Gigabyte mz32-ar0 - Terrible performance if a PSU fails

[Harmony]

I have an issue with my mz32-ar0 (rev3) inside a Supermicro 826. Tested this with multiple servers now after noticing. If I pull the plug on one of my 2 PSU (so that it buzzes loudly) the system starts to perform very slowly. These are Supermicro PWS-920P-SQ 920W so there is plenty of extra power.

sysbench --test=memory --memory-block-size=4G --memory-total-size=32G run

1PSU
(1170.13 MiB/sec)
2 PSU
(19055.43 MiB/sec)

If I pull the turned off PSU out so the buzzing stops the system works fine.

Is there something I can adjust in the BIOS to fix this?

 

[oneplane]

Does it have PMBus integration, and if so, does it have any configuration for it? Sometimes it's a pretty dumb setup where it does talk to the PDB (and sometimes the PSU) but doesn't actually do anything useful with the data except put a cap on power usage when redundancy is lost.

 

[Harmony]

Does it have PMBus integration, and if so, does it have any configuration for it? Sometimes it's a pretty dumb setup where it does talk to the PDB (and sometimes the PSU) but doesn't actually do anything useful with the data except put a cap on power usage when redundancy is lost.

Removing the pmbus cable does fix this.

I don't know how I can adjust the configuration you're talking about.

 

[Harmony]

Does it have PMBus integration, and if so, does it have any configuration for it? Sometimes it's a pretty dumb setup where it does talk to the PDB (and sometimes the PSU) but doesn't actually do anything useful with the data except put a cap on power usage when redundancy is lost.

What do you need from me to check compatibility?

 

[RolloZ170]

you can set the PSU modes.
redundancy or wattage adding.
but the supermicro PSUs SMBus is not comaptible with gigabyte BMC (mostly not)
the lack of one PSU is interpreted as a critical event and PROSHOT# issued, the rsult is the system thottling

 

[Markess]

but the supermicro PSUs SMBus is not comaptible with gigabyte BMC (mostly not)
the lack of one PSU is interpreted as a critical event and PROSHOT# issued, the rsult is the system thottling

Good to know! I'm getting ready to put a Gigabyte motherboard in a CSE-826. Different model (MF51-ES0), but also with Aspeed AST2500 for the BMC. I will have to keep watch for this behavior.

 

[RolloZ170]

but also with Aspeed AST2500 for the BMC

it is not the AST2500 which is incompatible. GB and SMC run different Firmware on it.

check the GB servers for power supply and PDB part numbers.

 

[Markess]

it is not the AST2500 which is incompatible. GB and SMC run different Firmware on it.

check the GB servers for power supply and PDB part numbers.

I suppose I should more properly have said that even though they are different models, both Gigabyte motherboards use the same Gigabyte firmware for the BMC. I routinely run a single PSU in my 2U chassis, so will be interesting to see what the behavior is.

 

[RolloZ170]

correct.
if you have supermicro power-supply, you have supermicro Power Distribution Unit.
another SMBus device.

 

[Harmony]

you can set the PSU modes.
redundancy or wattage adding.
but the supermicro PSUs SMBus is not comaptible with gigabyte BMC (mostly not)
the lack of one PSU is interpreted as a critical event and PROSHOT# issued, the rsult is the system thottling

What if I do not plug the PMBus cable in at all? Is there any major downsides

 

[RolloZ170]

What if I do not plug the PMBus cable in at all? Is there any major downsides

no.
SMBus in your case. (supermicro management bus)

 

[oneplane]

Not plugging it in mostly just means you can't receive signals or measure anything, but if it's incompatible anyway there isn't much lost.

 

[Harmony]

Not plugging it in mostly just means you can't receive signals or measure anything, but if it's incompatible anyway there isn't much lost.

Ok thank you both will leave it unplugged. The PSU are not being detected either way in IPMI