CPU and DRAM Ez Debug LEDs (B650M-A WIFI and 7600x)

[nanit155102dc]

(for setup details see bottom)
Upon first bench assembly, red and orange LEDs are lit. Tried numerous things:
- updated bios via quick flash button & usb
- leave it on for half an hour for 'memory training'
- RAM in numerous positions, both dual and single sticks
- reseat CPU numerous times (always confirming no bent pins)
- try different power cables / ports from power supply
- short the JBAT1 jumper to clear CMOS numerous times
- tighter / looser screws from cooler

After hours of all that and just before giving up hope, it did finally POST and load up the BIOS, when I had a single stick of ram in the DIMMA2 position. I rejoiced but noticed RAM was on 4800 MHz so I turned on EXPO profile to 6000 MHz, and it booted up quickly with 6000MHz (more rejoicing). I then shut down and installed the second ram stick into DIMMB2 position, at which point it permanently stopped working -- even after I removed that stick. I cleared CMOS, waited for 'memory training', reseat CPU, adjust cooler ad nauseam. I give up, and to be honest I don't trust this system to last long even if it did manage to boot!

From reading this forum and elsewhere this seems to likely be a memory communications issue that comes from motherboard or CPU, maybe RAM, but may be sensitive to the AM5 socket clamping (depending on contact frame and cooler pressure). Unfortunately I have no spare parts to isolate the issue myself.

Fortunately the mobo/cpu/ram were bought in a bundle from a physical store, just a couple of days ago. They invited me back there so they can identify and exchange the defective part, and send me home with a known-working assembly. I'll post the conclusion later.

Motherboard: PRO B650M-A WIFI, date code "...B2405..."
CPU: Ryzen 7600x . date code reads "2436PGK"
RAM: T-FORCE Vulcan DDR5 6000MHz CL30 XMP 3.0 Dual Channel RAM Kit (2 x 16GB) PN: FLBD532G6000HC30DC01
PSU: Corsair RM850x
Cooler: Phantom Spirit 120

 

[nanit155102dc]

Next day: I took it into the shop. The board came to life after they swapped out for another stick of RAM, and then continued working when switching back to my RAM.

However, the problem returned. But I have found a highly reproducible way to make the CPU & DRAM LEDs come on again, and it explains almost everything so far!

To brick my PC (works every time):
- Turn off the PSU switch, or unplug the PSU, while the PC is on.
- Turn PSU and PC back on.
- PC is now apparently in a permanent bad state. CPU and DRAM lights are lit on every boot, unless...

To unbrick my PC (works every time):
- Turn off the PSU.
- Remove the CMOS battery (a bit tricky first time because they wedge a plastic rectangle to keep the spring clip holding the battery, I guess for transport).
- Wait (dunno how long, 20 minutes is enough), or, short together the terminals of the CMOS battery holder.
- Reinstall CMOS battery.
- Turn on PSU and PC.
- CPU and DRAM LEDs do light up but only for 10 seconds, then just the DRAM light, then it's all good.
- On subsequent boots I never see the CPU LED.
- NOTE -- SHORTING THE JBAT1 JUMPER IS INSUFFICIENT TO CLEAR THIS BAD STATE. THE CMOS BATTERY NEEDS TO BE PHYSICALLY REMOVED!

I can turn the EXPO to maximum (6000 MHz) or anything and the PC is completely happy. I can do 'soft' shutdowns from the OS, or holding down the power button, and those do not cause the CPU led to return. Only when the PSU shuts down hard does it cause some kind of corruption of the CMOS (which JBAT1 cannot clear!).

So either the motherboard or PSU are bad, or they somehow just don't like each other. I am going to try another PSU and see if that helps, if not, then try to exchange the motherboard.

 

[Alan J T]

- Turn off the PSU switch, or unplug the PSU, while the PC is on.
- Turn PSU and PC back on.

I have to do this when mine wont post so it will

- NOTE -- SHORTING THE JBAT1 JUMPER IS INSUFFICIENT TO CLEAR THIS BAD STATE. THE CMOS BATTERY NEEDS TO BE PHYSICALLY REMOVED!

Unplug PSU and short Jbat1 for at least 60 seconds to see if this makes a difference.

So either the motherboard or PSU are bad, or they somehow just don't like each other. I am going to try another PSU and see if that helps, if not, then try to exchange the motherboard.

