ASRock Rack E3C252D4U-2T (OEM/new?) stuck at POST code 00, shuts off after ~10-20 seconds - Xeon E-2324G (used) - exhausted all ideas send help

[etorix]

SoC boards (e.g. Xeon D series) that are incredibly expensive, hard to find, and lack upgrade flexibility

There are plenty of X10SDV boards on eBay: Neither hard to find not super expensive—RAM might be an issue now, but not the board.
Lack of upgrade flexibility comes with the mini-ITX form factor, but with onboard 10G and an x16 slot which can bifurcate x8x4x4 to hold a pair of NVMe drives and either a HBA or an Arc A310 for transcoding purposes, these little boards can do a lot.

I still need to figure out a better backup strategy than having my most important files on an external 12TB hard drive

At least you already have a backup strategy. That's a good point!

Seems they are super expensive too.

This site posted many reviews of cheap switches with a few 1/2.5G ports and one or two 10G ports, which should fit a small home network with a 10G NAS, one priviledged desktop and a splatter of connected "smart" things.
Moving a step up in quality, I see Mikrotik CRS-305 (SFP+) from 120 €, CRS-304 (10GBase-T) from 157 €, or QNAP QSW-M408-2C (8*1G, 2 SFP+, 2 combo) from 316 € (not to mention the many further M400/M2100 variants on this theme) on geizahls.de. None of that is "super" expensive in my book.

 

[MSameer]

Has anyone managed to reach deeper c-states? I can only get to C3

 

[TransistorResultConsumer]

Has anyone managed to reach deeper c-states? I can only get to C3

It's been over a week CPU is still not here. I can try once it's here, which is hopefully tomorrow.

 

[TransistorResultConsumer]

I was really worried but the CPU works just fine thankfully.


Will update on the c-state levels once I've gotten everything into the case and set up finally.

BIOS is still showing the G6400, not sure why but that's no real problem for now.

 

[MSameer]

I got a response from Asrock Rack.

I was given a more recent BIOS (5.07) which I have flashed successfully via BMC. I was also given a BMC image which failed to flash with the dreaded error "image verification unsuccessful". I emailed them back so hopefully it will be solved.

I am unable to attach the images here. They seem to be too large for the server to handle but here is the reply I got:

the /OVH variant is using OEM firmware. I’ve uploaded the latest
original lab/beta firmware for your mainboard here:
E3C252D4U-2T - Google Drive

Please note that this link is only valid for a few days.

Please update the BMC first via the webinterface and then update the
BIOS via the webinterface as well.

 

[TransistorResultConsumer]

Thank you for providing those. Did you go on and patch to 5.07, or still on 5.05?

I'm having the same issue with the BMC

 

[MSameer]

I updated to 5.07

 

[MSameer]

More progress. I received socflash from asrock rack and I managed to update BMC using that tool.

I initially used Debian stable but the tool crashed.

I then tried again using latest proxmox with kernel 6.17.2-1-pve and I did manage to flash BMC
./socflash_x64 if=E3C252D4U-2T_3.02.00.ima

I will ask if they have anything to say about package c-state being stuck at C2.

 

[RolloZ170]

I will ask if they have anything to say about package c-state being stuck at C2.

package C can not go deeper if any core is in higher state.
pkg C3 is like a dead processor
if "memory self refresh, memory clock stop" not supported, you can't get C3.

PKG C3

Cannot be requested explicitly by the Software. All cores in C6 or deeper + Processor Graphics in RC6, LLC may be flushed and turned off, memory in self refresh, memory clock stopped. The processor will enter Package C3 when: All IA cores in C6 or deeper + Processor Graphic cores in RC6. The platform components/devices allows proper LTR for entering Package C3.

All processor IA cores in C6 or deeper. Processor Graphics in RC6. memory in self refresh, memory clock stopped. LLC may be flushed and turned off.

source

 

[MSameer]

Thank you.

That is interesting. I am not really a hardware guy so not sure how to map such info to something I can try to investigate.

My CPU cores reach c7 so there is something.

 

[RolloZ170]

My CPU cores reach c7 so there is something.

hardware or BIOS does not support to kill memory controller. older can, newer modern(fast) can not i guess.
afaik memory self refresh can be switched by ME (like required in S3)

 

[RolloZ170]

My CPU cores reach c7 so there is something.

you know how much time it takes to wake from C6 or deeper ? if you have no jobs maybe better use S3 instead ?

