From dcf8b8558c99e0f2f89e2b46ca337b864fc80307 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Guillermo=20Juli=C3=A1n?= Date: Fri, 4 Sep 2026 11:32:26 +0200 Subject: [PATCH] Update GPU metric metadata --- gpu/metadata.csv | 131 ++++++++++++++++++++++++++++------------------- 1 file changed, 78 insertions(+), 53 deletions(-) diff --git a/gpu/metadata.csv b/gpu/metadata.csv index aa7a20f1038c3..27a39374f1bbb 100644 --- a/gpu/metadata.csv +++ b/gpu/metadata.csv @@ -1,7 +1,7 @@ metric_name,metric_type,interval,unit_name,per_unit_name,description,orientation,integration,short_name,curated_metric,sample_tags -gpu.c2c.errors.interrupt,gauge,16,,,C2C interrupt error counter total,0,gpu,c2c.errors.interrupt,, -gpu.c2c.errors.replay,gauge,16,,,C2C replay error counter total,0,gpu,c2c.errors.replay,, -gpu.c2c.errors.replay.b2b,gauge,16,,,C2C back-to-back replay error counter total,0,gpu,c2c.errors.replay.b2b,, +gpu.c2c.errors.interrupt,gauge,16,,,C2C interrupt error counter total. The counter reports the device lifetime total reported by NVML.,0,gpu,c2c.errors.interrupt,, +gpu.c2c.errors.replay,gauge,16,,,C2C replay error counter total. The counter reports the device lifetime total reported by NVML.,0,gpu,c2c.errors.replay,, +gpu.c2c.errors.replay.b2b,gauge,16,,,C2C back-to-back replay error counter total. The counter reports the device lifetime total reported by NVML.,0,gpu,c2c.errors.replay.b2b,, gpu.clock.speed.graphics,gauge,16,megahertz,,Current clock speed for the graphics domain,0,gpu,clock.speed.graphics,, gpu.clock.speed.graphics.max,gauge,16,megahertz,,Maximum clock speed for the graphics domain,0,gpu,clock.speed.graphics.max,, gpu.clock.speed.memory,gauge,16,megahertz,,Current clock speed for the memory,0,gpu,clock.speed.memory,, @@ -13,79 +13,104 @@ gpu.clock.speed.video.max,gauge,16,megahertz,,Maximum clock speed for the video gpu.clock.throttle_reasons.applications_clocks_setting,gauge,16,,,GPU clocks that are throttled due to application settings,0,gpu,clock.throttle_reasons.applications_clocks_setting,, gpu.clock.throttle_reasons.display_clock_setting,gauge,16,,,GPU clocks that are throttled due to display clock settings,0,gpu,clock.throttle_reasons.display_clock_setting,, gpu.clock.throttle_reasons.gpu_idle,gauge,16,,,GPU clocks that are throttled due to the GPU being idle,0,gpu,clock.throttle_reasons.gpu_idle,, +gpu.clock.throttle_reasons.hw_power_brake_slowdown,gauge,16,,,GPU clocks that are throttled due to an external power brake assertion,0,gpu,clock.throttle_reasons.hw_power_brake_slowdown,, +gpu.clock.throttle_reasons.hw_slowdown,gauge,16,,,GPU clocks that are throttled by hardware slowdown,0,gpu,clock.throttle_reasons.hw_slowdown,, +gpu.clock.throttle_reasons.hw_thermal_slowdown,gauge,16,,,GPU clocks that are throttled by hardware thermal slowdown,0,gpu,clock.throttle_reasons.hw_thermal_slowdown,, gpu.clock.throttle_reasons.none,gauge,16,,,GPU clocks that are not throttled,0,gpu,clock.throttle_reasons.none,, gpu.clock.throttle_reasons.sw_power_cap,gauge,16,,,GPU clocks that are throttled to avoid exceeding power limits,0,gpu,clock.throttle_reasons.sw_power_cap,, gpu.clock.throttle_reasons.sw_thermal_slowdown,gauge,16,,,GPU clocks that are throttled to avoid exceeding temperature limits,0,gpu,clock.throttle_reasons.sw_thermal_slowdown,, gpu.clock.throttle_reasons.sync_boost,gauge,16,,,GPU clocks that are throttled to match clock speed of another GPU in the current sync boost group,0,gpu,clock.throttle_reasons.sync_boost,, +gpu.clock.throttled_while_active,gauge,16,,,Whether