I have come across a BIOS that potentially allows Milan support, just wondering if there are other out-in-the-wild BIOS’ that allow unofficial support?
I am a hobbyist but am always interested in shortcuts or unofficial workarounds for certain boards ever since intel 1151 screwed me out of using a 9900k on my z170x G1 Gigabyte board.
A BIOS that allows Milan support on what?
I think there is something rather crucial that is missing here.
Do you have a motherboard with an SP3 socket that lacks official Zebra/Milan support?
If so, knowing which one would be absolutely essential here.
Edit: Nevermind. I googled, and apparently EpycD8 is a motherboard I had never heard of.
Yes it has been done before.
No it’s not a smart thing unless you know exactly what you are doing.
Also you don’t need to mind you’ll lose the board if things go south.
Let me quickly tell some related stuff.
AMD made a huge deal when Epyc 7003 launched and the gave a BS story how 7003 is electrical different from 7002 yada.
They just wanted to force everyone to buy a new server.
Because it’s all firmware gated.
Some 7003 cpu’s did have a 100W higer tdp so those had to use the new boards.
But a lot of perfect boards just did not het a new bios for 7003.
It’s a open secret Epyc 7001, 7002, 7003 have no electrical incompatibiliteit and all fit in any SP3 socket.
You usually see 7001 and 7002 are both supported on specific boards, and the newer boards had 7002 and 7003 support.
This 2 by 2 split had a practical reason, Epyc serverboards had 2 bioses onboard.
So depending on which cpu was installed it loaded the required bios.
What currently is most common is people jus grabbing a bios file for a newer boards sku and yolo it.
But than they forget to do the BMC and the boards bricks.
From my understanding there is very little difference between the EPYCD8 and ROMED8.
So the bioses and BMC are technically interchangeable, but not fully and needs more resears before atempting it my advice.
I was wrong the EPYCD8 and ROMED8 do differ a bit.
A second to add support for 7003 is to extract the uefi rom of a new board.
And with some C hackery It’s possible to add the 7003 microcide binary in a bios for the old board.
Both ways have had sucess but a walk in the park it isn’t
It’s to late and almost bed time.
I might remember more about this stuff tomorrow.
But I vaguely remember the EPYCD8 only needs a microcore adjustment.
And the BMC firmware needs to be left alone.
Btw you did luck out with a asrock rack board, the way more populair Supermicro H11 can not be bios modded as Supermiro signs al their bioses.
Not exactly true.
An Addition with 7002 was another pcie interface they called WAFL.
It was meant to allow connecting a BMC via pcie without loosing a bunch of lanes.
Because prior, you had to reserve a bunch of lanes to hook up the BMC. More lanes then just go to the BMC.
So boards who do not use WAFL would work with all of them.
Boards that use WAFL will not work with 7001.
H11 as far as i am aware, can be modded freely.
H12 has this new “root of trust” on the lattice that verifies the flashed bios.
It is not really secure, but it puts up a good initial barrier
I seem to have made some wrong rushed assumption in heading to bed.
Now you mention the WAFL, I start to remember yes.
I guess it’s a elegant wat to get the BMC hoocked up.
But it’s not like Epyc is shy on PCIE lanes, of well it’s one more ssd you can connect.
I probably have mixed up the H12 and H11, it might have been the H11 got a root of trust in one of it’s later bios updates.
I vaguely remember a angry serverthehome topic with a bunch of people being mad they can’t flash a modded bios to enable REbar anymore.
I used to have a Supermicro H12, but I didn’t have much luck getting RoT disabled.
From my understanding It was implemented in hardware by a dedicated chip form Insyde, even when flashing a modded bios with a SPI programmer directly to rom chip the system triggers a fail safe mode.
Each boot Insyde checks it full boot stack has all signatures.
But this is why I like Asrock rack and gigabyte server stuff, it’s reasonable open still.
H12 ROT i was told can be circumvented by a simple TocTou.
It checks the eeprom content once at powerup.
If it fails, it flashes back known good images from another eeprom.
Funnily enough, because of that, it has two eeproms for the BIOS and two for the BMC.
Who knows what happens when you fiddle with the second one who is the source of the flashed back image.
There might have been a role back prevention on later H11 but one only needs to swap an eeprpm to get past that.
There is two possible hurdles for the EPYCD8 to not support Milan.
If a Milan posts at all, its not an issue with the straps.
Though that alone will need a bios change due to the way epyc has an 16MB image per CPU generation.
Oh very interesting, very intrestring indeed.
I need to explore the subject a bit more when my H12 board is back form repair.
You mean the AGESA version right?
I was looking in to some related stuff for it.
CPLD and microcode should probably not be a very large issue.
As with some funess we can use the example in milan bios.
If we do hit a wall to bad, it was brain puzzel.
No, not the AGESA, that only lives inside the “BIOS” eeprom. As part of it i mean.
The Lattice / what ever else programmable logic on the boards can be an issue. The HW programmed to work as a SuperIO controller.
Asus had some known cases where newer CPUs wouldn’t work because the FW on those wasn’t compatible.
No idea what specifically, not that they tell us.
Oh yah those, could have sworn the BMC also has SuperIO, oh the BMC can redirect the signalmen to get serial ipkvm.
I think ASUS has some real bad time with Threadripper 9000 yes.
Did tried to help one dude on the for but no disce.
Could als have been flaky CPU mounting pressure, some people got really annoyed getting that question.
And I’m not being a ass about it, I just aks when I suspect It.
I ask if they happen to know about the issue and it is more common most people think.
So they happen to used a torque key/driver the CPU install.
I’m pretty sure one time I tried to help debug and just about all things had been checked except the CPU, I had asked before when I had enough suspicions, and after a lot of wasted testing was done. Guy didn’t even answer the question once and RMA a good board.
The BMC can act as a SIO i think, but the OEMs not necessarily use it for that or fully.
But welp.
Diagnostics for issues could be way better.
I still wonder why we are stuck with postcodes.
Do note that there are EPYC8D/R32 and EPYC8D-2T/R32 variants. The R32 denotes a 32MB bios instead of 16MB.
I have one with an 7551 if you need me to test anything.
I would add that removing the physical bios chip from the epyc8d-2t board is very simple. It is in a sop-8 holder/socket, and not soldered down. You just peal the sticker away with the original bios version and it has two clamshell doors that opens up to 8-sop. Super easy to remove and keep/replace with a different nor flash chip. Mine originally came with a MX25L25673GM2I-08G nor flash chip.