Why Is the Number of Queried Graphics Cards Different from the Actual One?

Symptom

The number of graphics cards queried by running the nvidia-smi command is less than the actual number of graphics cards.

In this figure, the nvidia-smi command output indicates that there are seven graphics cards, but the ECS actually has eight graphics cards.

Checking the Number of Graphics Cards

Run the following command. If the number of the queried graphics cards is the same as the actual number of the graphics cards and the graphics card's status is normal (rev a1), go to Solution. If the graphics card cannot be found or the status is rev ff, refer to Fault Diagnosis and Handling of Graphics Cards. You can query the number of the graphics card defined by the flavors by referring to GPU-accelerated ECSs.

lspci | grep -i nvidia

Solution

• In non-CCE clusters, reinstall the driver or upgrade the driver and run the nvidia-smi command to check whether the fault persists. If the fault persists, refer to Fault Information Collection and contact technical support.
• In CCE clusters, collect fault information by referring to Fault Information Collection and contact technical support.