Bug in Intel Atom C2000 series processors?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

bfarnam

New Member
May 20, 2020
19
23
3
St Louis
It seems that all of the resistors are marked 12A but ohm out around 130... when measuring the PCI CLK it is rock steady at 25M but the LPC0 and LPC1 are AROUND 25M (as well as LPC2 and LPC3).

I confirmed that ALL the resistors are between 3.3V (which measures out at 3.46 on my mobo) and the clock signal.

20211125_152846_c.jpg

Also interesting is the last recorded failure in the IPMI log...

1637885896512.png

So I removed the heatsink and the thermal pad was in pieces. I applied some artic silver, reattached the heatsink, and then pulled it back off to confirm fitment. The heatsink has standoffs in order to limit the pressure on the thermal pad and the thermal paste never touched the heatsink making me wonder if this is really a "toasted" CPU and not a clock issue. I replaced the thermal pad with one I had laying around but I still can not get video or a boot. In fact this is what the main page of the IPMI reads:

1637886054401.png

I am wondering with other CPUs that have the temperature error if this is unrelated to the clock issue.

Has anyone confirmed that the temp error is caused by the bad LPC clock?

20211125_153329_s.jpg
 
Last edited:

RageBone

Active Member
Jul 11, 2017
654
169
43
@bfarnam have you checked and verified that the Bios-eeprom is still good and contains the right bios?
Are you able to get postcodes from the Board?

What Ram are you using?
My SM Atoms only take ECC Udimms, so that might be a reason.
 

bfarnam

New Member
May 20, 2020
19
23
3
St Louis
@RageBone I actually have a fully functioning C2750D4I that is a later version with a C0 stepping CPU. The C0 stepping CPU has a different heatsink design which screws in from the bottom with springs for tension. There are no standoffs on the CPU heatsink. This is probably the biggest giveaway between the two CPUs and this is why I am thinking this is a heat related failure.

I have verified that the memory I am using works fine in my working system. I have also "hot swapped" the BIOS to flash the bad board BIOS with the latest BIOS. The IPMI/BMC BIOS is up to date since you can flash that from IPMI.

The CPU is definitely generating more heat in the heatsink since I replaced the thermal pad - but even a dead CPU will generate heat.

I do not have a device to read post codes - I have just about everything else . . . LOL I need to get one. However, when looking at the PCB, the components for the audio output seem to be populated, there is just no header. I will take another closer look and if everything is there, I might just solder in a header and plug in a speaker to see if it sends out post code beeps.

To buy a post code reader, should I buy a TPMS version or PCIe version?

What would you recommend I should take as my next steps?

BTW, I am seriously contemplating purchasing a C0 stepping CPU and swapping out the CPU. I do a lot of SMT assembly with QFPs, just never attempted a BGA. I found the correct BGA stencil (and holder) to use with the C2000 series CPU and even found a C0 C2750... shouldn't be too hard. In regards to SMT, I have been using SA305T4 paste and have been extremely happy with it. Have had no issues with QFP 100 packages and even tolerates poor stencil quality with no solder balls. So this probably would be ok for the BGA. The biggest challenge will be proper final reflow peak temperature without harming the plastic components... not undoable, but don't think I will be able to utilize my reflow oven for final peak temperature. I will probably have to result in using my hot air workstation.

Loctite GC 10 SA305T4 Data Sheet

EDIT:
Some other notes: The power switch turns "on" the mobo. A long hold on the power switch turns "off" the mobo. I have no way of knowing if the reset button works. Haven't checked the state of the system fans. Normally, the fans come on high, then throttle back during POST. I will check that when I check the audio output.
 
Last edited:

bfarnam

New Member
May 20, 2020
19
23
3
St Louis
And some more thoughts: I went back to the Intel Datasheet after measuring around 25M on each of the LAD pins at the TPMS header. I also looked at the TPMS specification. TPMS uses the SPI serial communication through the SMB bus pins and uses FOUR data lines, LAD0-3. There are only four clock/GPIO pins that output 25M from the C2000 chips: LPC_CLKOUT0/GPIOS_26, LPC_CLKOUT1/GPIOS_27, FLEX_CLK_SE0/GPIOS_19, and FLEX_CLK_SE1/GPIOS_20. The outputs of the LPC Clocks are controlled in part, but not exclusively, by the LPC_CLKRUNB signal in order to switch the LCP Clocks from output to input.

