Hello - in two of our mature network security appliances, we’re using an MP2953A as a VR12.5 controller - one for a Haswell-EP CPU and one for a Broadwell-D SoC. In both cases, we’re monitoring IOUT & POUT via PMBus with our BMC, and the Intel CPU monitors IOUT via SVID. These products have been shipping since 2017 and 2019, respectively.
We were recently told by MPS that we needed to migrate to the MP2953B device, as the A-device is not being produced anymore.
In migrating to the MP2953B device, we are seeing that the PMBus-reported IOUT/POUT is 2~3x higher than the SVID-reported IOUT/POUT under the same conditions. This is with no other HW changes…just the MP2953 is changing from the A-rev to the B-rev device.
Can MPS explain why we would be seeing this difference? I have not been able to find anything in the PMBus-accessible register map that would account for this.
I know for more recent VR controllers, the IOUT reported via PMBus and SVID are in fact different and use different gain and offset values, programmable by the user. I can’t find any evidence of this for the older VR12.5 devices, although I’m sure they exist, just not documented?
EDIT: For clarification - the power consumed by the CPU is in fact the same in both cases, and matches the SVID-reported IOUT in both cases. The issue is only in the values read via PMBus for IOUT and POUT. The PMBus-reported values match the SVID-reported values for the MP2953A. But, the PMBus-reported values are 2~3x higher than the correct SVID-reported values using the MP2953B device. Through all of our testing and qualification, there are no other issues that we have identified.
Thanks,
Ed