Henrik Carlqvist

Thanks for not giving up and finally being able to verify that it works as expected! Even though it has not yet been tested with an older version of nvidia-smi the changes to the code are so trivial that I expect to release version 1.3 later today. Thanks again!

Interesting! And the snmp daemon has been restarted since nvgpu-snmp.so was installed as you rebooted. However, you wrote something about uininstalling and reinstalling net-snmp, could it be that the running snmpd is using modules from some other directory than /usr/x86_64-linux-gnu/snmp/dlmod ? In what way did the compilation of nvgpu-snmp interfer with your snmp installation? During compilation, netsnmp-config is used to figure out where to install nvgpu-snmp.so, maybe that directory changes if...

I now realize that during "make install", the "install" command is called with the switch "-s" when installing nvgpu-snmp.so which strips the binary from some symbols, making it smaller. However, there is still one way to check that the new version is installed: strings /usr/x86_64-linux-gnu/snmp/dlmod/nvgpu-snmp.so | grep power The above command should show the new fields average_power_draw and instant_power_draw, if not, the installed nvgpu-snmp.so is some older version. If those fields does not...

Thanks! With that line in snmpd.conf you should not have nggpu_agentxd running as a separate process. Instead, please check that your /usr/x86_64-linux-gnu/snmp/dlmod/nvgpu-snmp.so has a size that matches your newly compiled nvgpu-snmp.so Also, just to make sure, check that the timestamp of /usr/x86_64-linux-gnu/snmp/dlmod/nvgpu-snmp.so is the same or newer than your newly compiled file.

In lack of any nvidia-smi I made a quick hack to my code to read the XML data provided by you instead of calling nvidia-smi. I then got the following output: tuxedo:~> snmpwalk -c public -v2c localhost 1.3.6.1.4.1.2021.13.42.2 UCD-SNMP-MIB::ucdExperimental.42.2.1.1.0 = INTEGER: 0 UCD-SNMP-MIB::ucdExperimental.42.2.1.1.1 = INTEGER: 1 UCD-SNMP-MIB::ucdExperimental.42.2.1.2.0 = STRING: "NVIDIA GeForce RTX 2060" UCD-SNMP-MIB::ucdExperimental.42.2.1.2.1 = STRING: "NVIDIA GeForce RTX 4060 Ti" UCD-SNMP-MIB::ucdExperimental.42.2.1.3.0...

Thanks for your attempt! I was hoping that my changes would help and use average_power_draw and instant_power_draw for your version of nvidia-smi. Just to double-check that the right version is running: ps -ef | grep nvgpu_agentxd | grep -v grep (just to make sure that only one such process is running), and then: ls -alH /proc/`ps -ef | grep nvgpu_agentxd | grep -v grep | awk '{print $2}'`/exe And make sure that the size and timestamp matches with your newly compiled binary. If size and timestamp...

Fine! I have pushed my changes now, but not been able to test anymore than to see that the code still compiles. I have also fixed another bug report about code no longer compiling with newer versions of net-snmp.. Please let me know if this version works better for you.

Thanks a lot for the xml data! Unfortunately I am not sure where to find any matching xml data in the file. Was your card used rather heavily when you did produce the file? If so, maybe the field "instant_power_draw" is the one supposed to be used now instead. That field says 144.88 W for one of your cards and 156.89 W for the other card. However those values are far higher than the 12 W in your first example output from nvidia-smi and 10 W in the example snmpwalk output. It seems as if average_power_draw...