Supermicro h11dsi rev 2.0, after 8h the system freezes.

[Billyherington]

0.00612429

 

[pricklypunter]

Sounds like a cooling issue to me. Is this in an SM case with fans and shrouds etc?

 

[RolloZ170]

When I installed 14 pcs of RAM out of 16 pcs, then the system could work for more than 8 hours.
If install 16 pieces of RAM from 16 pieces, the system freezes after 8 hours.

ECC error count threshold

 

[RolloZ170]

Is there any way to disable the ecc function?

your memory is bad or your have bad contacts(CPU/socket/RAMslots), just solve.

 

[RolloZ170]

If that's the case, then why does the system work for 8 hours and not less? After all, if there is a problem with the contact, then the problem should appear faster and not after 8 hours of active work.

ECC (corrected) error thresholds is only idea i have if you just add (ecc error issueing) RAM.

After all, if there is a problem with the contact, then the problem should appear faster and not after 8 hours of active work.

no i mean the corrected EC errors. threshold.
search in BIOS for this setting, maybe you can increase count or disable and check for change of 8 hour time.

 

[sko]

Both processors are overclocked

oh dear...

the case cover is removed.

so no controlled airflow through the case (and hence cpu coolers and memory modules)?

So my bet would also be temperatures...

As was already stated: check for ECC errors! Real operating systems show those in their logs (no idea about windows), otherwise check the BMC, supermicro IPMI usually also shows ECC/memory errors (and others that might lead to crashes/freezes) in the health event logs.

Usually with too many ECC errors (or other malfunction that can be detected) the module (or all on that lane if necessary) gets shut down by the BMC, which should trigger a panic + reboot in the OS. The system then runs with the remaining modules (if not below minimal supported configuration). Not sure if this behaviour is similar on AMD platforms though.

 

[sko]

Which component's temperature should be checked?

If you can already link the problem to memory: How about the memory modules?

 

[sko]

ipmitool. Or simply look at the sensor readings in the ipmi web interface (where you already checked the health event log, right?)

 

[RolloZ170]

I did not find the ecc shutdown function in ipmi, there is nothing like that there.

its is the BIOS.
IPMI log can show ecc err's

 

[RolloZ170]

If install 16 pieces of RAM from 16 pieces, the system freezes after 8 hours.
Both processors are overclocked via ZenStates

your RAM(at least the last two DIMMs) may not like OC. downgrade the RAM clock to stock values.

 

[RolloZ170]

16pcs of memory have a frequency of 2666, so it is unclear why decided that this is a frequency issue.

with OC RAM is clocked higher too but latency is not increased as it should.
your RAM is not binned, some can, some not.

 

[RolloZ170]

i remember a freeze can also caused by a BIOS bug, EFI uses memory but not declared in table.
can you try other BIOS version to check ?

 

[RolloZ170]

If this is an error, then why did the system work for more than 8 hours with 14 pcs of RAM installed?

with less memory the EFI uses memory which is not used by OS. it is strange but happens.
just a guess.

 

[RolloZ170]

The program "ZenStates" overclocks the processor, why should "ZenStates" secretly overclock RAM?

IF clock = RAM clock.

 

[RolloZ170]

If the program does overclocking on the bus, then why did the system work for more than 8 hours with 15 PCs of RAM installed?

the 15 pcs can do the (little?) overclock, the others not.
just do not overclock and see what happens.

its hard to debug without spare mobo, CPUs and RAM.

 

[RolloZ170]

If the program does overclocking on the bus, then why did the system work for more than 8 hours with 15 PCs of RAM installed?

hey: now 15 working, one more than yesterday.

 

[RolloZ170]

With 15 dies, it worked stably, I remember it, when I put 16 dies, then instability appeared, I returned 15 dies back, but stability did not return.

does the RAM slot matter or can you remove anyone ?

 

[RolloZ170]

With 15 dies, it worked stably, I remember it, when I put 16 dies, then instability appeared, I returned 15 dies back, but stability did not return.

the BIOS doesn't like swapping/adding/removing RAM.
between config.changes you should clear CMOS to prevent issues.

 

[erock]

• Does the freeze occur at exactly 8 hrs or approximately 8 hrs?
• What CPU's are installed?
• Are the CPU's new or used?
• Are you checking for memory errors using rasdaemon (see STH posts for scripts)?
• What is the load distribution prior to freezing?
• What do your CPU and VRM temperatures look like prior to the freeze?

 

[RolloZ170]

If that's the case, then why does the system work for 8 hours and not less?

i remember a similar issue with Intel ME, if any fatal error(e.g. incompatible pch:me version) the unit turns OFF after 30 minutes exactly.
maybe there is a similar mechanism in the EPYC SoC PSP.