From 65d09d89fa951e1855f8bdb285a7a8389abc84c1 Mon Sep 17 00:00:00 2001
From: Sven Rebhan <36194019+srebhan@users.noreply.github.com>
Date: Wed, 5 Mar 2025 16:23:40 +0100
Subject: [PATCH] feat(inputs.nvidia_smi): Add new power-draw fields for v12
 scheme (#16579)

---
 plugins/inputs/nvidia_smi/nvidia_smi_test.go  |  50 +++
 .../inputs/nvidia_smi/schema_v12/parser.go    |   3 +
 plugins/inputs/nvidia_smi/schema_v12/types.go |   6 +
 .../nvidia_smi/testdata/rtx-3060-v12.xml      | 319 ++++++++++++++++++
 4 files changed, 378 insertions(+)
 create mode 100644 plugins/inputs/nvidia_smi/testdata/rtx-3060-v12.xml

diff --git a/plugins/inputs/nvidia_smi/nvidia_smi_test.go b/plugins/inputs/nvidia_smi/nvidia_smi_test.go
index 94510c3e7..1260bc347 100644
--- a/plugins/inputs/nvidia_smi/nvidia_smi_test.go
+++ b/plugins/inputs/nvidia_smi/nvidia_smi_test.go
@@ -426,6 +426,56 @@ func TestGatherValidXML(t *testing.T) {
 					time.Unix(0, 0)),
 			},
 		},
+		{
+			name:     "RTC 3060 schema v12",
+			filename: "rtx-3060-v12.xml",
+			expected: []telegraf.Metric{
+				testutil.MustMetric(
+					"nvidia_smi",
+					map[string]string{
+						"compute_mode": "Default",
+						"index":        "0",
+						"name":         "NVIDIA GeForce RTX 3060",
+						"arch":         "Ampere",
+						"pstate":       "P8",
+						"uuid":         "GPU-d6889ff6-2523-9142-ca3c-1ca3f396a625",
+					},
+					map[string]interface{}{
+						"clocks_current_graphics":       210,
+						"clocks_current_memory":         405,
+						"clocks_current_sm":             210,
+						"clocks_current_video":          555,
+						"cuda_version":                  "12.8",
+						"display_active":                "Disabled",
+						"display_mode":                  "Disabled",
+						"driver_version":                "570.124.04",
+						"encoder_stats_average_fps":     0,
+						"encoder_stats_average_latency": 0,
+						"encoder_stats_session_count":   0,
+						"fbc_stats_average_fps":         0,
+						"fbc_stats_average_latency":     0,
+						"fbc_stats_session_count":       0,
+						"fan_speed":                     0,
+						"power_draw":                    11.63,
+						"memory_free":                   11806,
+						"memory_total":                  12288,
+						"memory_used":                   116,
+						"memory_reserved":               368,
+						"pcie_link_gen_current":         1,
+						"pcie_link_width_current":       16,
+						"temperature_gpu":               42,
+						"utilization_gpu":               0,
+						"utilization_jpeg":              0,
+						"utilization_memory":            0,
+						"utilization_encoder":           0,
+						"utilization_decoder":           0,
+						"utilization_ofa":               0,
+						"vbios_version":                 "94.04.71.00.69",
+					},
+					time.Unix(1689872450, 0),
+				),
+			},
+		},
 		{
 			name:     "RTC 3080 schema v12",
 			filename: "rtx-3080-v12.xml",
diff --git a/plugins/inputs/nvidia_smi/schema_v12/parser.go b/plugins/inputs/nvidia_smi/schema_v12/parser.go
index bafd88401..dbf8745ba 100644
--- a/plugins/inputs/nvidia_smi/schema_v12/parser.go
+++ b/plugins/inputs/nvidia_smi/schema_v12/parser.go
@@ -77,10 +77,13 @@ func Parse(acc telegraf.Accumulator, buf []byte) error {
 		common.SetIfUsed("int", fields, "clocks_current_memory", gpu.Clocks.MemClock)
 		common.SetIfUsed("int", fields, "clocks_current_video", gpu.Clocks.VideoClock)
 		common.SetIfUsed("float", fields, "power_draw", gpu.PowerReadings.PowerDraw)
+		common.SetIfUsed("float", fields, "power_draw", gpu.PowerReadings.InstantPowerDraw)