the GPU clock is throttled for a reason other than the GPU being idle,0,gpu,clock.throttled_while_active,, gpu.core.limit,gauge,16,core,,Number of GPU cores that the process/container/device has available,0,gpu,core.limit,, gpu.decoder_active,gauge,16,percent,,Percentage of time the decoder was active,0,gpu,decoder_active,, gpu.decoder_utilization,gauge,16,percent,,"Percentage of time the decoder was active. Only emitted on Agent 7.76 and previous versions, replaced by `gpu.decoder_active` since 7.77",0,gpu,decoder_utilization,, -gpu.device.total,gauge,16,device,,"Number of GPU devices found on the host. Aggregate across hosts with sum; the metric is per-device without interpolation.",0,gpu,device.total,, -gpu.device.unhealthy,gauge,16,,,"Indicates whether the GPU device is reporting an unhealthy state (for example, XID errors, driver communication failure, or persistent throttling). Reported as 1 when unhealthy, 0 otherwise.",0,gpu,device.unhealthy,, +gpu.device.needs_recovery,gauge,16,,,"Whether the GPU requires a recovery action (e.g. reset or reboot) as reported by NVML's GPU recovery action field. 0 means no action is needed, 1 means a recovery action is required. The recovery_action tag indicates the specific action.",0,gpu,device.needs_recovery,, +gpu.device.total,gauge,16,device,,Number of GPU devices found in the host,0,gpu,device.total,, +gpu.device.unhealthy,gauge,16,,,Whether the GPU device is unhealthy,0,gpu,device.unhealthy,, gpu.dram_active,gauge,16,percent,,Percentage of time the DRAM was active,0,gpu,dram_active,, gpu.ecc.repair_pending.channel,gauge,16,,,Whether a channel repair is pending due to ECC errors,0,gpu,ecc.repair_pending.channel,, gpu.ecc.repair_pending.tpc,gauge,16,,,Whether a TPC repair is pending due to ECC errors,0,gpu,ecc.repair_pending.tpc,, gpu.encoder_active,gauge,16,percent,,Percentage of time the encoder was active,0,gpu,encoder_active,, gpu.encoder_utilization,gauge,16,percent,,"Percentage of time the encoder was active. Only emitted on Agent 7.76 and previous versions, replaced by `gpu.encoder_active` since 7.77",0,gpu,encoder_utilization,, -gpu.errors.ecc.corrected.total,gauge,16,,,"Corrected ECC (Error Correcting Code) errors indicate GPU memory corruption that was detected and successfully corrected by the hardware. While these errors do not immediately impact application correctness, recurring corrected errors can be an early sign of degrading memory or underlying hardware issues.",0,gpu,errors.ecc.corrected.total,, -gpu.errors.ecc.uncorrected.total,gauge,16,,,"Uncorrectable ECC (Error Correcting Code) errors indicate GPU memory corruption that could not be automatically fixed. These errors can lead to data corruption, application crashes, or incorrect computation results.",0,gpu,errors.ecc.uncorrected.total,, -gpu.errors.xid.total,gauge,16,error,,"XID errors are NVIDIA driver error codes that indicate GPU hardware or software issues, ranging from thermal throttling to hardware defects. Tagged by `origin` (driver/hardware) and `type` (XID number; see the NVIDIA XID catalog). Investigate immediately as XID errors often precede total GPU failures.",0,gpu,errors.xid.total,, +gpu.errors.ecc.corrected.total,gauge,16,,,"Corrected ECC (Error Correcting Code) errors indicate GPU memory corruption that was detected and successfully corrected by the hardware. While these errors do not immediately impact application correctness, recurring corrected errors can be an early sign of