help with gigabyte MZ73-lm0 motherboard and genoa-x 2x 9684x help

[Venturi]

new gigabyte MZ73-LM0 motherboards
new retail Epy 9684x cpus
new Samsung ram (1.5TB in 64 dimms x24)
2050w power also tested using Ax1600i
Heatsinks put on correctly
..and made the appropriate amount of burnt offerings before I tried to power up.


So I have gone through 4 (back to back ) Rev 2 gigabyte MZ73-LM0 motherboards. All ne, from three different suppliers.

Using BMC, all boards were flashed to the latest F29 bios. built it not in the case, to omit any issues, on test bench set up

Issue - all are paperweights - won't power on (all 4 boards) at this point.

Then:



with both cpu's in, there is no initial boot, power or response, the BMC lights are on, but no fans, no action or anything anything. Used BMC power up function, No response.

Testing was done with only 1 stick of ram per cpu, no other accessories, used onboard video (behavior the same for 4 motherboards)

All power leads connected, the 24, the two 8s and the single 6 pin, even replaced and reset all new cables, and switched to other power supplies.



BMC,for all 4 motherboards, under inventory, reports: "engineering samples" for the AMD epyc 9684x at 2.15ghx,

They are not of course ES, they are retail unlocked, 2.55ghz. I have 4 cpus on hand, two that I tested worked fine in a supermicro H13 motherboard, no issue, and from windows and CPUz you can see they are retail. The other two are also new but I did not test on supermicro.
Unfortunately supermicro does not make a workstation dual 9004 board

Tried / Removed oner processor, and the other stick of ram (the ram Samsung, from gigabytes approved QVL)

Hurray it turns on!
Got to a post and got into bios (F29) with one cpu and 1 stick of ram

Reinstalled 2nd processor and the other stick of ram (resisted urge to drop in the 24 dimms)

Hurray, it boots now with 2 cpus , screen displays "initializing chipset' (for a long time 10 minutes, then get into bios (after the long wait) (behavior is the same for all 4 motherboards)

I set date and time, reboot before making any other changes (save and exit), figured and get the rest of the settings after a reboot.

rebooted, and now it stays on the "initializing chipset" screen "forever" - never does anything else. (same behavior all 4 motherboards)

Experienced this through 4 motherboards (all new MZ73-LM0 rev 2 (the latest) and all 4 mother boards end with stuck on "initializing chipset" after the first bios save.

100% reproducible behavior

Called gigabyte, so far, no solution. Except they were asking for me to send them a pair of CPUs because they don't have any - I'm not funding their research as they posted a QVL where the CPUs are supported. Strange that supermicro shows retail on same cpus, but gigabyte BMC says ES and wrong ghz, was thinking they actually never tested on real 9684x.... microcode?.....


what am I missing, what am I doing wrong? What is the answer? I've returned 2 boards, returned the unused 2nd pair of CPUs, and I'm still trying to make it work with what I have left.

QVL approved ram, 4 cpus (9684x), all power connected, plenty of power (2050w)
F29 Bios (latest), 4 identical behaviors from 4 motherboards, never get past initializing chipset screen after initial power up

No peripherals, just cpus, 2 sticks of ram, all power connected

(No power initially if one tries to start with 2 cpus in the sockets fresh out of the board box)

All the micro socket little pins are beautifully lined up without any issues across all 8 sockets.

I think I'm relatively savvy on building this stuff, but this has me stumped.

What am I not seeing or not trying???

AAAARRRRGH!

HELP!

..tired of banging my head

pics









The prior build, dual 7773x was simple compared to this.

 

[MasterControl]

Hi Venturi,

This is a long shot, but did you try updating the BMC? Latest version is 13.05.09
Did you use the correct torque for the CPU's?

Older motherboard, but similar issue. Turned out to be one of the CPU's:
Gigabyte MD50-LS0 "Please wait for chipset initialize..." at boot

Ram was possibly not fully inserted:
https://www.reddit.com/r/gigabyte/comments/1ahxw7h
Same motherboard, same issue, not resolved:
Question - Home Server Troubles

If all else fails, try one of these:
https://www.asrockrack.com/general/productdetail.asp?Model=BERGAMO2D16-2T#Specifications

 

[Venturi]

Sir. BMC is updated to latest version, have swapped in and out 4cpus, torqued “perfectly”
I even used a gauge checking all the around the gap and distance as well as the count of amount of turns, I have a micro torque wrench as well

I checked the ram insertion with both clips equally clicking in on both sides

I noticed on the link you referenced, that someone had rinse repeat same issue without resolve

however I this case, when those CPUs were put in Supermicro H13 the issue did not manifest

but I appreciate and I am mindful of your suggestion, and there are no long shots, willing to try anything

 

[MasterControl]

Reason I mentioned updating the BMC, your screenshot shows v13.04.16, which is not the most current.

Disregard if you already updated to latest version.

I really feel your pain though... The Asrock board I linked is preliminary and not yet available. I would have thought Supermicro would have put out a dual proc Genoa/Bergamo (EATX form factor), but only offering UP at this time.

 

[Venturi]

Reason I mentioned updating the BMC, your screenshot shows v13.04.16, which is not the most current.