 		common.SetIfUsed("float", fields, "power_limit", gpu.PowerReadings.PowerLimit)
 		common.SetIfUsed("float", fields, "power_draw", gpu.GpuPowerReadings.PowerDraw)
+		common.SetIfUsed("float", fields, "power_draw", gpu.GpuPowerReadings.InstantPowerDraw)
 		common.SetIfUsed("float", fields, "power_limit", gpu.GpuPowerReadings.PowerLimit)
 		common.SetIfUsed("float", fields, "module_power_draw", gpu.ModulePowerReadings.PowerDraw)
+		common.SetIfUsed("float", fields, "module_power_draw", gpu.ModulePowerReadings.InstantPowerDraw)
 		acc.AddFields("nvidia_smi", fields, tags, timestamp)
 
 		for _, device := range gpu.MigDevices.MigDevice {
diff --git a/plugins/inputs/nvidia_smi/schema_v12/types.go b/plugins/inputs/nvidia_smi/schema_v12/types.go
index a65c7c8a1..40c666fc0 100644
--- a/plugins/inputs/nvidia_smi/schema_v12/types.go
+++ b/plugins/inputs/nvidia_smi/schema_v12/types.go
@@ -114,6 +114,8 @@ type smi struct {
 			MaxPowerLimit       string `xml:"max_power_limit"`
 			MinPowerLimit       string `xml:"min_power_limit"`
 			PowerDraw           string `xml:"power_draw"`
+			AveragePowerDraw    string `xml:"average_power_draw"`
+			InstantPowerDraw    string `xml:"instant_power_draw"`
 			PowerLimit          string `xml:"power_limit"`
 			PowerState          string `xml:"power_state"`
 			RequestedPowerLimit string `xml:"requested_power_limit"`
@@ -180,6 +182,8 @@ type smi struct {
 			MaxPowerLimit       string `xml:"max_power_limit"`
 			MinPowerLimit       string `xml:"min_power_limit"`
 			PowerDraw           string `xml:"power_draw"`
+			AveragePowerDraw    string `xml:"average_power_draw"`
+			InstantPowerDraw    string `xml:"instant_power_draw"`
 			PowerState          string `xml:"power_state"`
 			RequestedPowerLimit string `xml:"requested_power_limit"`
 		} `xml:"module_power_readings"`
@@ -221,6 +225,8 @@ type smi struct {
 			PowerState         string `xml:"power_state"`
 			PowerManagement    string `xml:"power_management"`
 			PowerDraw          string `xml:"power_draw"`
+			AveragePowerDraw   string `xml:"average_power_draw"`
+			InstantPowerDraw   string `xml:"instant_power_draw"`
 			PowerLimit         string `xml:"power_limit"`
 			DefaultPowerLimit  string `xml:"default_power_limit"`
 			EnforcedPowerLimit string `xml:"enforced_power_limit"`
diff --git a/plugins/inputs/nvidia_smi/testdata/rtx-3060-v12.xml b/plugins/inputs/nvidia_smi/testdata/rtx-3060-v12.xml
new file mode 100644
index 000000000..fe442f419
--- /dev/null
+++ b/plugins/inputs/nvidia_smi/testdata/rtx-3060-v12.xml
@@ -0,0 +1,319 @@
+<?xml version="1.0"?>
+<!DOCTYPE nvidia_smi_log SYSTEM "nvsmi_device_v12.dtd">
+<nvidia_smi_log>
+	<timestamp>Sun Mar 2 15:02:02 2025</timestamp>
+	<driver_version>570.124.04</driver_version>
+	<cuda_version>12.8</cuda_version>
+	<attached_gpus>1</attached_gpus>
+	<gpu id="00000000:2D:00.0">
+		<product_name>NVIDIA GeForce RTX 3060</product_name>
+		<product_brand>GeForce</product_brand>
+		<product_architecture>Ampere</product_architecture>
+		<display_mode>Disabled</display_mode>
+		<display_active>Disabled</display_active>
+		<persistence_mode>Enabled</persistence_mode>
+		<addressing_mode>None</addressing_mode>
+		<mig_mode>
+			<current_mig>N/A</current_mig>
+			<pending_mig>N/A</pending_mig>
+		</mig_mode>
+		<mig_devices>
+			None
+		</mig_devices>
+		<accounting_mode>Disabled</accounting_mode>
+		<accounting_mode_buffer_size>4000</accounting_mode_buffer_size>
+		<driver_model>
+			<current_dm>N/A</current_dm>
+			<pending_dm>N/A</pending_dm>
+		</driver_model>
+		<serial>N/A</serial>