 

[MSameer]

you know how much time it takes to wake from C6 or deeper ? if you have no jobs maybe better use S3 instead ?

I nees the server to be running. Not to be in S3.

Was hoping to shave off some power. I idle at around 30W but I can live with that. The cost of the hobby

 

[RolloZ170]

I nees the server to be running. Not to be in S3.

but you know if clocks are stoppen, caches ar flushed and RAM disabled, it is not like a runing machine.
try to find who draws how much. you can save power e.g. by lowering RAM clock.
chipset and BMC needs some. not that you screw on the wrong places.

 

[randomusername1]

More progress. I received socflash from asrock rack and I managed to update BMC using that tool.

I initially used Debian stable but the tool crashed.

I then tried again using latest proxmox with kernel 6.17.2-1-pve and I did manage to flash BMC
./socflash_x64 if=E3C252D4U-2T_3.02.00.ima

I will ask if they have anything to say about package c-state being stuck at C2.

How did you reach out to them? I've sent them multiple messages via e-mail and the support form (ASRock Rack > Support Request Form) but no response to-date.

Also, where did you find the serial number on your board for contacting them? My board doesn't appear to have the label for one!

Wary of downloading and flashing a random BIOS from online strangers, so hoping they may one day send me the damn bios/bmc firmware.

 

[MSameer]

How did you reach out to them? I've sent them multiple messages via e-mail and the support form (ASRock Rack > Support Request Form) but no response to-date.

Also, where did you find the serial number on your board for contacting them? My board doesn't appear to have the label for one!

Wary of downloading and flashing a random BIOS from online strangers, so hoping they may one day send me the damn bios/bmc firmware.

I used that exact same form you used. I did not include the serial number. I even had the BIOS version incorrect.

The form was sent on 11.03 and I got the reply on 31.03

The links I posted were the ones Asrock sent me. They use online storage.

I can upload the images to my server if that makes you more comfortable. Just drop me a PM.

 

[TransistorResultConsumer]

Regarding the C-state question @MSameer

I contacted ASRock Rack support with the same concern and got a reply directly from their engineer.

The _ACPI naming in Linux sysfs is just an abstraction layer.
C1_ACPI maps to CPU C1
C2_ACPI maps to C6/C7
C3_ACPI maps to C8/C9/C10

So deeper C-states are actually working correctly, just named differently than expected.

One thing worth checking on your boards: Package C State Support was set to Disabled by default on my OVH variant. Changing it to Auto in the BIOS makes sense for better idle power management.

@randomusername1 I also used the support form and got a response after a few weeks for my other and after a couple of days for this request.

 

[MSameer]

The _ACPI naming in Linux sysfs is just an abstraction layer.
C1_ACPI maps to CPU C1
C2_ACPI maps to C6/C7
C3_ACPI maps to C8/C9/C10

That sounds strange.
Still I cannot get lower than package P2 but the cores themselves get to c6/c7. I am using powertop to check. How does that map to ACPI C-statrs then?

One thing worth checking on your boards: Package C State Support was set to Disabled by default on my OVH variant. Changing it to Auto in the BIOS makes sense for better idle power management.

I have it set to auto. I recall it did not make a difference wrt idle consumption though.

Anyway I might be booting win10 in the upcoming days on that board (GAH :-() and can do some checks.

Thank you for updating me.

 

[RolloZ170]

That sounds strange.
Still I cannot get lower than package P2 but the cores themselves get to c6/c7.

package C6/C7 and you have 300 µs / 400 µs latency to C0, do you want a lazy system ?
every LAN package will wake up package, in some worloads package can't go deeper C because of work.

 

[MSameer]

package C6/C7 and you have 300 µs / 400 µs latency to C0, do you want a lazy system ?
every LAN package will wake up package, in some worloads package can't go deeper C because of work.

I have no issues with a lazy system.

This will be my home server running light workloads and mostly idle. I know it is overkill but it is also a hobby (and i stick to server boards only). I am not obsessed with low power states. It's just nice to have.

It burns around 30 watts currently (60 with my llm 3090 slotted in). I can live with that. A 10w difference is roughly 1 euro. Not a problem.

I am just tinkering.