nvidia_smi utilization_memory not correct

[Qedr1]

Relevant telegraf.conf

[[inputs.nvidia_smi]]
bin_path = "/usr/bin/nvidia-smi"
timeout = "5s"
fieldinclude=["utilization_gpu","memory_total","memory_used","utilization_memory"]
tagexclude=["pstate","uuid","compute_mode"]

Logs from Telegraf

"fields": {               
    "gpu_used_percent": 0,  
    "utilization_memory": 0,  
    "index": 0,             
    "memory_total": 11178,  
    "memory_used": 3843,    
}

System info

Telegraf 1.33, Nvidia 1080Ti

Docker

No response

Steps to reproduce

1. Add to telegraf.conf
   [[inputs.nvidia_smi]]
   bin_path = "/usr/bin/nvidia-smi"
   timeout = "5s"
   fieldinclude=["utilization_gpu","memory_total","memory_used","utilization_memory"]
   tagexclude=["pstate","uuid","compute_mode"]

Expected behavior

"utilization_memory": 34.38
// utilization_memory= memory_used/memory_total*100= 3843/11178 *100

Actual behavior

"utilization_memory": 0

Additional info

No response

[skartikey]

The discrepancy between utilization_memory (0%) and your calculation (34.38%) is due to these metrics measuring different things:

• utilization_memory represents the percentage of time the memory controller is busy (time-based metric), not the percentage of memory capacity used
• NVIDIA's NVML documentation defines it as "the percentage of time during which global (device) memory was being read from or written to"

This explains why you can have significant memory allocated (3843/11178 MiB) but 0% memory utilization if there are no active read/write operations at the moment of measurement.

If you want to track the percentage of memory capacity used, you can add a processor plugin to calculate it:

[[processors.converter]]
  [processors.converter.fields]
    measurement = "*"
    [[processors.converter.fields.calculated]]
      new_field = "memory_used_percent" 
      formula = "(memory_used / memory_total) * 100.0"

This will create a field that matches your expected calculation.

[skartikey]

@Qedr1, We provided a response but haven't heard back. Closing this issue due to inactivity.
Feel free to reopen if you're still experiencing this problem and can provide additional details.
Thanks!