Brian,
I normally tell customers that has such issues to use Universal Device Poller (UnDP) to isolate the issue.
First, check if the In or Out bandwidth is showing within the values you expect. The one that is showing up issues should have the OID monitored using UnDP.
Create the UnDP under Interfaces (not Nodes), so that it will mirror the Interface's Statistics Polling Interval.
By default, Nodes is polled for statistics at 10mins, whereas Interfaces are at 9mins
** Note that IfInOctets/IfOutOctets are COUNTER type OIDs. You will have to expand the advanced options of UnDP and ensure that it is COUNTER.
** If you are using 64-bit counters, use IfHCInOctets/IfHCOutOctets instead.
You should be able to see the values that are returned from the OIDs before going through transformation (How To Calculate Bandwidth Utilization Using SNMP [IP Application Services] - Cisco Systems)
But you do not normally need to do it. The pattern should be similar to the output of the charts in NPM.
If they are, we are reporting correctly, and the outputs of the OID might have issues that need to have the hardware vendor notified.
HTH.
Cheers!
Andy