degrading memory or underlying hardware issues. The counter reports errors seen for the entire lifetime of the device.",0,gpu,errors.ecc.corrected.total,, +gpu.errors.ecc.corrected.volatile,gauge,16,,,Corrected ECC (Error Correcting Code) errors since the last driver load or device reset. Recurring corrected errors can be an early sign of degrading memory or underlying hardware issues.,0,gpu,errors.ecc.corrected.volatile,, +gpu.errors.ecc.sram.threshold_exceeded,gauge,16,,,Whether the SRAM ECC error threshold has been exceeded,0,gpu,errors.ecc.sram.threshold_exceeded,, +gpu.errors.ecc.sram.uncorrected_by_subtype.total,gauge,16,,,Aggregate SRAM uncorrectable ECC errors split by subtype. The counter reports errors seen for the entire lifetime of the device.,0,gpu,errors.ecc.sram.uncorrected_by_subtype.total,, +gpu.errors.ecc.sram.uncorrected_by_subtype.volatile,gauge,16,,,"SRAM uncorrectable ECC errors since the last driver load or device reset, split by subtype.",0,gpu,errors.ecc.sram.uncorrected_by_subtype.volatile,, +gpu.errors.ecc.uncorrected.total,gauge,16,,,"Uncorrectable ECC (Error Correcting Code) errors indicate GPU memory corruption that could not be automatically fixed. These errors can lead to data corruption, application crashes, or incorrect computation results. The counter reports errors seen for the entire lifetime of the device.",0,gpu,errors.ecc.uncorrected.total,, +gpu.errors.ecc.uncorrected.volatile,gauge,16,,,"Uncorrectable ECC (Error Correcting Code) errors since the last driver load or device reset. These errors can lead to data corruption, application crashes, or incorrect computation results.",0,gpu,errors.ecc.uncorrected.volatile,, +gpu.errors.xid,rate,16,,,XID errors received during the last collection interval. XID errors are NVIDIA driver error codes that indicate GPU hardware or software issues like thermal throttling to hardware defects. Investigate these errors immediately as they often precede total GPU failures.,0,gpu,errors.xid,, +gpu.errors.xid.total,gauge,16,error,,XID errors are NVIDIA driver error codes that indicate GPU hardware or software issues like thermal throttling to hardware defects. Investigate these errors immediately as they often precede total GPU failures. The counter reports XID errors received since the Agent started collecting events for this device.,0,gpu,errors.xid.total,, gpu.fan_speed,gauge,16,percent,,Configured fan speed as a percentage of its maximum,0,gpu,fan_speed,, gpu.fp16_active,gauge,16,percent,,Percentage of the time that the 16-bit floating point calculation engine was active. Only for Hopper and newer GPUs,0,gpu,fp16_active,, gpu.fp32_active,gauge,16,percent,,Percentage of the time that the 32-bit floating point calculation engine was active. Only for Hopper and newer GPUs,0,gpu,fp32_active,, gpu.fp64_active,gauge,16,percent,,Percentage of the time that the 64-bit floating point calculation engine was active. Only for Hopper and newer GPUs,0,gpu,fp64_active,, -gpu.gr_engine_active,gauge,16,percent,,"Percentage of time the graphics engine was active. Broader than `sm_active`: the graphics engine includes SM, texture, and raster work. Prefer `sm_active` for ML/compute workloads; use this metric for rendering workloads.",0,gpu,gr_engine_active,, +gpu.gr_engine_active,gauge,16,percent,,Percentage of time that the graphics engine was active,0,gpu,gr_engine_active,, gpu.integer_active,gauge,16,percent,,Percentage of the time that the integer calculation engine was active. Only for Hopper and newer