like dcgmi dmon -e 1002,1004,.... ,I want to include all field ids but, Got error while creating a Field Group: Bad parameter passed to function

[hyoonseo159357]

I want to include all field ids like dcgmi dmon -e 1002,1004..............

However, if I enter more than a certain number, the following error occurs:

Got error while creating a Field Group: Bad parameter passed to function

Is there a way to print all fields through dcgmi dmon ? No matter how much I search, I can't find it.

[nikkon-dev]

@hyoonseo159357,

As of today there are several limitations related to the number of fields you can collect simultaneously via a single connection (single dcgmi command):

1. There can be only 64 field groups with 128 fields in each group. A signle dcgmi command creates a single fields group. That means you will not be able to collect more than 128 fields due to this limitation alone.
2. The communication protocol limit a single message size and all watched fields come in a signle message. That usually does not lead to the error you observe and instead you do not get all fields for all GPUs you wanted.
3. Your GPU may not support some fields you request, and the logic is that either all requested fields are available or the whole batch fails.

You could try to use multiple dcgmi commands with different set of fields, or use DCGM API directly to overcome the limitations.

WBR, Nik