This pretty much kills the idea of generating an external clock signal for those that thought about that.
 

bfarnam

New Member
May 20, 2020
19
23
3
St Louis
OK Speaker header is missing components... so that leaves me back to a POST card via TPMS or PCIe....?

20211126_105827_c_s.jpg

I have verified that all fan headers work and are reported in IPMI/BMC. I have also verified that reset switch also works (via the fan cycling) and that IPMI/BMC has full control over the "host" being able to power up, down, and reset.
 
Last edited:

Jyrki

New Member
May 14, 2013
11
3
3
components:
resistor to fet/bjt control ; resistor to limit current to speaker ; the fet/bjt
diode to protect motherboard components from the speaker, which may be coil based (or piezo speaker)

"speaker" is driven with pwm signal. use a oscilloscope to measure the signal coming from "?" in your picture and no need to solder components. i.e. from the left most pad of left most unpopulated resistor
 

RageBone

Active Member
Jul 11, 2017
654
169
43
Just because those parts are missing, it doesn't mean that the speaker won't work.
At least i would not just assume such without trying.

Most boards don't send postcodes on pcie.
There are some, and some even have bios settings to enable that, but those usually are default off.

Another issue with the TPM Ports is that the ports differ between OEM and Generation.
Your picture shows the old and general ones that was in use till like two years ago.

There are "all" in one debug cards that have pci, pcie, and multiple TPM port versions.
Though, trhose should not work on current SPI/TPM ports as far as i think.
Elmore Labs also offers a debug thingy.

If you are getting in the thick of it, a simple 2 digit code will not be enough.
There are more then just two digits and codes change way quicker then we can see.
So a Log would be needed as well and the only current options would be Software on the BMC or a selfmade , likely FPGA based solution.
 

RageBone

Active Member
Jul 11, 2017
654
169
43
You could also measure crystals / oscillators (marked with x) to figure out if they are working.
I had a few cases of dead ones.
Their output voltage should be something weird.
Not 0 or 3.3V.

Another desperate attempt would be to freeze the board.
I'd say that in most cases where baking helps,freezing does so as well.
 

bfarnam

New Member
May 20, 2020
19
23
3
St Louis
@Jyrki I work with pwm audio a lot... i don't know why I didn't think of putting my scope on there. That's how I got the freqs of the clock signals... I have a nice, albeit old, as in green crt old, Tektronix 100M scope that also has a digital multifunction display for freq, among other things... I will do that sometime today.

@RageBone yes i looked at elmore labs devices... and yes this board is tpm is not the newer spi version. I guess i could try the freezing.... have to get some canned air.. lol
 

RageBone

Active Member
Jul 11, 2017
654
169
43
@RageBone yes i looked at elmore labs devices... and yes this board is tpm is not the newer spi version. I guess i could try the freezing.... have to get some canned air.. lol
[/QUOTE]

A ziplock bag and a normal freezer are usually enough.
I haven't really found science papers on it other then cryo-freezing, so i can't say more then just to try it.
 

fynchy

New Member
Jul 28, 2022
1
5
3
Posting a reply for future individuals in case they come across this issue. I have a Supermicro SYS-5018A-TN4 from 2016 which bricked during a reboot. I contacted Supermicro today about the issue. They are still honoring fixing all C2000 boards even if you have no warranty, although you have to pay shipping cost to them ($45ish for me). Just make sure you tell them specifically it is the issue with C2000 boards.
 

ericloewe

Active Member
Apr 24, 2017
327
153
43
31
Some further notes to clarify some of the details of this mess:

Supermicro at least, and probably all vendors are using separate clocks for the TPM and the BMC. The TPM clock luckily seems to be commonly the one driven by the pin the SoC uses to choose the boot mode (SPI or LPC) or internally affects said pin - so systems can be quickly bodged to boot, as widely known.
This explains why some systems will boot but not have the connection to the BMC working. This also explains why adding two resistors is Intel's workaround for this - it's one resistor for each clock.