GPUs,0,gpu,integer_active,, gpu.memory.bar1.free,gauge,16,byte,,Unallocated BAR1 memory (in bytes),0,gpu,memory.bar1.free,, gpu.memory.bar1.total,gauge,16,byte,,Total BAR1 memory (in bytes).,0,gpu,memory.bar1.total,, gpu.memory.bar1.used,gauge,16,byte,,Allocated used memory (in bytes),0,gpu,memory.bar1.used,, gpu.memory.free,gauge,16,byte,,Unallocated device memory (in bytes).,0,gpu,memory.free,, -gpu.memory.limit,gauge,16,byte,,"Total device memory (framebuffer). This is always the device-level memory limit; the `pid` and `container_id` tags are present to enable per-process and per-container utilization formulas, but the value itself does not change.",0,gpu,memory.limit,, -gpu.memory.reserved,gauge,16,byte,,Device memory (in bytes) reserved for system use (driver or firmware).,0,gpu,memory.reserved,, +gpu.memory.limit,gauge,16,byte,,The maximum amount of memory a process/container/device could allocate,0,gpu,memory.limit,, +gpu.memory.reserved,gauge,16,byte,,Device memory (in bytes) reserved for system use (driver or firmware)..,0,gpu,memory.reserved,, gpu.memory.temperature,gauge,16,degree celsius,,Temperature of the memory chip,0,gpu,memory.temperature,, +gpu.memory.utilization,gauge,16,fraction,,Ratio of used memory compared to total memory,0,gpu,memory.utilization,, gpu.nccl.collective.algo_bandwidth_gbps,gauge,16,gigabyte,second,Algorithmic bandwidth of a collective operation per rank,0,gpu,nccl.collective.algo_bandwidth_gbps,, gpu.nccl.collective.bus_bandwidth_gbps,gauge,16,gigabyte,second,Bus bandwidth of a collective operation per rank,0,gpu,nccl.collective.bus_bandwidth_gbps,, gpu.nccl.collective.exec_time_us,gauge,16,microsecond,,Execution time of a collective operation per rank,0,gpu,nccl.collective.exec_time_us,, gpu.nccl.collective.msg_size_bytes,gauge,16,byte,,Message size of a collective operation per rank,0,gpu,nccl.collective.msg_size_bytes,, gpu.nccl.rank.seconds_since_last_event,gauge,16,second,,Seconds since the last NCCL event was received for a rank. Used for hang detection.,0,gpu,nccl.rank.seconds_since_last_event,, -gpu.nvlink.ber.effective,gauge,16,,,NVLink effective error counter total for all links (errors not corrected by FEC/recovery mechanisms).,0,gpu,nvlink.ber.effective,, -gpu.nvlink.ber.symbol,gauge,16,,,Symbol bit error rate for all NVLINK links,0,gpu,nvlink.ber.symbol,, +gpu.nvlink.ber.effective,gauge,16,,,Per-link effective bit error rate for the NVLink port,0,gpu,nvlink.ber.effective,, +gpu.nvlink.ber.symbol,gauge,16,,,Per-link symbol bit error rate for the NVLink port,0,gpu,nvlink.ber.symbol,, gpu.nvlink.count.active,gauge,16,,,Number of active nvlinks for the device,0,gpu,nvlink.count.active,, gpu.nvlink.count.inactive,gauge,16,,,Number of inactive nvlinks for the device,0,gpu,nvlink.count.inactive,, gpu.nvlink.count.total,gauge,16,,,Number of total nvlinks for the device,0,gpu,nvlink.count.total,, -gpu.nvlink.errors.buffer.overrun,gauge,16,,,NVLink buffer overrun error counter total for all links,0,gpu,nvlink.errors.buffer.overrun,, -gpu.nvlink.errors.crc.data,gauge,16,,,NVLink data CRC error counter total for all links,0,gpu,nvlink.errors.crc.data,, -gpu.nvlink.errors.crc.flit,gauge,16,,,NVLink flow control CRC error counter total for all links,0,gpu,nvlink.errors.crc.flit,, -gpu.nvlink.errors.ecc,gauge,16,,,NVLink ECC error counter total for all links,0,gpu,nvlink.errors.ecc,, -gpu.nvlink.errors.effective,gauge,16,,,NVLink effective error counter total for all links,0,gpu,nvlink.errors.effective,, -gpu.nvlink.errors.local.link.integrity,gauge,16,,,NVLink