GPUM: changes and additions in gpu metrics (#21148)

* added new gpu memory metrics

* updated memory.usage and core.usage metrics names

Author: val06

gpu/CHANGELOG.md

@@ -1,5 +1,12 @@
 # CHANGELOG - GPU
 
+## 0.4.1
+
+***Added***:
+
+* Added GPU device level memory metrics: `gpu.memory.free`, `gpu.memory.reserved`.
+* Renamed gpu.core.usage to gpu.process.core.usage for naming consistency.
+* Renamed gpu.memory.usage to gpu.process.memory.usage for naming consistency.
 
 ## 0.4.0
 

gpu/metadata.csv

@@ -15,7 +15,6 @@ gpu.clock.throttle_reasons.sw_power_cap,gauge,,,,GPU clocks that are throttled t
 gpu.clock.throttle_reasons.sw_thermal_slowdown,gauge,,,,GPU clocks that are throttled to avoid exceeding temperaturelimits,0,gpu,clock.throttle_reasons.sw_thermal_slowdown,,
 gpu.clock.throttle_reasons.sync_boost,gauge,,,,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.core.limit,gauge,,core,,Number of GPU cores that the process/container/device has available,0,gpu,core.limit,,
-gpu.core.usage,gauge,,core,,Average number of GPU cores that a process was using in the interval. Only emitted when processes are active.,0,gpu,core.usage,,
 gpu.decoder_utilization,gauge,,percent,,Percentage of time the decoder was active,0,gpu,decoder_utilization,,
 gpu.device.total,gauge,,,,Number of devices active in the host,0,gpu,device.total,,
 gpu.dram_active,gauge,,percent,,Percentage of time the DRAM was active,0,gpu,dram_active,,
@@ -29,9 +28,10 @@ gpu.integer_active,gauge,,percent,,Percentage of the time that the integer calcu
 gpu.memory.bar1.free,gauge,,byte,,Unallocated BAR1 memory (in bytes),0,gpu,memory.bar1.free,,
 gpu.memory.bar1.total,gauge,,byte,,Total BAR1 memory (in bytes).,0,gpu,memory.bar1.total,,
 gpu.memory.bar1.used,gauge,,byte,,Allocated used memory (in bytes),0,gpu,memory.bar1.used,,
+gpu.memory.free,gauge,,byte,,Unallocated device memory (in bytes).,0,gpu,memory.free,,
 gpu.memory.limit,gauge,,byte,,The maximum amount of memory a process/container/device could allocate,0,gpu,memory.limit,,
+gpu.memory.reserved,gauge,,byte,,Device memory (in bytes) reserved for system use (driver or firmware)..,0,gpu,memory.reserved,,
 gpu.memory.temperature,gauge,,degree celsius,,Temperature of the memory chip,0,gpu,memory.temperature,,
-gpu.memory.usage,gauge,,byte,,The memory used by this process at the point the metric was given. Only emitted when processes are active.,0,gpu,memory.usage,,
 gpu.nvlink.count.active,gauge,,,,Number of active nvlinks for the device,0,gpu,,,
 gpu.nvlink.count.inactive,gauge,,,,Number of inactive nvlinks for the device,0,gpu,,,
 gpu.nvlink.count.total,gauge,,,,Number of total nvlinks for the device,0,gpu,,,
@@ -52,9 +52,11 @@ gpu.pci.throughput.tx,gauge,,byte,second,Bytes transmitted through PCI to the GP
 gpu.performance_state,gauge,,,,Returns the current performance state of the device,0,gpu,performance_state,,
 gpu.power.management_limit,gauge,,milliwatt,,Upper boundary for the device power draw.,0,gpu,power.management_limit,,
 gpu.power.usage,gauge,,milliwatt,,"Power usage for the GPU device. On GA100 and older architectures this is the instantaneous power at that moment, in newer ones it represents the average power draw over one second",0,gpu,power.usage,,
+gpu.process.core.usage,gauge,,core,,Average number of GPU cores that a process was using in the interval. Only emitted when processes are active.,0,gpu,process.core.usage,,
 gpu.process.decoder_utilization,gauge,,percent,,Percentage of time the decoder was active for a specific process,0,gpu,process.decoder_utilization,,
 gpu.process.dram_active,gauge,,percent,,Percentage of time the DRAM was active for a specific process,0,gpu,process.dram_active,,
 gpu.process.encoder_utilization,gauge,,percent,,Percentage of time the encoder was active for a specific process,0,gpu,process.encoder_utilization,,
+gpu.process.memory.usage,gauge,,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,,percent,,Percentage of time the streaming multiprocessor was active for a specific process,0,gpu,process.sm_active,,
 gpu.remapped_rows.correctable,count,,,,Number of rows remapped due to correctable errors,0,gpu,remapped_rows.correctable,,
 gpu.remapped_rows.failed,count,,,,Number of rows that failed remapping,0,gpu,remapped_rows.failed,,