More info and high-resolution pictures of a homebrew copy of Supermicro's fix on A1SAi/A1SRi boards, as well a convenient test point for the BMC LPC clock over at the TrueNAS forums.

Posting a reply for future individuals in case they come across this issue. I have a Supermicro SYS-5018A-TN4 from 2016 which bricked during a reboot. I contacted Supermicro today about the issue. They are still honoring fixing all C2000 boards even if you have no warranty, although you have to pay shipping cost to them ($45ish for me). Just make sure you tell them specifically it is the issue with C2000 boards.
Can confirm as of late last year.
 
Last edited:
  • Like
Reactions: abq

RZSN

New Member
Mar 4, 2023
12
6
3
What I can tell you is that the "platform" fix that was applied included one surface mount resistor under the CPU in addition to multiple 120 ohm 1% resistors along the underside of the TPM header.
I have manged to bring my C2750D4I back to life by soldering the 5+2 resistors - mine missed two and your pictures shows soldering on those two places as well - one of them is in fact a TPM signal (IRQ), but they preferred to populate the unpopulated location within the fix, the remaining signals likely had no such DNP pullup part in the design. So I have used seven 120R in 0402 size (even for the TPM header).

The board was a little picky about the memory population order - reading the manual for this main-board is strongly advised. Fully works now! Thanks a lot for sharing those pictures.
 
  • Like
Reactions: cj1

zac1

Well-Known Member
Oct 1, 2022
432
362
63
Posting a reply for future individuals in case they come across this issue. I have a Supermicro SYS-5018A-TN4 from 2016 which bricked during a reboot. I contacted Supermicro today about the issue. They are still honoring fixing all C2000 boards even if you have no warranty, although you have to pay shipping cost to them ($45ish for me). Just make sure you tell them specifically it is the issue with C2000 boards.
SM is still fixing C2000 boards as of March 2023. My fanboy level is now over 9,000.
 
  • Like
  • Haha
Reactions: abq and Markess

cj1

New Member
Mar 5, 2023
3
1
1
NL
I have manged to bring my C2750D4I back to life by soldering the 5+2 resistors - mine missed two and your pictures shows soldering on those two places as well - one of them is in fact a TPM signal (IRQ), but they preferred to populate the unpopulated location within the fix, the remaining signals likely had no such DNP pullup part in the design. So I have used seven 120R in 0402 size (even for the TPM header).
Thanks for sharing your C2750D4I repairs. And compliments for soldering 0402's behind the TPM(-S) header.

Bfarnam picture of the 5 resistors under the TPM-S header was clear. LAD0, LAD3, PCICLK, LAD1 and LAD2 should all have a resistor, connecting to a shared point. Although I have no clue to what kind of trace the 5 resistors connect to. Any hint?

TPM-S seems easy to fix. I've ordered a few 9 pin female headers with 2 mm pitch (W+P 316-09-1) to solder the resisters onto.

Would you be so kind to share a picture of the situation after you have soldered the 2 resistors under the CPU. Bfarnam's CPU picture is a little unsharp. It would be even more helpfull if you would like to mark your 2 added resistors in the picture.
 

RZSN

New Member
Mar 4, 2023
12
6
3
Sure, here are the pictures. I have modified a C2750D4I rev 1.02, but also own a C2550D4I rev 1.03. That one had one of the resistors already there (first in group of three) and it measures 131.6 ohms (in-circuit), so my mods using 120ohm probably wont break anything. That signal is the TPM header pin 16: SERIRQ# and I have read somewhere a note that if it is unused, it should be pulled up (DS / errata / forum post).

The signal trace which the pullups around TPM are connected is same as the pin 9 (it is ugly on my pictures as it was soldered on mine from a previous LPCCLK only pullup, but I opted for recreation of the modification as pictured - you can of course use that pin from the tpm header and bend the wire around as necessary).

The wire looks thick, but its a piece of a wire from an old 80-pin PATA cable (25mil pitch ribbon), tinned before use. Also I need to clean this a bit, but I usually go to testing stuff before, as my flux is marked "no clean flux" and thus should not be electrically interfering.
 

Attachments