My Money is on the RAM as the main cause
Swap them with each other's ram slot and see if that helps

 

[nanit155102dc]

Interesting, so you are saying that hard shutoffs actually help you?
Well, since my previous post, I've actually installed this system into a case, and installed an OS. Now I've found the PSU hard shutdowns to act differently, no longer reliable, maybe more like yours. Now the hard shutdowns only brick the system sometimes, and frequently they will unbrick the system! I have no idea which change caused this? I now have more things plugged in (more fans an SSD, and front panel connectors) and the board is now vertical and mechanically affixed to the case. I also had to reinstall the cooler and reattach power cables in the process. But I did not touch the RAM nor loosen the CPU, and it's the same PSU.

---
A further thing I found now: in the bricked state (i.e. CPU + DRAM LEDs on), I can still perform a soft shutdown by holding the power button for 5 seconds. A soft shutdown never unbricks the system (and never bricks it, as I observed earlier). But when the system is in soft-off state and then I turn off the PSU, then it can unbrick too! Note that when you switch from soft-off to hard-off, the only thing that gets turned off is the '5V standby' rail, so that is interesting -- what is getting powered by 5V standby?

 

[nanit155102dc]

Some further investigations: I removed the SSD and extra fans (leaving just 1 CPU fan, and keeping case panel off for cooling), and the bricking is now as reliable as my second post. My working theory is that the low power drain in my system (I have no discrete GPU) compared to the powerful PSU (850W) is causing something odd to occur during the shutdowns. For example, maybe the motherboard is unhappy if 5V standby rail drops too soon compared to other rails.

Also regarding the unbricking using first soft-off (holding power for 5s) and then switching PSU off or unplugging: this seems to be perfectly reliable. So I am doing this now instead of removing CMOS battery, as it's easier and lets me keep my BIOS settings.

Again in bold for anyone skimming through the thread: I can also unbrick my PC by holding power button for 5s to perform soft-off, then turning off the PSU.

I'm going to check moving RAM around in next comment.

 

[nanit155102dc]

OK, now for some playing around with RAM positions and settings. In summary, nothing important happened and you can skip to the conclusion.

I am doing some playing around with the Memory Context Restore (MCR) feature in the BIOS. Previously it was Auto, which seems to mean Enabled (this is the Nov 22 2024 bios version E7D77AMS.1N2).
With MCR = Disabled:
- During a successful boot from soft-off state the DRAM and CPU leds are always illuminated for 28 seconds. (scary at first)
- During restart the DRAM led is illuminated for 28 seconds; the CPU led may flash briefly at the end.
- Hard-offs still always brick the PC (DRAM and CPU leds remain illuminated for at least a few minutes then I give up), and never unbrick it.
- Soft-offs still never brick the PC and never help unbrick it, unless I do first soft-off then PSU off, then it always unbricks.

Playing around with RAM (I am going to leave MCR disabled. Note that EXPO 1 of 6000MT/s 30-36-36-76 1.35V, is still on), it appears the two RAM sticks and two relevant slots are indistinguishable:
- So far I have had the stick with serial number #...374 in DIMMA2 and the stick with serial number #...373 in DIMMB2.
- With just stick #374 in DIMMA2, the LEDs light for 27s, whether restart (only DRAM LED) or boot (DRAM and CPU LEDs).
- With just stick #373 in DIMMA2, the LEDs light for 27s, whether restart or boot.
- With stick #373 in DIMMA2 and #374 in DIMMB2 (reverse of before), it takes 28s, whether restart or boot.
- With just stick #374 in DIMMB2, the LEDs light for 27s, whether restart or boot.
- I now checked the hard power offs with just one RAM stick, and they still brick; I unbrick using soft-off then PSU off.
- Now let's try something against MSI's recommendation: one RAM stick in slot DIMMA1 (#374). The PC refuses to start, CPU and DRAM LEDs lit for at least 3 minutes. OK good to know that recommendation really matters. I'll skip testing DIMMB1 as it's likely the same.
- Returning #374 to DIMMA2 we are happy again and back to the 27s startup.

As an aside, playing with EXPO setting and one ram stick:
- EXPO off (4800MT/s 40-40-40-77 1.10V) takes 17s
- EXPO profile 2 (5600MT/s 40-40-40-84 1.20V) takes 29s
- EXPO profile 1 (6000MT/s 30-36-36-76 1.35V) takes 27s
- Note there seems to be some bugs with the EXPO selector in this BIOS - sometimes the '1' and '2' are not lit but it is actually in EXPO mode.
- Even with EXPO off, hard-offs still brick the system.
Now with two RAM sticks:
- EXPO off takes 18s
- EXPO off with MCR Enabled takes 2.5s
- EXPO profile 2 takes 4s
- EXPO profile 1 takes 4s
- Note when first Enabling MCR on or switching EXPO modes, the first time takes as long as MCR disabled.

