Epyc 9755 ES and MZ73-LM2 problems

[terryww]

Having MZ73-LM2 with latest BMC version and BIOS version R15_F39, using 1 CPU (AMD Eng Sample: 100-000001245-08), I see on the monitor only "Please wait for chipset to initialize...". I tried to update the BIOS as the latest version is R19_F40 (web UI -> file from downloaded ZIP: RBU/image.RBU), says it flashed successful, but the BIOS version stays the same even after waiting for 1h and then removing all power so that the BMC reboots (I don't remove the battery in the motherboard). Tried to power cycle the machine with same result. Another strange BIOS thing I notice is if I change a setting in the current BIOS and save it, the changes don't persist.

Somebody mentioned ES work only in pairs, so tried to run using single rail LC Power 1650W PSU which just spins the fans a bit and then shuts down. I assume I need more power.

Talking to ChatGPT it mentions some vendors remove CPU IDs from later BIOS versions so I tried to flash with the oldest version I could find, but the behavior is the same as when flashing with the latest version - BMC says flash successful, but the version stays the same.

 

[RolloZ170]

using 1 CPU (AMD Eng Sample: 100-000001245-08), I see on the monitor only "Please wait for chipset to initialize...".

AMD EPYC ES work in pairs only. buy a second ES or buy a single socket motherboard.

BMC says flash successful, but the version stays the same.

inventory and BIOS version info is updated by successfull POST.
no POST = old data.

 

[terryww]

AMD EPYC ES work in pairs only. buy a second ES or buy a single socket motherboard.

Oh so single socket only if also motherboard single socket. Is there more I can read about this since nowhere on the internet I saw this info?

inventory and BIOS version info is updated by successfull POST.
no POST = old data.

Ok, so I've flashed the latest BIOS now multiple times and it actually wrote it, just it needs to POST? How I wish this to be true . OK, apparently then need a more powerful PSU to be able to start both CPUs.

 

[RolloZ170]

Is there more I can read about this since nowhere on the internet I saw this info?

we must have learned the hard way.

https://forums.servethehome.com/index.php?threads/finally-overclocking-epyc-rome-es.28111/

Ok, so I've flashed the latest BIOS now multiple times and it actually wrote it, just it needs to POST?

the BIOS sends the BIOS version to the BMC. but if you get no successfull POST you see old BIOS version.
you are the 1000000000 user who asked why the BIOS is not updated....

 

[RolloZ170]

if you run ES it is generaly not a good idea to flash latest BIOS.
ES support is not added, it is (maybe) removed some day.

 

[terryww]

Y, ChatGPT said that vendors sometimes remove support for ES CPUs in later BIOS versions or that a special version was needed. Somebody on Reddit said that in their case the CPU POSTed only with the latest BIOS version and was stuck before at "Please wait for chipset to initialize...".

we must have learned the hard way.

Thanks for the source. Yeah, expensive hobby to have I see now...

 

[RolloZ170]

Somebody on Reddit said that in their case the CPU POSTed only with the latest BIOS version and was stuck before at "Please wait for chipset to initialize...".

most gigabyte's support intel ES still with latest BIOS, but i know other vendors not.
if AMD removes support or doo block, all vendors will be affected because they have to use AMD's AGESA.

 

[Squuiid]

We couldn't get a single (non-ES) 9845 to post on this board. Two of them works fine, once you have the latest R19_F40 in place. We tested R19_F40 while it was in BETA and it does resolve the eternal 'waiting for chipset to initialise' message, particularly when loaded up with 24x 128GB DIMMs and two H200s. Prior to this BIOS it would just not progress beyond that message, no matter what we did. Gigabyte were pretty good at working with us on this.

 

[terryww]

We couldn't get a single (non-ES) 9845 to post on this board. Two of them works fine, once you have the latest R19_F40 in place.

Thanks for the info. That's crazy to me how they can release things without mentioning any of it anywhere. Guess the next motherboard will be anything but Gigabyte.

Looks like I need to get a single rail PSU with correct power sequencing. I see a lot of server PSUs on Ebay, but wonder how to get the right ATX power connectors. It looks like the way to go would be 2x CRPS PSUs with a cage with cabling, but can't find the specs on cables for cages like these.

