Finally: Overclocking EPYC Rome ES

[dassiq]

@ExecutableFix : Awesome work my friend I had couple of zs1406e2vjug5 , that I can now test, thanks to your hard work and persistence.

1 ) I have couple of H11SSl-i (Rev 1.01) boards, I am going to try this on those now.

2) I will bring a new board in this equation , The Tyan S8026, currently I get 'FFFF' post code with zs1406e2vjug5. Board officially supports Rome.

 

[jpmomo]

Try it through the IPMI method.

You may need a IPMI license ( 20$ -30$ from SM) to activate Bios update feature. There is a hack available (Google it) , not sure if it works on H11 /H12 boards.

why do you think that flashing the bios via the ipmi would overcome the ROMID issue? I don't mind spending the money for the ipmi license but I am not sure if this would help. Is there something technically different when flashing via ipmi vs efi command line? I am willing to give it a try if you think it might help get the 2S cpu to boot. I also have a tyan s8030 that I am using with a retail 7742. I haven't tried to boot the 2S cpu on that board yet as I didn't want to mess with that board unless I had to. I am currently using that board with the 7742 and a few pci gen 4 NICs for another project.

 

[ExecutableFix]

why do you think that flashing the bios via the ipmi would overcome the ROMID issue? I don't mind spending the money for the ipmi license but I am not sure if this would help. Is there something technically different when flashing via ipmi vs efi command line? I am willing to give it a try if you think it might help get the 2S cpu to boot. I also have a tyan s8030 that I am using with a retail 7742. I haven't tried to boot the 2S cpu on that board yet as I didn't want to mess with that board unless I had to. I am currently using that board with the 7742 and a few pci gen 4 NICs for another project.

The IPMI is actually different and it accepts more files. I wouldn't buy that license tho, you're able to get the key for free

 

[dassiq]

Hi JP

Probably this obvious: In addition to what ExecutableFix said, Biggest advantage of using the BMC / IPMI functionality is that you don't need a bootable cpu to flash the bios. ExecutableFix has mentioned this few times in different posts that I have seen on the EPYC ES topics here at STH forum.

With IPMI /BMC , you can force overrides during Bios updates. Here is an example using the 'SUM" command line tool which talks to the BMC over the network.

One of the howtoguides from Patrick. Link

Notice the reply from 'i386' .

I can confirm H11SSL IPMI is able to activate the Bios update feature, using the 'hack' or free 'key generation' I mentioned. earlier

 

[jpmomo]

The IPMI is actually different and it accepts more files. I wouldn't buy that license tho, you're able to get the key for free

you were correct. I found the ipmi bios update license and gave it a try. It was able to load your custom bios and turn my rev 1.1 mb into a rev 2.0. I still couldn't get my 2S cpu to boot but that may still be some issue with the cpu. I did test a retail 7452 and it booted properly and ran windows 10 without issue. When I was upgrading to the custom rome bios, I left the option to Preserve SMBIOS. I am not sure if it would help to get the 2S cpu to boot by unchecking that option. I am not sure what other things I can try to get this cpu to boot. It may just be bad as it was sold as is for $100 so I wasn't too optimistic but figured I would give it a go. Thanks again for everyone's help and especially to executablefix as his bios was at least able to upgrade my mb! and now I also have an easier way to update the bios!
jp

 

[dassiq]

You can actually mess with Tyan as well, using BMC/IPMI. You don't even need a Bios update key for Tyan

Side note : Tyan's BMC interface is even more beautiful then SM

Also, before attempting anything, it will be best to update SM or Tyan IPMI with the latest firmware. redfish --xxx---x

SM IMPI default user/password = ADMIN / ADMIN
Tyan = root/superuser

Saved you an hour there

why do you think that flashing the bios via the ipmi would overcome the ROMID issue? I don't mind spending the money for the ipmi license but I am not sure if this would help. Is there something technically different when flashing via ipmi vs efi command line? I am willing to give it a try if you think it might help get the 2S cpu to boot. I also have a tyan s8030 that I am using with a retail 7742. I haven't tried to boot the 2S cpu on that board yet as I didn't want to mess with that board unless I had to. I am currently using that board with the 7742 and a few pci gen 4 NICs for another project.