Finally, moving RAM does not fix brick state:
- If I'm in the bricked state, and I perform a hard shutdown, then remove a ram stick, it does not unbrick. Possibly if I installed another brand of RAM then it might unbrick, but I have none to test.

So in conclusion:
- The RAM sticks are equal. EXPO is good. MCR is good. Hard power offs (swiching PSU off while PC is on) cause the bad bricked state regardless of RAM configuration. Fortunately I now have good workarounds to unbrick, but I don't like that my PC will get bricked every time we have a power outage, or if power strip gets turned off while running.
- During first boots the memory training happens with both CPU and DRAM LEDs lit. This may happen with the full memory training time (up to 29s in my case) in which case it can be scary because it looks bricked.
- During restarts the memory training happens with only DRAM led lit. I feel like it's a firmware bug that the CPU LED gets lit for first boot, yet not for restarts.
- Installing RAM in the wrong slots, in violation of the manual's instructions, will also cause CPU and DRAM LEDs to be lit.

 

[nanit155102dc]

Unplug PSU and short Jbat1 for at least 60 seconds to see if this makes a difference.

I tried it - it didn't unbrick the system. After I unbricked it (soft-off then PSU off) I could see that shorting JBAT1 did mostly reset the CMOS: EXPO turned off, my fan curves wiped. Interestingly though my saved OC Profiles are still there, and the realtime clock did not get wiped, nor did it reenable Secure Boot (which I have turned off for sake of running memtest86+).

In contrast, when I pop the CMOS battery and short the battery holder terminals, I also lose the realtime clock (set back to 2022/01/01). And, it does unbrick the system. Otherwise it acts like when I shorted JBAT1 -- the OC Profiles are still saved and Secure Boot remains disabled.

Conclusion
- Shorting JBAT1 is insufficient. Shorting the CMOS battery holder does everything that JBAT1 does, but also unbricks the bad state and erases clock.



Some other things I noticed, maybe not as important:

The first successfuly boot happens differently after resetting CMOS (either JBAT1 or shorting battery holder terminals): both CPU and DRAM LEDs light for ~18s, then both LEDs turn off for 7 seconds, then only DRAM relit for 18s doing the DDR5-4800 memory training. (Whereas normally when I boot up and it needs to memory train, I have both CPU and DRAM LEDs lit during the memory training.). I guess after a CMOS reset it does a normal boot then a warm reset and boots up again, but both times it has to do full memory training.
-Note that if they system is bricked and I only short JBAT1 and so it is still bricked on the next startup, the CPU and DRAM LEDs are lit solid and so the system is not even making it through the 'first boot'. Only after I unbrick it (soft-off then PSU off) do I get to see the double boot thing.

I also noticed something funny when I was loading the OC Profile -- it stopped doing warm reboots (ACPI S0->S0). Instead every restart was a power cycle (soft off then soft on, ACPI S0->S5->S0). I can hear something, I think the PSU fan relay, clicking 'off' during every soft-off and also clicking 'on' a few seconds after every soft-on. And for these power cycle reboots I see both LEDs light up during memory training. In order to get warm resets back, I have to fully turn off the PC (soft-off and PSU off). I don't think it's any particular setting that does it, because it happens even when I do 'Discard changes' then 'Save And Reboot' (though not if I do Discard And Reboot). It even happens when I merely save settings to a new OC profile and then load that profile, then do Save And Reboot. It seems there is some hidden persistent flag that disables warm resets whenever you load an OC Profile no matter what is in it.

 

[nanit155102dc]

OK now for something meaningful:

Without changing anything else, I installed the SSD (M.2 NVME, it is a Teamgroup T-FORCE Z44A7 1TB SSD). Just popped it in and turned the little quick latch. I then repeated a process of: Turn on the PC, wait for memtest86 to boot off USB drive (earlier I had set USB drive to priority -- the SSD does not boot), then turn the PSU off. Wait a few seconds or minutes, then turn PSU back on and repeat.
Only 2 out of 15 hard shutdowns bricked the system (causing the CPU and DRAM lights to get stuck on the next boot)! And even when bricked, another hard shutdown was enough to unbrick it.