local link integrity error counter total for all links,0,gpu,nvlink.errors.local.link.integrity,, -gpu.nvlink.errors.malformed.packet,gauge,16,,,NVLink malformed packet error counter total for all links,0,gpu,nvlink.errors.malformed.packet,, -gpu.nvlink.errors.recovery,gauge,16,,,NVLink recovery error counter total for all links,0,gpu,nvlink.errors.recovery,, -gpu.nvlink.errors.replay,gauge,16,,,NVLink replay error counter total for all links,0,gpu,nvlink.errors.replay,, -gpu.nvlink.errors.rx,gauge,16,,,NVLink receive error counter total for all links,0,gpu,nvlink.errors.rx,, -gpu.nvlink.errors.rx.general,gauge,16,,,NVLink general receive error counter total for all links,0,gpu,nvlink.errors.rx.general,, -gpu.nvlink.errors.rx.remote,gauge,16,,,NVLink remote receive error counter total for all links,0,gpu,nvlink.errors.rx.remote,, -gpu.nvlink.errors.symbol,gauge,16,,,NVLink symbol error counter total for all links,0,gpu,nvlink.errors.symbol,, -gpu.nvlink.nvswitch_connected,gauge,16,,,Number of NVLinks connected to the NVSwitch,0,gpu,nvlink.nvswitch_connected,, -gpu.nvlink.plr.codes_loss,gauge,16,,,NVLink PLR codes loss counter for the port,0,gpu,nvlink.plr.codes_loss,, -gpu.nvlink.plr.rx.code_err,gauge,16,,,NVLink PLR RX code error counter for the port,0,gpu,nvlink.plr.rx.code_err,, -gpu.nvlink.plr.rx.codes,gauge,16,,,NVLink PLR RX codes counter for the port,0,gpu,nvlink.plr.rx.codes,, -gpu.nvlink.plr.rx.uncorrectable_code,gauge,16,,,NVLink PLR RX uncorrectable code counter for the port,0,gpu,nvlink.plr.rx.uncorrectable_code,, -gpu.nvlink.plr.tx.codes,gauge,16,,,NVLink PLR TX codes counter for the port,0,gpu,nvlink.plr.tx.codes,, -gpu.nvlink.plr.tx.retry_codes,gauge,16,,,NVLink PLR TX retry codes counter for the port,0,gpu,nvlink.plr.tx.retry_codes,, -gpu.nvlink.plr.tx.retry_events,gauge,16,,,NVLink PLR TX retry events counter for the port,0,gpu,nvlink.plr.tx.retry_events,, -gpu.nvlink.plr.tx.retry_events_within_t_sec_max,gauge,16,,,Maximum NVLink PLR TX retry events within the configured time window for the port,0,gpu,nvlink.plr.tx.retry_events_within_t_sec_max,, -gpu.nvlink.plr.tx.sync_events,gauge,16,,,NVLink PLR TX sync events counter for the port,0,gpu,nvlink.plr.tx.sync_events,, -gpu.nvlink.recovery.events.failed,gauge,16,,,Failed NVLink recovery events total for all links,0,gpu,nvlink.recovery.events.failed,, -gpu.nvlink.recovery.events.successful,gauge,16,,,Successful NVLink recovery events total for all links,0,gpu,nvlink.recovery.events.successful,, -gpu.nvlink.rx.packets,gauge,16,,,NVLink received packet counter total for all links,0,gpu,nvlink.rx.packets,, -gpu.nvlink.speed,gauge,16,megabit,second,"Negotiated NVLink speed in megabits per second. On Agent 7.80 and previous versions, this is the common speed shared by all active NVLINK links on the device. Since Agent 7.81, this is reported per link and tagged by `nvlink_port`.",0,gpu,nvlink.speed,, -gpu.nvlink.throughput.data.rx,gauge,16,kilobyte,second,Total RX data throughput of all NVLINK links,0,gpu,nvlink.throughput.data.rx,, -gpu.nvlink.throughput.data.tx,gauge,16,kilobyte,second,Total TX data throughput of all NVLINK links,0,gpu,nvlink.throughput.data.tx,, -gpu.nvlink.throughput.raw.rx,gauge,16,kilobyte,second,Total RX of all NVLINK links,0,gpu,nvlink.throughput.raw.rx,, -gpu.nvlink.throughput.raw.tx,gauge,16,kilobyte,second,Total TX of all NVLINK links,0,gpu,nvlink.throughput.raw.tx,, -gpu.nvlink.tx.discards,gauge,16,,,NVLink transmitted