+		<uuid>GPU-d6889ff6-2523-9142-ca3c-1ca3f396a625</uuid>
+		<minor_number>0</minor_number>
+		<vbios_version>94.04.71.00.69</vbios_version>
+		<multigpu_board>No</multigpu_board>
+		<board_id>0x2d00</board_id>
+		<board_part_number>N/A</board_part_number>
+		<gpu_part_number>2487-150-A1</gpu_part_number>
+		<gpu_fru_part_number>N/A</gpu_fru_part_number>
+		<platformInfo>
+			<chassis_serial_number>N/A</chassis_serial_number>
+			<slot_number>N/A</slot_number>
+			<tray_index>N/A</tray_index>
+			<host_id>N/A</host_id>
+			<peer_type>N/A</peer_type>
+			<module_id>1</module_id>
+			<gpu_fabric_guid>N/A</gpu_fabric_guid>
+		</platformInfo>
+		<inforom_version>
+			<img_version>G001.0000.03.03</img_version>
+			<oem_object>2.0</oem_object>
+			<ecc_object>N/A</ecc_object>
+			<pwr_object>N/A</pwr_object>
+		</inforom_version>
+		<inforom_bbx_flush>
+			<latest_timestamp>N/A</latest_timestamp>
+			<latest_duration>N/A</latest_duration>
+		</inforom_bbx_flush>
+		<gpu_operation_mode>
+			<current_gom>N/A</current_gom>
+			<pending_gom>N/A</pending_gom>
+		</gpu_operation_mode>
+		<c2c_mode>N/A</c2c_mode>
+		<gpu_virtualization_mode>
+			<virtualization_mode>None</virtualization_mode>
+			<host_vgpu_mode>N/A</host_vgpu_mode>
+			<vgpu_heterogeneous_mode>N/A</vgpu_heterogeneous_mode>
+		</gpu_virtualization_mode>
+		<gpu_reset_status>
+			<reset_required>Requested functionality has been deprecated</reset_required>
+			<drain_and_reset_recommended>Requested functionality has been deprecated</drain_and_reset_recommended>
+		</gpu_reset_status>
+		<gpu_recovery_action>None</gpu_recovery_action>
+		<gsp_firmware_version>570.124.04</gsp_firmware_version>
+		<ibmnpu>
+			<relaxed_ordering_mode>N/A</relaxed_ordering_mode>
+		</ibmnpu>
+		<pci>
+			<pci_bus>2D</pci_bus>
+			<pci_device>00</pci_device>
+			<pci_domain>0000</pci_domain>
+			<pci_base_class>3</pci_base_class>
+			<pci_sub_class>0</pci_sub_class>
+			<pci_device_id>248710DE</pci_device_id>
+			<pci_bus_id>00000000:2D:00.0</pci_bus_id>
+			<pci_sub_system_id>36583842</pci_sub_system_id>
+			<pci_gpu_link_info>
+				<pcie_gen>
+					<max_link_gen>4</max_link_gen>
+					<current_link_gen>1</current_link_gen>
+					<device_current_link_gen>1</device_current_link_gen>
+					<max_device_link_gen>4</max_device_link_gen>
+					<max_host_link_gen>4</max_host_link_gen>
+				</pcie_gen>
+				<link_widths>
+					<max_link_width>16x</max_link_width>
+					<current_link_width>16x</current_link_width>
+				</link_widths>
+			</pci_gpu_link_info>
+			<pci_bridge_chip>
+				<bridge_chip_type>N/A</bridge_chip_type>
+				<bridge_chip_fw>N/A</bridge_chip_fw>
+			</pci_bridge_chip>
+			<replay_counter>0</replay_counter>
+			<replay_rollover_counter>0</replay_rollover_counter>
+			<tx_util>550 KB/s</tx_util>
+			<rx_util>500 KB/s</rx_util>
+			<atomic_caps_outbound>N/A</atomic_caps_outbound>
+			<atomic_caps_inbound>N/A</atomic_caps_inbound>
+		</pci>
+		<fan_speed>0 %</fan_speed>
+		<performance_state>P8</performance_state>
+		<clocks_event_reasons>
+			<clocks_event_reason_gpu_idle>Active</clocks_event_reason_gpu_idle>
+			<clocks_event_reason_applications_clocks_setting>Not Active</clocks_event_reason_applications_clocks_setting>
+			<clocks_event_reason_sw_power_cap>Not Active</clocks_event_reason_sw_power_cap>
+			<clocks_event_reason_hw_slowdown>Not Active</clocks_event_reason_hw_slowdown>
+			<clocks_event_reason_hw_thermal_slowdown>Not Active</clocks_event_reason_hw_thermal_slowdown>
+			<clocks_event_reason_hw_power_brake_slowdown>Not Active</clocks_event_reason_hw_power_brake_slowdown>