Disregard if you already updated to latest version.

I really feel your pain though... The Asrock board I linked is preliminary and not yet available. I would have thought Supermicro would have put out a dual proc Genoa/Bergamo (EATX form factor), but only offering UP at this time.

Yes sir, you are correct, that was an early screenshot, the bmc has been updated since then.

Asrock and Supermicro have not produced a dual workstation board this time around. Gigabyte support is not even in the same league as Supermicro. This has been a long discussion with gigabyte, none of the people on the tickets are experts and they are always suggesting lame stuff and no one has gotten me a single “expert” on the phone or even on email chain.

Every call seems to start from scratch, no progress

 

[RolloZ170]

Strange that supermicro shows retail on same cpus, but gigabyte BMC says ES and wrong ghz, was thinking they actually never tested on real 9684x.... microcode?.....

you need a full successfull POST with BMC<>BIOS data exchange to update IPMI inventory.
you see old working config, the seller of the MB tested with the CPU you see.

 

[RolloZ170]

Using BMC, all boards were flashed to the latest F29 bios. built it not in the case, to omit any issues, on test bench set up

older BIOS is worth a try.

 

[Venturi]

older BIOS is worth a try.

sir, that is a good suggestion, on their website there is an F25 bios, the new bios is F29

the birds came with F25. When the issues manifested I used the BMC to do the bios update to F29

and this status quo is where I am

it was a logical suggestion and I thought the same in the process

thank you

 

[RolloZ170]

I have 4 cpus on hand, two that I tested worked fine in a supermicro H13 motherboard, no issue, and from windows and CPUz you can see they are retail. The other two are also new but I did not test on supermicro.
Unfortunately supermicro does not make a workstation dual 9004 board

so it can be you have issues with the socket to socket bridge, because they work solo on the Mz73.

 

[Venturi]

so it can be you have issues with the socket to socket bridge, because they work solo on the Mz73.

Possibly, my head can't comprehend 4x motherboards doing it unless there is a bios or hardware design issue

I did attempt on the one instance to set the link width and size and speed of the cpu to cpu comms and even tried to set things in the cbs menu
however nothing comes back after the one and only entry into the bios, and save and exit action renders the next boot to ne stuck on "initializing chipset"

the statistical issue is that the problem has been 100% reproducible behavior across 4 motherboards and 4 cpus.....

 

[Venturi]

Maybe a better question would be:
does anyone have a gigabyte and 9684x that is actually running? Anyone have the same issues?

...anyone?
...anyone?

...Bueller?
...Bueller?

 

[RolloZ170]

Hurray, it boots now with 2 cpus , screen displays "initializing chipset' (for a long time 10 minutes, then get into bios (after the long wait) (behavior is the same for all 4 motherboards)

I set date and time, reboot before making any other changes (save and exit), figured and get the rest of the settings after a reboot.

rebooted, and now it stays on the "initializing chipset" screen "forever" - never does anything else. (same behavior all 4 motherboards)

"figured and get the rest of the settings"
clearly one or more of this settings makes trouble then.

 

[Venturi]

could be, but it must be a default setting causing it as the only thing I ever messed with was date and time and if data and time cause the issue then gigabyte has bigger problems....

its so fragile, that every little change makes for a permanent "initializing chipset" on the screen till I reset the cmos,
there has to be an easier way.

I asked gigabyte to walk me through the bios settings, in 28 days, gigabyte has not provided anyone.

 

[RolloZ170]

its so fragile, that every little change makes for a permanent "initializing chipset" on the screen till I reset the cmos,

EPYC is SoC, can be anything.
i recommend to check with singel CPU and at least 4 RDIMMs(populated according to the manual)
and check if your BIOS settings(still secret) work there.
btw: i can't imagen how dual EPYC can work with 1 DIMM per CPU at all...

 

[Venturi]

I agree with you. I tried it with 24, and theried it with 8 (4 each). Gigabyte asked to try one per cpu.
There is a provision for placement in the manual for only 1 per cpu and I tried in those specific sockets

I have not had a chance to really do anything with the bios, never get past that second boot

 

[RolloZ170]

Gigabyte asked to try one per cpu.

makes no sense

There is a provision for placement in the manual for only 1 per cpu and I tried in those specific sockets

historical.
some SNC configs are not supported by BIOS, even if its possible in theory

 

[Venturi]

You are correct. I did that mostly to get past gigabytes initials set of troubleshooting requests

At this point I think it’s a bios revision that should be the least effort where Genoa x and Bergamo are updated

 

[RolloZ170]

2050w power also tested using Ax1600i

new or old with dried cap's ?
also i remenbered the Corsair 'i' have OCP per plug enabled by default, can give issues at peak currrents.

 

[Venturi]

I’m not sure how to answer the question about the caps, but the Corsair ax1600i is set to “single rail” mode without any ocp settings, the 2050w is also a “single rail “ mode

The ax1600i was only used as test, the power supply for the build is the 2050w

thank you, very good advice for others that may not know to set that

 

[RolloZ170]

but the Corsair ax1600i is set to “single rail” mode without any ocp settings

ok, Corsair changed that default later. but worth to mention thought.