discard counter total for all links,0,gpu,nvlink.tx.discards,, -gpu.nvlink.tx.packets,gauge,16,,,NVLink transmitted packet counter total for all links,0,gpu,nvlink.tx.packets,, +gpu.nvlink.errors.buffer.overrun,gauge,16,,,Per-link NVLink buffer overrun error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.buffer.overrun,, +gpu.nvlink.errors.crc.data,gauge,16,,,Per-link NVLink data CRC error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.crc.data,, +gpu.nvlink.errors.crc.flit,gauge,16,,,Per-link NVLink flow control CRC error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.crc.flit,, +gpu.nvlink.errors.ecc,gauge,16,,,Per-link NVLink ECC error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.ecc,, +gpu.nvlink.errors.effective,gauge,16,,,Per-link NVLink effective error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.effective,, +gpu.nvlink.errors.fec.heavy,gauge,16,,,Per-link rate of NVLink messages with heavy FEC errors for the NVLink port. The heavy threshold is configurable and defaults to 4 or more FEC errors.,0,gpu,nvlink.errors.fec.heavy,, +gpu.nvlink.errors.fec.light,gauge,16,,,Per-link rate of NVLink messages with light FEC errors for the NVLink port. The light threshold is configurable and defaults to 1 through 3 FEC errors.,0,gpu,nvlink.errors.fec.light,, +gpu.nvlink.errors.fec.none,gauge,16,,,Per-link rate of NVLink messages with no FEC errors for the NVLink port,0,gpu,nvlink.errors.fec.none,, +gpu.nvlink.errors.local.link.integrity,gauge,16,,,Per-link NVLink local link integrity error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.local.link.integrity,, +gpu.nvlink.errors.malformed.packet,gauge,16,,,Per-link NVLink malformed packet error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.malformed.packet,, +gpu.nvlink.errors.recovery,gauge,16,,,Per-link NVLink recovery error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.recovery,, +gpu.nvlink.errors.replay,gauge,16,,,Per-link NVLink replay error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.replay,, +gpu.nvlink.errors.rx,gauge,16,,,Per-link NVLink receive error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.rx,, +gpu.nvlink.errors.rx.general,gauge,16,,,Per-link NVLink general receive error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.rx.general,, +gpu.nvlink.errors.rx.remote,gauge,16,,,Per-link NVLink remote receive error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.rx.remote,, +gpu.nvlink.errors.symbol,gauge,16,,,Per-link NVLink symbol error counter for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.errors.symbol,, +gpu.nvlink.nvswitch_connected,gauge,16,,,NVSwitch connection count for the GPU.,0,gpu,nvlink.nvswitch_connected,, +gpu.nvlink.plr.codes_loss,gauge,16,,,Per-link NVLink PLR codes loss counter for the NVLink port,0,gpu,nvlink.plr.codes_loss,, +gpu.nvlink.plr.rx.code_err,gauge,16,,,Per-link NVLink PLR RX code error counter for the NVLink port,0,gpu,nvlink.plr.rx.code_err,, +gpu.nvlink.plr.rx.codes,gauge,16,,,Per-link NVLink PLR RX codes counter for the NVLink port,0,gpu,nvlink.plr.rx.codes,, +gpu.nvlink.plr.rx.uncorrectable_code,gauge,16,,,Per-link NVLink PLR RX uncorrectable code counter for the NVLink port,0,gpu,nvlink.plr.rx.uncorrectable_code,, +gpu.nvlink.plr.tx.codes,gauge,16,,,Per-link NVLink PLR TX codes counter for the NVLink port,0,gpu,nvlink.plr.tx.codes,, +gpu.nvlink.plr.tx.retry_codes,gauge,16,,,Per-link NVLink PLR TX retry codes counter for the NVLink port,0,gpu,nvlink.plr.tx.retry_codes,, +gpu.nvlink.plr.tx.retry_events,gauge,16,,,Per-link