErrNoGPUDevices = errors.New("no GPU devices found")

-func GetGPUDeviceByLUID(adapterLUID windows.LUID) (GPUDevice, error) {
-	open := D3DKMT_OPENADAPTERFROMLUID{
-		AdapterLUID: adapterLUID,
-	}
-
-	if err := D3DKMTOpenAdapterFromLuid(&open); err != nil {
-		return GPUDevice{}, fmt.Errorf("D3DKMTOpenAdapterFromLuid failed: %w", err)
-	}
-
-	errs := make([]error, 0)
-
-	gpuDevice, err := GetGPUDevice(open.HAdapter)
-	if err != nil {
-		errs = append(errs, fmt.Errorf("GetGPUDevice failed: %w", err))
-	}
-
-	if err := D3DKMTCloseAdapter(&D3DKMT_CLOSEADAPTER{
-		HAdapter: open.HAdapter,
-	}); err != nil {
-		errs = append(errs, fmt.Errorf("D3DKMTCloseAdapter failed: %w", err))
-	}
-
-	if len(errs) > 0 {
-		return gpuDevice, fmt.Errorf("errors occurred while getting GPU device: %w", errors.Join(errs...))
-	}
-
-	gpuDevice.LUID = adapterLUID
-
-	return gpuDevice, nil
-}
-
 func GetGPUDevice(hAdapter D3DKMT_HANDLE) (GPUDevice, error) {
 	var gpuDevice GPUDevice

@@ -118,6 +89,18 @@ func GetGPUDevice(hAdapter D3DKMT_HANDLE) (GPUDevice, error) {

 	gpuDevice.AdapterString = windows.UTF16ToString(info.AdapterString[:])

+	var deviceIDs D3DKMT_QUERY_DEVICE_IDS
+
+	query.queryType = KMTQAITYPE_PHYSICALADAPTERDEVICEIDS
+	query.pPrivateDriverData = unsafe.Pointer(&deviceIDs)
+	query.privateDriverDataSize = uint32(unsafe.Sizeof(deviceIDs))
+
+	if err := D3DKMTQueryAdapterInfo(&query); err != nil && !errors.Is(err, windows.ERROR_FILE_NOT_FOUND) {
+		return gpuDevice, fmt.Errorf("D3DKMTQueryAdapterInfo (Device IDs) failed: %w", err)
+	}
+
+	gpuDevice.DeviceID = formatPNPDeviceID(deviceIDs)
+
 	return gpuDevice, nil
 }

@@ -151,7 +134,7 @@ func GetGPUDevices() ([]GPUDevice, error) {
 	// Process each adapter
 	for i := range enumAdapters.NumAdapters {
 		adapter := pAdapters[i]
-		// Validate handle before using it
+		// Validate the handle before using it.
 		if adapter.HAdapter == 0 {
 			errs = append(errs, fmt.Errorf("adapter %d has null handle", i))

@@ -190,3 +173,13 @@ func GetGPUDevices() ([]GPUDevice, error) {

 	return gpuDevices, nil
 }
+
+func formatPNPDeviceID(deviceIDs D3DKMT_QUERY_DEVICE_IDS) string {
+	return fmt.Sprintf("PCI\\VEN_%04X&DEV_%04X&SUBSYS_%04X%04X&REV_%02X",
+		uint16(deviceIDs.DeviceIds.VendorID),
+		uint16(deviceIDs.DeviceIds.DeviceID),
+		uint16(deviceIDs.DeviceIds.SubSystemID),
+		uint16(deviceIDs.DeviceIds.SubVendorID),
+		uint8(deviceIDs.DeviceIds.RevisionID),
+	)
+}
--- a/internal/headers/gdi32/types.go
+++ b/internal/headers/gdi32/types.go
@@ -73,9 +73,22 @@ type D3DKMT_ADAPTERADDRESS struct {
 	FunctionNumber win32.UINT
 }

+type D3DKMT_QUERY_DEVICE_IDS struct {
+	PhysicalAdapterIndex win32.UINT
+	DeviceIds            struct {
+		VendorID    win32.UINT
+		DeviceID    win32.UINT
+		SubVendorID win32.UINT
+		SubSystemID win32.UINT
+		RevisionID  win32.UINT
+		BusType     win32.UINT
+	}
+}
+
 type GPUDevice struct {
 	AdapterString             string
 	LUID                      windows.LUID
+	DeviceID                  string
 	DedicatedVideoMemorySize  uint64
 	DedicatedSystemMemorySize uint64
 	SharedSystemMemorySize    uint64
--- a/internal/headers/win32/types.go
+++ b/internal/headers/win32/types.go
@@ -18,6 +18,7 @@
 package win32

 import (
+	"strconv"
 	"unsafe"

 	"golang.org/x/sys/windows"
@@ -32,8 +33,8 @@ type (
 	LPWSTR    struct {
 		*uint16
 	}
-	ULONG = uint32 // ULONG is a 32-bit unsigned int in Win32
-	UINT  = uint32 // UINT is a 32-bit unsigned int in Win32
+	ULONG uint32 // ULONG is a 32-bit unsigned int in Win32
+	UINT  uint32 // UINT is a 32-bit unsigned int in Win32
 )

 // NewLPWSTR creates a new LPWSTR from a string.
@@ -60,3 +61,7 @@ func (s *LPWSTR) Pointer() uintptr {
 func (s *LPWSTR) String() string {
 	return windows.UTF16PtrToString(s.uint16)
 }
+
+func (u *UINT) String() string {
+	return strconv.FormatUint(uint64(*u), 10)
+}
--- a/internal/headers/win32/utils.go
+++ b/internal/headers/win32/utils.go
@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: Apache-2.0
+//
+// Copyright The Prometheus Authors
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package win32
+
+// ParseMultiSz splits a UTF-16 encoded MULTI_SZ buffer (Windows style) into
+// individual UTF-16 string slices.
+//
+// A MULTI_SZ buffer is a sequence of UTF-16 strings separated by single null
+// terminators (0x0000) and terminated by an extra null (i.e., two consecutive
+// nulls) to mark the end of the list.
+//
+// Example layout in memory (UTF-16):
+//
+//	"foo\0bar\0baz\0\0"
+//
+// Given such a []uint16, this function returns a [][]uint16 where each inner
+// slice is one null-terminated string segment without the trailing null.
+//
+// The returned slices reference the original buffer (no copying).
+func ParseMultiSz(buf []uint16) [][]uint16 {
+	var (
+		result [][]uint16
+		start  int
+	)
+
+	for i := range buf {
+		if buf[i] == 0 {
+			// Found a null terminator.
+			if i == start {
+				// Two consecutive nulls → end of list.
+				break
+			}
+
+			// Append current string slice (excluding null).
+			result = append(result, buf[start:i])
+			// Move start to next character after null.
+			start = i + 1
+		}
+	}
+
+	return result
+}