Finally: Overclocking EPYC Rome ES

[heavyarms2112]

Looks like CBS is unlocked (kind of) and it does give me some additional controls, but I am still not able to push this cpu any harder than the stock rom with determinism set to power. Its nice having control of FCLK and UCLK controls now so I can see what makes a difference performance wise (which appears to be nothing lol). What I think I am running into now though is still a lack of power. Has me wondering what on this board is adhering to whatever that cpu is telling the bios that its power limit is. Shame voltage control isnt in there since that would give me some more leeway if it continues to refuse higher ppt/edc values.

did you not have cTDP control in the CBS menu?

 

[SHD_Tech]

its there but for my specific processor, 7D12, the max cTDP is 105w. You can change the value to whatever you want, but the processor will not consume more than that. I find this strange since the KRPA U16 apparently doesnt care and will allow 3ghz all core boost. All that was changed on the asus board is core optimizer and engine boost in bios.

 

[heavyarms2112]

its there but for my specific processor, 7D12, the max cTDP is 105w. You can change the value to whatever you want, but the processor will not consume more than that. I find this strange since the KRPA U16 apparently doesnt care and will allow 3ghz all core boost. All that was changed on the asus board is core optimizer and engine boost in bios.

I thought you need to bump both cTDP and package power limit.

 

[SHD_Tech]

nah. What my guess is asus boards let you ignore the ppt of epycs and lets them boost harder. Would explain why they set so many records with those boards when they launched them.

 

[Layla]

Wanted to to the test again with SMT enabled, but the system hangs during POST with code 0x55 (Memory not installed) after changing back to SMT Enabled. Anybody encountered this?

Note that I've already updated Windows Server 2019 and applied the SP3 driver package from SuperMicro.

A few years later, I've also found yet another reason why the system might appear to "freeze" with POST code 55; except this time, it turned out to not actually be frozen, and a totally phantom "issue" caused by iKVM:

Supermicro H11DSi-NT BIOS POST Phantom Issue | ServeTheHome Forums

 

[Layla]

After installing new RAM (512GB) and increasing some of the CPU power limits, I also may have just killed my boot SSD (SM953 M.2 22110 with PLP) during an overclocked run of Cinebench which black-screened (for a few minutes, I thought I'd killed the CPUs in the manner discussed by the OP, as the system wouldn't boot up via IPMI at all without hard cycling the power supply itself). After getting the system to POST again, the SSD (and Windows Boot Manager) now no longer shows up in the BIOS boot menu. I'll try to see if it's just in a bad state which can be repaired tomorrow, but given the PLP, I would have expected it to be more resilient against that...

 

[NablaSquaredG]

I also may have just killed my boot SSD (SM953 M.2 22110 with PLP)

It is known among Ryzen overclockers, that Ryzen is very susceptible to USB & SATA issues when being overclocked, especially when messing with Infinity Fabric clock etc...

 

[Layla]

I'm not surprised there would be stability issues (that SSD is NVme, and the pci-e traffic goes over the overclocked IF) but it seems like this might have actually done real damage (voltage issue?) - is that also a thing known to regularly happen?

 

[Layla]

I also may have just killed my boot SSD (SM953 M.2 22110 with PLP) during an overclocked run of Cinebench which black-screened (for a few minutes, I thought I'd killed the CPUs in the manner discussed by the OP, as the system wouldn't boot up via IPMI at all without hard cycling the power supply itself). After getting the system to POST again, the SSD (and Windows Boot Manager) now no longer shows up in the BIOS boot menu. I'll try to see if it's just in a bad state which can be repaired tomorrow, but given the PLP, I would have expected it to be more resilient against that...

Apparently the SM953 doesn't actually have proper PLP:
Samsung SM963 full PLP m.2 nvme ? | ServeTheHome Forums

Maybe this explains why it died when overclocking IF? I put a PM963 in, which should have full PLP, and am in the process of re-installing everything from scratch

 

[Layla]

Unfortunately, I provide bare servers to customers . Most servers will install windows server 2016/2019 operating system. So I will worry about this problem .
If I change ES CPUs to retail CPUs or QS CPUs, can I avoid this problem?

Selling remote access to ES/QS CPUs seems like a very bad idea - there are likely several unpatched bugs in ES CPUs that could impact both security and the reliability of computations...

 

[Layla]

Someone found a way to boot ES cpu on latest bios for the h11dsi ? I know that it's possible to boot on latest bios version on other sm board by downgrading agesa ,I've tried here dowgrading agesa and NBIO pei related driver but without success...

Also is it possible to lower P1_SOCRUN ? Mine are running freaking hot and hard to cool.

Which specific changes did you make which did not work?

 

[RolloZ170]

Selling remote access to ES/QS CPUs seems like a very bad idea

note QS have intel warranty, not intended for sale&re-sale. beside that official QS is same than retail except ES bit set.
QS is usualy same stepping/microcode than retail one. QS are free gifts to engineers(need intel NDA)

 

[SirenCylinder]

Just picked up a single 7702 ES chip, before realizing I'd need another to properly boot. That's fine, but can I mix a non ES CPU with a retail version out of curiosity? Unfortunately I'm finding it a challenge to get another 7702 ES.

 

[RolloZ170]

Just picked up a single 7702 ES chip, before realizing I'd need another to properly boot.

never heart about that. where have you read that ?
on a dual socket board you can not use some PCIe slot and/or SATA/NVMe outputs, but with CPU in first socket(check manual) it runs.

 

[SirenCylinder]

never heart about that. where have you read that ?
on a dual socket board you can not use some PCIe slot and/or SATA/NVMe outputs, but with CPU in first socket(check manual) it runs.

Ah, could be that I just misread that as meaning it won't post at all without a second CPU, not just that it won't post with a single CPU in the second slot.

Either way, the good news is that I found another ES with the same model number so it shouldn't matter for my purposes. It would satisfy a curiousity of mine (and maybe help someone in the future) if I knew if you could mix ES with retail models, though.

 

[RolloZ170]

if I knew if you could mix ES with retail models, though.

have no clear information on AMD EPYC platforms. Intel makes clear rules, but if BIOS follows them is another question.

 

[RolloZ170]

Ah, could be that I just misread that as meaning it won't post at all without a second CPU, not just that it won't post with a single CPU in the second slot

"single CPU in the second slot"
NEVER NEVER try that. on some actual platform's this will burn-out CPU&Motherboard.
in the good OLD day's the memory controller was a chipset (Northbridge) and a dual socket board worked with signle CPU doesn't matter in which socket because the Southbridge connected to both sockets.

 

[NablaSquaredG]

NEVER NEVER try that. on some actual platform's this will burn-out CPU&Motherboard.

Wait what the hell? This shouldn't be possible. If it is, very bad design.

Can you be more specific on that?

 

[RolloZ170]

Wait what the hell? This shouldn't be possible. If it is, very bad design.

Can you be more specific on that?

Ice Lake

 

[brickmover]

New to the party, is the overclocking on Milan stable now?

I have a 32-core Milan and a 64-core Milan ES/QS paired with Gigabyte MZ32 and a SuperMicro H12SSL board.

Ready to get some nice free performance : ]