NVLink PLR TX retry events counter for the NVLink port,0,gpu,nvlink.plr.tx.retry_events,, +gpu.nvlink.plr.tx.retry_events_within_t_sec_max,gauge,16,,,Per-link maximum NVLink PLR TX retry events within the configured time window for the NVLink port,0,gpu,nvlink.plr.tx.retry_events_within_t_sec_max,, +gpu.nvlink.plr.tx.sync_events,gauge,16,,,Per-link NVLink PLR TX sync events counter for the NVLink port,0,gpu,nvlink.plr.tx.sync_events,, +gpu.nvlink.recovery.events.failed,gauge,16,,,Per-link failed NVLink recovery events for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.recovery.events.failed,, +gpu.nvlink.recovery.events.successful,gauge,16,,,Per-link successful NVLink recovery events for the NVLink port. The counter reports the device lifetime total reported by NVML.,0,gpu,nvlink.recovery.events.successful,, +gpu.nvlink.rx.packets,gauge,16,,,Per-link NVLink received packet counter for the NVLink port,0,gpu,nvlink.rx.packets,, +gpu.nvlink.speed,gauge,16,megabit,second,Per-link speed of the NVLink port,0,gpu,nvlink.speed,, +gpu.nvlink.throughput.data.rx,gauge,16,kilobyte,second,Per-link RX data throughput for the NVLink port,0,gpu,nvlink.throughput.data.rx,, +gpu.nvlink.throughput.data.rx.total,gauge,16,kilobyte,second,Total RX data throughput of all NVLINK links,0,gpu,nvlink.throughput.data.rx.total,, +gpu.nvlink.throughput.data.tx,gauge,16,kilobyte,second,Per-link TX data throughput for the NVLink port,0,gpu,nvlink.throughput.data.tx,, +gpu.nvlink.throughput.data.tx.total,gauge,16,kilobyte,second,Total TX data throughput of all NVLINK links,0,gpu,nvlink.throughput.data.tx.total,, +gpu.nvlink.throughput.raw.rx,gauge,16,kilobyte,second,Per-link raw RX throughput for the NVLink port,0,gpu,nvlink.throughput.raw.rx,, +gpu.nvlink.throughput.raw.rx.total,gauge,16,kilobyte,second,Total RX of all NVLINK links,0,gpu,nvlink.throughput.raw.rx.total,, +gpu.nvlink.throughput.raw.tx,gauge,16,kilobyte,second,Per-link raw TX throughput for the NVLink port,0,gpu,nvlink.throughput.raw.tx,, +gpu.nvlink.throughput.raw.tx.total,gauge,16,kilobyte,second,Total TX of all NVLINK links,0,gpu,nvlink.throughput.raw.tx.total,, +gpu.nvlink.tx.discards,gauge,16,,,Per-link NVLink transmitted discard counter for the NVLink port,0,gpu,nvlink.tx.discards,, +gpu.nvlink.tx.packets,gauge,16,,,Per-link NVLink transmitted packet counter for the NVLink port,0,gpu,nvlink.tx.packets,, +gpu.pci.link.speed.current,gauge,16,byte,second,Current usable bandwidth for the PCI link in bytes per second,0,gpu,pci.link.speed.current,, +gpu.pci.link.speed.degraded,gauge,16,,,Whether the current PCI link speed is lower than the max PCI link speed,0,gpu,pci.link.speed.degraded,, +gpu.pci.link.speed.max,gauge,16,byte,second,Max usable bandwidth for the PCI link in bytes per second,0,gpu,pci.link.speed.max,, +gpu.pci.link.width.current,gauge,16,,,Current lane width for the PCI link,0,gpu,pci.link.width.current,, +gpu.pci.link.width.degraded,gauge,16,,,Whether the current PCI link width is lower than the max PCI link width,0,gpu,pci.link.width.degraded,, +gpu.pci.link.width.max,gauge,16,,,Max lane width for the PCI link,0,gpu,pci.link.width.max,, gpu.pci.replay_counter,gauge,16,,,PCI replay counter,0,gpu,pci.replay_counter,, gpu.pci.throughput.rx,gauge,16,byte,second,Bytes received through PCI to the GPU device per second,0,gpu,pci.throughput.rx,, gpu.pci.throughput.tx,gauge,16,byte,second,Bytes transmitted through PCI from the GPU device per second,0,gpu,pci.throughput.tx,, @@ -100,14 +125,14 @@ gpu.process.encoder_active,gauge,16,percent,,Percentage of time the encoder was gpu.process.encoder_utilization,gauge,16,percent,,"Percentage of time the encoder was active for a specific process. Only emitted on Agent 7.76 and previous versions, replaced by `gpu.process.encoder_active` since 7.77",0,gpu,process.encoder_utilization,, gpu.process.memory.usage,gauge,16,byte,,The memory used by this process at the point the metric was given. Only emitted when processes are active.,0,gpu,process.memory.usage,, gpu.process.sm_active,gauge,16,percent,,Percentage of time the streaming multiprocessor was active for a specific process,0,gpu,process.sm_active,, -gpu.remapped_rows.correctable,gauge,16,,,Number of rows remapped due to correctable errors,0,gpu,remapped_rows.correctable,, -gpu.remapped_rows.failed,gauge,16,,,Number of rows that failed remapping,0,gpu,remapped_rows.failed,, -gpu.remapped_rows.pending,gauge,16,,,Number of rows pending remapping,0,gpu,remapped_rows.pending,, -gpu.remapped_rows.uncorrectable,gauge,16,,,Number of rows remapped due to uncorrectable errors,0,gpu,remapped_rows.uncorrectable,, +gpu.remapped_rows.correctable,gauge,16,,,Number of rows remapped due to correctable errors. The counter reports the device lifetime total reported by NVML.,0,gpu,remapped_rows.correctable,, +gpu.remapped_rows.failed,gauge,16,,,Whether row remapping has failed. This value reflects the current remapping status reported by NVML.,0,gpu,remapped_rows.failed,, +gpu.remapped_rows.pending,gauge,16,,,Whether row remapping is pending. This value reflects the current remapping status reported by NVML.,0,gpu,remapped_rows.pending,, +gpu.remapped_rows.uncorrectable,gauge,16,,,Number of rows remapped due to uncorrectable errors. The counter reports the device lifetime total reported by NVML.,0,gpu,remapped_rows.uncorrectable,, gpu.slowdown_temperature,gauge,16,degree celsius,,Slowdown temperature,0,gpu,slowdown_temperature,, -gpu.sm_active,gauge,16,percent,,"Percentage of time at least one warp was resident on a Streaming Multiprocessor (temporal measure). Distinct from `sm_utilization` (fraction of SMs with resident warps, spatial) and `sm_occupancy` (warp-slot fill ratio). Source: NVML SM_UTIL / GetProcessUtilization.",0,gpu,sm_active,, -gpu.sm_occupancy,gauge,16,percent,,Percentage of the warps available in the Streaming Multiprocessors that were active in the interval,0,gpu,sm_occupancy,, -gpu.sm_utilization,gauge,16,percent,,"Percentage of Streaming Multiprocessors with at least one resident warp during the interval (spatial measure). Distinct from `sm_active` (fraction of time the SM is active, temporal) and `sm_occupancy` (warp-slot fill ratio).",0,gpu,sm_utilization,, +gpu.sm_active,gauge,16,percent,,Percentage of time the streaming multiprocessor was active,0,gpu,sm_active,, +gpu.sm_occupancy,gauge,16,percent,,Percentage of the Streaming Multiprocessors that were active in the interval,0,gpu,sm_occupancy,, +gpu.sm_utilization,gauge,16,percent,,Percentage of the Streaming Multiprocessors that were in use in the interval,0,gpu,sm_utilization,, gpu.temperature,gauge,16,degree celsius,,Temperature of the GPU device,0,gpu,temperature,, gpu.tensor_active,gauge,16,percent,,Percentage of the time that the tensor calculation engine was active. Only for Hopper and newer GPUs,0,gpu,tensor_active,, gpu.total_energy_consumption,gauge,16,millijoule,,Total energy consumed by the device since the driver was reloaded,0,gpu,total_energy_consumption,,