Then I took the SSD back out and tried again 5 times... we are back to problems: the hard shutdowns always cause the system to get bricked, and they never help unbrick.

SSD back in - the first start is bricked. Then hard off, and the system is unbricked.

How could this be? How could SSD influence motherboard CMOS corruption? Here's my theory:
M.2 NVME are one of the very few things that still get powered by 3.3V rail on modern PCs. Almost everything else goes off the 12V rail through the VRMs. So what happens if basically nothing is using the 3.3V rail? My theory is that the 3.3V rail is discharging too slowly when I hard-shutdown the PSU. It is maybe discharging even slower than the 5V standby. And somehow my motherboard really doesn't like it when the 5V standby disappears before the 3.3V. Then if I put an SSD in the system, it draws some decent power and so the 3.3V rail will usually discharge faster than the 5V standby. In other words, with the SSD in, then even when I do a hard shutdown, it gets more of a soft shutdown first.

Anyway, I am interested to see how an alternative PSU performs. I'm also interested to try just squeezing a resistor on the 3.3V pin on the back of the ATX power connector, and maybe dust off the old oscilloscope to see what those voltage discharges actually look like.

The thing I still don't understand is, what is this nonvolatile state that is getting corrupted during the hard shutdowns, and yet gets repaired by doing a soft shutdown then PSU off?

 

[Alan J T]

How could this be? How could SSD influence motherboard CMOS corruption?

I do not know but I swapped out my 1TB M.2 with a 4TB M.2 and it hard bricked my system was stuck on Code 20.
Stripped it back to the bare system, and swapped RAM, CPU and PSU Flashback refused to do anything and would not even power up the system like it normally does.
And only way I got it back was to Wright a new BIOS to the BIOS chip via a programmer, Not something most people have just lying around.

 

[nanit155102dc]

OK now for some load testing on the 3.3V rail.

I found a couple of old ~1 ohm 5W power resistors and soldered them in series, and jammed them into pins 12 and 24 of the back of the ATX power connector (3.3V and COM). It's a bit annoying, they aren't very tight in there. I attached a voltmeter to the midpoint so it should read half of the 3.3V only if both pins are being contacted. I am reading 1.635 V so they're in good contact. That's total of 2.1 ohms across 3.3V so they are passing 1.6 amps and burning about 3 W. And I removed the M.2 SSD.

As I had theorized, over several trials I confirm that this load on the 3.3V rail is sufficient to prevent the DRAM+CPU leds from coming back on when I shut PSU off while PC is running.

Interestingly, when I turn the PSU off while running, then PSU back on, I see that the voltmeter reads 0.042 V even though the system is off and 3.3V rail should not be energized yet. That means there is 0.084V on the 3.3V rail, and it's driving 40mA through these resistors! That could be the PSU leaking or is the motherboard backfeeding from 5V standby line?
- And yet if I turn the PC on and then soft-off, the 3.3V rail properly drops to 0.000 V.
- If I turn the PC fully off but gently (soft-off then PSU off), then turn PSU back on, the voltmeter reads 0.000 V.
- This is with the resistors still in, so I'm not getting the CPU+DRAM LEDs but somehow it is remembering whether my last shutdown was hard or soft.


Now with the resistors out (hard power offs cause CPU+DRAM LEDs), I'm going to directly measure pin 2 (also +3.3V) where my voltmeter probe will fit into.
- I turned on the PSU and voltage climbed up to 1.55 V! I pushed power button and it started normally.
- Soft power off. Voltage drops very slowly. Takes several seconds to 0.20 V then eventually 0.004V. Note that before with the resistors, the voltage dropped very fast.
- PSU off then on. Voltage stays at 0.004V.
- Push power button, PC boots normally to 3.286 V.
- Hard power off. Voltage drops very slowly yet again. Takes several seconds to reach 0.2V.
- PSU on again, voltage climbed up to 1.55V. I push power button and this time the LEDs are stuck on. 3.29 V.
- Soft off, drops slowly to 0.004V. Soft on, still LEDs are stuck on. Soft off, slow drop to 0.004V again, then PSU off.