 

[jpmomo]

You can actually mess with Tyan as well, using BMC/IPMI. You don't even need a Bios update key for Tyan

Side note : Tyan's BMC interface is even more beautiful then SM

Also, before attempting anything, it will be best to update SM or Tyan IPMI with the latest firmware. redfish --xxx---x

SM IMPI default user/password = ADMIN / ADMIN
Tyan = root/superuser

Saved you an hour there

Thanks again for the helpful info. I also like the tyan as this was the first pcie gen 4 board that didn't give me any issues. I guess it couldn't hurt to try the problem child 2S cpu in that mb. the tyan board that I have is pretty new so I am not sure if any other bios would work but I could at least see if the 2S cpu works with the current tyan bios.
thank you,
jp

 

[ExecutableFix]

@ExecutableFix : Awesome work my friend I had couple of zs1406e2vjug5 , that I can now test, thanks to your hard work and persistence.

1 ) I have couple of H11SSl-i (Rev 1.01) boards, I am going to try this on those now.

2) I will bring a new board in this equation , The Tyan S8026, currently I get 'FFFF' post code with zs1406e2vjug5. Board officially supports Rome.

I think the same method applies to the Tyan boards, just duplicate the bios. Let me know if it works

 

[111alan]

Mark, this is good. Does this work on the principle of modifying the MSR registers?

 

[EagerToLearn]

Can the ES/QS Rome CPUs be used in a single configuration on a dual board?

 

[ExecutableFix]

Can the ES/QS Rome CPUs be used in a single configuration on a dual board?

No, unfortunately not as with all the engineering samples

 

[EagerToLearn]

I was hoping, this issue may have been resolved or would not apply to QS samples. But thank you

 

[tesla100]

No, unfortunately not as with all the engineering samples

Do you mean all Epyc Rome ES/QS, or Naples too? On the Intel side, as far as I know, there's no such a limitation.

How do I know if the ES/QS Epyc is for a single or dual motherboard?

Thanks

 

[ExecutableFix]

Do you mean all Epyc Rome ES/QS, or Naples too? On the Intel side, as far as I know, there's no such a limitation.

How do I know if the ES/QS Epyc is for a single or dual motherboard?

Thanks

They will work on dual socket motherboards if they're installed with two sockets populated, not if only one is populated. This holds true for both Naples and Rome engineering samples

 

[tesla100]

They will work on dual socket motherboards if they're installed with two sockets populated, not if only one is populated. This holds true for both Naples and Rome engineering samples

And the same CPUs will work in single socket motherboards? Thank you

 

[ExecutableFix]

And the same CPUs will work in single socket motherboards? Thank you

Of course

 

[tesla100]

Of course

Great, thank you!

 

[Brainbug]

@I.nfraR.ed

I'm surprised and impressed that you were able to identify the SMU addresses and commands that would be required. I looked into it briefly with my ES setup, and the SMU returns are less than helpful.

Very much appreciate the time and effort you put into it. (Both yourself and ExecutableFix)

 

[111alan]

After trying this on some CPUs I have a suggestion.

Could you set the maximum allowed multiplier to 38, so people won't accidentally kill the CPU? The same goes to the voltage, a cap or a confirmation when modifying should be nice.

Thank you for your effort.

 

[ExecutableFix]

After trying this on some CPUs I have a suggestion.

Could you set the maximum allowed multiplier to 38, so people won't accidentally kill the CPU? The same goes to the voltage, a cap or a confirmation when modifying should be nice.

Thank you for your effort.

No, as you can safely set the voltage/frequency higher as long as you're not using the EDC "bug". The 32-core doesn't have the same limits as the 64-core as well.