 

[Squuiid]

Thanks for the info. That's crazy to me how they can release things without mentioning any of it anywhere. Guess the next motherboard will be anything but Gigabyte.

Good luck finding another board that can do dual proc, 500W, 12 channel DIMM, EPYCs though, AND fit in a standard case.
Kudos to Gigabyte for making this sku!

 

[terryww]

I wonder how problematic it would be to use TURIN2D24G-2L+/500W in an E-ATX case. This to me looks like the ultimate motherboard one could wish for - connect as many GPUs, HDD backplanes as you want + each CPU has a separate exhaust vs in case of MZ73-LM2 one CPU will be a few degrees higher when on air cooling. Guess the safest option would be finding a case that has this motherboard and PSU already.

 

[RolloZ170]

I wonder how problematic it would be to use TURIN2D24G-2L+/500W in an E-ATX case

ASRock Rack ? you can kill them also from your list.
TURIN2D24G-2L+/500W proprietary form factor. i doubt you can buy a bare mobo.
no BIOS to DL. you have to beg for it. they are late with motherboards and for that reason almost no ES support.

 

[terryww]

Rly wonder what small SMBs are using as everything not bought by components starts at a price of a new car.

 

[RolloZ170]

Rly wonder what small SMBs are using as everything not bought by components starts at a price of a new car.

that exp. parts is most with good support and good warranty conditions.
anything a company buy/pay is subtracted from the profit, and only the profit is taxed.

 

[terryww]

Update: tested with FSP Cannon 2500W, 208A on single 12V rail, same behavior - fans start spinning and then I hear the PSU shutting off. And a LED starts blinking on the motherboard. According to ChatGPT, the PSU doesn't provide proper power stepping and the motherboard gets power from all cables at once which triggers it's over current protection...

 

[RolloZ170]

And a LED starts blinking on the motherboard.

BMC heartbeat LED, start blink with standby power. normal behave.
if the OCP wriggers at 208A you have a short circuit.
remove CPU and inspect the socket.
some seller ship EPYC 180 degree wrong in carrier.

 

[terryww]

Tnx for the suggestions. Have checked that the triangle on the CPU aligns with the triangle on the blue plastic carrier. BMC LED is in the middle of the motherboard. After PSU shuts down it stops blinking and stays on, while a LED in right bottom (there's an array of about 4 LEDs together) starts blinking (and before power on it was solid on, no blinking). Video: PXL_20250828_082744734.mp4
Not sure if there's something with one of the CPUs as each gets to "Please wait for chipset to initialize..." when in single socket (CPU0). Would think there's sth going on in the second socket (CPU1), but looks ok (without using a microscope).
Or if it's really the power stepping ChatGPT talks about:

Electrical capacity ≠ signaling compliance

• The FSP Cannon Pro 2500W is an ATX/HEDT “consumer-class” unit.
• Your Gigabyte MZ73-LM2 is a server board, designed against the Intel SSI EPS/CRPS spec, not the looser ATX desktop spec.
• That means it doesn’t just want raw 12 V on the EPS plugs — it wants:
  • Precise ramp-up behavior (voltage rise time, monotonic ramp).
  • Correct PWOK (Power Good) timing relative to PS_ON#.
  • Stable hold-up time during short AC dips.
  • Remote sense (many server PSUs feed back voltage at the board).

If any of these are out of tolerance, the BMC/firmware will reject the PSU → blink the fault LED → cut power.

But wonder why then only 1 CPU can be powered. Now with that info added, ChatGPT thinks it's the transient load that's too high.

 

[RolloZ170]

Now with that info added, ChatGPT thinks it's the transient load that's too high.

should work with a 1200W PSU
but some can not the given power @120V AC
you have the 2x EPS 8pin connected ? ( P12V_AUX1/P12V_AUX2 not visible in the video)

 

[terryww]

I'm in EU, so from the main power cable comes 220V. Yes, everything connected (1x24pin, 2x 4+4pin CPU power cables, 3x 6-pin PCIe)

 

[RolloZ170]

in theory the PSU is bad, can't deliver more than one CPU needs.
remember: the item was good the second before it bricked.