Finally with the SSD in, and no resistors:
- PSU on. Power on - goes to 3.286 V. Starts normally.
- Hard power off. 3.3V rail drops very slowly.
- PSU back on again, voltage climbs to 1.55 V. Turn on and it is stuck. PSU hard off.
- PSU back on again, voltage climbs to 1.55 V. Turn on and it boots fine. PSU hard off.
- PSU back on again, voltage climbs to 1.55 V. Turn on and it boots fine. Soft power off (slow drop to 0 V) then PSU off.
- PSU back on again, voltage stays at 0 V. Turn on and it boots fine.
Finally, try transitioning back:
- PSU hard off. Remove SSD. PSU back on again, voltage climbs to 1.55 V. Turn on and it boots fine.

Addendum: Neither shorting JBAT1 nor shorting cmos battery holder will clear the 1.55 V state.

Conclusion:
- Both the M.2 SSD and resistive load help stop bricking behaviour, confirming my theory that loading 3.3V rail helps.
- The M.2 SSD is not a resistive load (no surprise) so it doesn't pull down the 3.3V rail as fast, at least as it goes to lower voltages. I think I'd need oscilloscope to tell any difference in the way it departs from 3.3V.
- Something naughty is energizing the 3.3V rail when PSU is first turned on, if the last power down was a hard shutdown. It drives 40mA closed circuit and 1.55 V open circuit. This always happens regardless of SSD/resistor/no load.
- The 3.3V rail being energized is not reliably related to CPU and DRAM LEDs, so probably not causal, but maybe related. Both are concerning issues though.

 

[nanit155102dc]

Okay, I finally got another PSU to try out: an MSI MAG 750GL. It is significantly noisier than the RM850x - the fan runs constantly and makes a noisy whirr. This is for a bedroom PC that I plan to leave on all the time and the idle noise is too much, so I don't think I can keep this one.

The 3.3V rail reads a bit high at 3.4V, but that's within tolerances.

Some tests:
- Hard shutdown (PSU off while running)
- 3.3V rail drops rapidly to 0.99V for several seconds and then drops to 0.
- PSU back on, voltage rises to 1.55 V.
- Turn on, CPU+DRAM LEDs turn off after a few seconds. Yay!
- Soft-off: voltage drops quickly to 0.16 V, then drops more slowly to 0.001 V and stays there indefinitely. (One time it went to 0.20 V, then stayed indefinitely at 0.078 V. Weird.)
- PSU off, voltage drops slowly to 0.
- PSU on, voltage stays at 0.

After dozens of attempts I did once get CPU+DRAM LEDs to light for a bit longer (>10s) but it never got stuck on.

Conclusion:
- The DRAM+CPU LEDs getting stuck on is only happening with RM850x supply and as seen so far, requires light loading on 3.3V rail.
- The 750GL supply seems to drop its 3.3V faster by itself.
- The motherboard seems to be responsible for energizing 3.3V rail when on standby. This happens with a hard shutdown of either power supply and any loading condition on the 3.3V rail.

Is it normal for a motherboard to backfeed onto 3.3V at standby? I am thinking I should exchange the motherboard for this point alone (I still have time).

 

[nanit155102dc]

I exchanged the motherboard and it behaves essentially the same, in all respects!

So it seems to be some kind of design flaw in this motherboard. Or it's a defect in this batch (the previous one's serial number was 49 lower than this one. Serial numbers are 601-7D77-030B2405001xxx). The only difference is that it backfeeds a slightly lower amount (1.53 V instead of 1.55 V).

I think I'll just live with it. I'm going to try to find a quiet PSU that doesn't trigger the DRAM+CPU LED problem. Anyway, future users will need to know if they see the DRAM+CPU LEDs stuck on for >30 seconds then they need to first hold power button down to do soft shutdown, then turn off PSU.

I am curious though whether anyone else sees their MSI motherboard backfeeding voltage onto the 3.3V rail? This is very reproducible for me -- I am seeing this happening with both motherboards and both power supplies, under all load conditions. I made a separate thread about this as it's more reproducible than the CPU+DRAM LEDs and may concern many other people.

 

[nanit155102dc]

OK got a third power supply (MSI A650N) and with it, hard shutdowns do not trigger CPU+DRAM LEDs. It still has the 1.5V backfeed thing after a hard shutdown, actually a bit lower at 1.51V, but I guess it is harmless. During a shutdown it behaves similarly to the other MSI power supply, with the 3.3V rail dropping quickly. Although this power supply does not go to zero RPM, the fan is at least very quiet, and so I'll be going with this.

In conclusion the MSI B650M-A WIFI motherboard does not like the Corsair RM850x power supply, at least not the unit that I got. I'll return that PSU and the noisy MSI one as well.