+			<clocks_event_reason_sync_boost>Not Active</clocks_event_reason_sync_boost>
+			<clocks_event_reason_sw_thermal_slowdown>Not Active</clocks_event_reason_sw_thermal_slowdown>
+			<clocks_event_reason_display_clocks_setting>Not Active</clocks_event_reason_display_clocks_setting>
+		</clocks_event_reasons>
+		<sparse_operation_mode>N/A</sparse_operation_mode>
+		<fb_memory_usage>
+			<total>12288 MiB</total>
+			<reserved>368 MiB</reserved>
+			<used>116 MiB</used>
+			<free>11806 MiB</free>
+		</fb_memory_usage>
+		<bar1_memory_usage>
+			<total>256 MiB</total>
+			<used>5 MiB</used>
+			<free>251 MiB</free>
+		</bar1_memory_usage>
+		<cc_protected_memory_usage>
+			<total>0 MiB</total>
+			<used>0 MiB</used>
+			<free>0 MiB</free>
+		</cc_protected_memory_usage>
+		<compute_mode>Default</compute_mode>
+		<utilization>
+			<gpu_util>0 %</gpu_util>
+			<memory_util>0 %</memory_util>
+			<encoder_util>0 %</encoder_util>
+			<decoder_util>0 %</decoder_util>
+			<jpeg_util>0 %</jpeg_util>
+			<ofa_util>0 %</ofa_util>
+		</utilization>
+		<encoder_stats>
+			<session_count>0</session_count>
+			<average_fps>0</average_fps>
+			<average_latency>0</average_latency>
+		</encoder_stats>
+		<fbc_stats>
+			<session_count>0</session_count>
+			<average_fps>0</average_fps>
+			<average_latency>0</average_latency>
+		</fbc_stats>
+		<dram_encryption_mode>
+			<current_dram_encryption>N/A</current_dram_encryption>
+			<pending_dram_encryption>N/A</pending_dram_encryption>
+		</dram_encryption_mode>
+		<ecc_mode>
+			<current_ecc>N/A</current_ecc>
+			<pending_ecc>N/A</pending_ecc>
+		</ecc_mode>
+		<ecc_errors>
+			<volatile>
+				<sram_correctable>N/A</sram_correctable>
+				<sram_uncorrectable_parity>N/A</sram_uncorrectable_parity>
+				<sram_uncorrectable_secded>N/A</sram_uncorrectable_secded>
+				<dram_correctable>N/A</dram_correctable>
+				<dram_uncorrectable>N/A</dram_uncorrectable>
+			</volatile>
+			<aggregate>
+				<sram_correctable>N/A</sram_correctable>
+				<sram_uncorrectable_parity>N/A</sram_uncorrectable_parity>
+				<sram_uncorrectable_secded>N/A</sram_uncorrectable_secded>
+				<dram_correctable>N/A</dram_correctable>
+				<dram_uncorrectable>N/A</dram_uncorrectable>
+				<sram_threshold_exceeded>N/A</sram_threshold_exceeded>
+			</aggregate>
+			<aggregate_uncorrectable_sram_sources>
+				<sram_l2>N/A</sram_l2>
+				<sram_sm>N/A</sram_sm>
+				<sram_microcontroller>N/A</sram_microcontroller>
+				<sram_pcie>N/A</sram_pcie>
+				<sram_other>N/A</sram_other>
+			</aggregate_uncorrectable_sram_sources>
+		</ecc_errors>
+		<retired_pages>
+			<multiple_single_bit_retirement>
+				<retired_count>N/A</retired_count>
+				<retired_pagelist>N/A</retired_pagelist>
+			</multiple_single_bit_retirement>
+			<double_bit_retirement>
+				<retired_count>N/A</retired_count>
+				<retired_pagelist>N/A</retired_pagelist>
+			</double_bit_retirement>
+			<pending_blacklist>N/A</pending_blacklist>
+			<pending_retirement>N/A</pending_retirement>
+		</retired_pages>
+		<remapped_rows>N/A</remapped_rows>
+		<temperature>
+			<gpu_temp>42 C</gpu_temp>
+			<gpu_temp_tlimit>N/A</gpu_temp_tlimit>
+			<gpu_temp_max_threshold>98 C</gpu_temp_max_threshold>
+			<gpu_temp_slow_threshold>95 C</gpu_temp_slow_threshold>
+			<gpu_temp_max_gpu_threshold>93 C</gpu_temp_max_gpu_threshold>
+			<gpu_target_temperature>83 C</gpu_target_temperature>
+			<memory_temp>N/A</memory_temp>
+			<gpu_temp_max_mem_threshold>N/A</gpu_temp_max_mem_threshold>
+		</temperature>
+		<supported_gpu_target_temp>
+			<gpu_target_temp_min>65 C</gpu_target_temp_min>
+			<gpu_target_temp_max>90 C</gpu_target_temp_max>
+		</supported_gpu_target_temp>
+		<gpu_power_readings>
+			<power_state>P8</power_state>
+			<average_power_draw>11.60 W</average_power_draw>
+			<instant_power_draw>11.63 W</instant_power_draw>
+			<current_power_limit>170.00 W</current_power_limit>
+			<requested_power_limit>170.00 W</requested_power_limit>
+			<default_power_limit>170.00 W</default_power_limit>
+			<min_power_limit>100.00 W</min_power_limit>
+			<max_power_limit>190.00 W</max_power_limit>
+		</gpu_power_readings>
+		<gpu_memory_power_readings>
+			<average_power_draw>N/A</average_power_draw>
+			<instant_power_draw>N/A</instant_power_draw>
+		</gpu_memory_power_readings>
+		<module_power_readings>
+			<power_state>P8</power_state>
+			<average_power_draw>N/A</average_power_draw>
+			<instant_power_draw>N/A</instant_power_draw>
+			<current_power_limit>N/A</current_power_limit>
+			<requested_power_limit>N/A</requested_power_limit>
+			<default_power_limit>N/A</default_power_limit>
+			<min_power_limit>N/A</min_power_limit>
+			<max_power_limit>N/A</max_power_limit>
+		</module_power_readings>
+		<power_smoothing>N/A</power_smoothing>
+		<power_profiles>
+			<power_profile_requested_profiles>N/A</power_profile_requested_profiles>
+			<power_profile_enforced_profiles>N/A</power_profile_enforced_profiles>
+		</power_profiles>
+		<clocks>
+			<graphics_clock>210 MHz</graphics_clock>
+			<sm_clock>210 MHz</sm_clock>
+			<mem_clock>405 MHz</mem_clock>
+			<video_clock>555 MHz</video_clock>
+		</clocks>
+		<applications_clocks>
+			<graphics_clock>N/A</graphics_clock>
+			<mem_clock>N/A</mem_clock>
+		</applications_clocks>
+		<default_applications_clocks>
+			<graphics_clock>N/A</graphics_clock>
+			<mem_clock>N/A</mem_clock>
+		</default_applications_clocks>
+		<deferred_clocks>
+			<mem_clock>N/A</mem_clock>
+		</deferred_clocks>
+		<max_clocks>
+			<graphics_clock>2145 MHz</graphics_clock>
+			<sm_clock>2145 MHz</sm_clock>
+			<mem_clock>7501 MHz</mem_clock>
+			<video_clock>1995 MHz</video_clock>
+		</max_clocks>
+		<max_customer_boost_clocks>
+			<graphics_clock>N/A</graphics_clock>
+		</max_customer_boost_clocks>
+		<clock_policy>
+			<auto_boost>N/A</auto_boost>
+			<auto_boost_default>N/A</auto_boost_default>
+		</clock_policy>
+		<voltage>
+			<graphics_volt>N/A</graphics_volt>
+		</voltage>
+		<fabric>
+			<state>N/A</state>
+			<status>N/A</status>
+			<cliqueId>N/A</cliqueId>
+			<clusterUuid>N/A</clusterUuid>
+			<health>
+				<bandwidth>N/A</bandwidth>
+				<route_recovery_in_progress>N/A</route_recovery_in_progress>
+				<route_unhealthy>N/A</route_unhealthy>
+				<access_timeout_recovery>N/A</access_timeout_recovery>
+			</health>
+		</fabric>
+		<supported_clocks>
+			<supported_mem_clock>
+				<value>7501 MHz</value>
+				<supported_graphics_clock>2145 MHz</supported_graphics_clock>
+
+			</supported_mem_clock>
+			<supported_mem_clock>
+				<value>7301 MHz</value>
+				<supported_graphics_clock>2145 MHz</supported_graphics_clock>
+			</supported_mem_clock>
+			<supported_mem_clock>
+				<value>5001 MHz</value>
+				<supported_graphics_clock>2145 MHz</supported_graphics_clock>
+			</supported_mem_clock>
+			<supported_mem_clock>
+				<value>810 MHz</value>
+				<supported_graphics_clock>2100 MHz</supported_graphics_clock>
+			</supported_mem_clock>
+		</supported_clocks>
+		<processes>
+		</processes>
+		<accounted_processes>
+		</accounted_processes>
+		<capabilities>
+			<egm>disabled</egm>
+		</capabilities>
+	</gpu>
+
+</nvidia_smi_log>
\ No newline at end of file