Overclock your AMD Epyc

alex_stief

Active Member
May 31, 2016
611
186
43
35
Using thermal paste instead of thermal pads is not necessarily better for VRM cooling. The issue here is that the components are not all at the exact same height, in contrast to e.g. the relatively flat surface of a CPU heatspreader. A thermal pad helps to even out the tolerances, making good contact with all the parts.

I am still curious about which parts of the VRM are actually cooled by the stock heatsink. On your picture it looks like the MOSFETs get no cooling at all. That might be worth looking into.

As for the super doctor software: not sure If I get the problem. I can just go here and enter some valid sounding credentials.
 

BryanH

New Member
Oct 28, 2019
9
0
1
Using thermal paste instead of thermal pads is not necessarily better for VRM cooling. The issue here is that the components are not all at the exact same height, in contrast to e.g. the relatively flat surface of a CPU heatspreader. A thermal pad helps to even out the tolerances, making good contact with all the parts.

I am still curious about which parts of the VRM are actually cooled by the stock heatsink. On your picture it looks like the MOSFETs get no cooling at all. That might be worth looking into.

As for the super doctor software: not sure If I get the problem. I can just go here and enter some valid sounding credentials.
So you think maybe the high temps in super doctor could refer to the MOSFETs rather than the chip itself? I wonder if the improvements were from the fan on the heat sink blowing on the MOSFETs. That would actually be a good thing because those wouldn't be too difficult to sink. Interesting... Thanks, more testing forthcoming.
 

alex_stief

Active Member
May 31, 2016
611
186
43
35
I still don't get why you would need to download SD from a dodgy site, but maybe that's just me.

Anyway, back to the VRMs: I only have second-hand knowledge here, but what I gathered over the years: those bulky components that the stock cooler seems to make contact with are the chokes. These are usually not the problem with overheating VRMs. On many motherboards that have heatsinks oon the VRMs, these components remain without additional cooling. It is the MOSFETS (the smaller components in between) that need to be cooled. A classic example with AMDs AM3 platform:
 

BryanH

New Member
Oct 28, 2019
9
0
1
I still don't get why you would need to download SD from a dodgy site, but maybe that's just me.

Anyway, back to the VRMs: I only have second-hand knowledge here, but what I gathered over the years: those bulky components that the stock cooler seems to make contact with are the chokes. These are usually not the problem with overheating VRMs. On many motherboards that have heatsinks oon the VRMs, these components remain without additional cooling. It is the MOSFETS (the smaller components in between) that need to be cooled. A classic example with AMDs AM3 platform:
After everything I have seen, I am convinced you are correct about this. I went ahead and ordered the MOSFET heat sinks and opted for the thermal glue instead of the tape that comes with them. These will not fit with the current VRM heatsink that comes with the motherboard, but they will fit with the water block I ordered.
Thanks again, I will post temps and speeds in a few days after these are all installed.
 

alex_stief

Active Member
May 31, 2016
611
186
43
35
I'm an idiot. Of course the heatsink makes contact with the MOSFETs. I messed up the scale of both photos in my head.
Sorry for the confusion.
 

BryanH

New Member
Oct 28, 2019
9
0
1
I'm an idiot. Of course the heatsink makes contact with the MOSFETs. I messed up the scale of both photos in my head.
Sorry for the confusion.
I understood what you meant, it makes contact with the FET not the choke but the logic is still sound. They both should be cooled.
VRMs.jpg
 

alex_stief

Active Member
May 31, 2016
611
186
43
35
Holy cow, is that a monoblock for both CPUs and the VRMs? I have been pondering a water cooling option for quite some time now, this might convince me to finally do it.
 

nero243

Active Member
Oct 28, 2018
108
76
28
Hello!
I've updated the ZenStates app with Rome CPUID, however I haven't figured out the P-States (or manual overclock) yet.
There is no BKDG for family 17h and setting the well-known P-State MSRs doesn't change the CPU frequency. I have no access to NDA stuff.
If I set the P-State, then read it back it shows the newly set multiplier, but the actual frequency is still the old one. Same with the Linux script.
So I'm currently scratching my head over this.

It would help if you report that at least it detects the CPU correctly.
Performance Bias options "might" work, perhaps some of the others, but not P-State :(
I'm not even sure if the MSR are shared between Matisse and Rome.
Hopefully it won't "break" your system, but apparently I have no way to test.
EPYC1 shared the CPUID with desktop parts, but now Rome has a new CPUID.


Download zip from my github releases page: irusanov/ZenStates

Changelog:
  • Added Rome support (P-States won't work, but at least it should start without complaining about unsupported CPU)
  • Added support for 32-bit OS (tested on XP SP3 x86)
  • Reduced .NET Framework version to 4.0 (lowest I can go for the current codebase)

Update 21 Aug:
I have managed to change the frequency with commands to SMU. Unfortunately there's no public document describing all commands. I've updated the readme with what I've been able to find so far mostly by trial and error method. Should be able to provide at least manual overclock soon.

irusanov/ZenStates
What are the CPUIDs for Rome retail and ES? Was just wondering if there's a possibility to get one of those 150$ Rome ES working on a H11 rev.1 Board. So i'll have to compare the microcodes and AGESA versions from the H12 series with the H11 ones.



Are there any news regarding the AMD CBS Bios menu? I was living under a rock for the last few month.

I kinda gave up my epyc platform, the horsepower for video editing in davinci resolve is insane, but i barely need it anymore.

If anyone is interested in my H11SSLi VRM waterblock or the USB005 Dongle let me know, i'll probably put it up for sale soon.

Edit:
So i did a quick check the CBS Menu is still present in the Supermicro H12 bios. So if anyone feels like picking up working on this again, it's still worth a shot.
 
Last edited:

charliehorse55

New Member
Oct 3, 2019
16
2
3
The $150 rome ES chips on ebay have serious physical damage, you can see the missing capacitor networks, pad damage and even scratches that might be deep enough to cut traces in the pcb!
 

nero243

Active Member
Oct 28, 2018
108
76
28
The $150 rome ES chips on ebay have serious physical damage, you can see the missing capacitor networks, pad damage and even scratches that might be deep enough to cut traces in the pcb!
Most of them look savable. But yeah, it's kind of inconvenient when you have to work against hardware damage and firmware incompatability at the same time.
 

charliehorse55

New Member
Oct 3, 2019
16
2
3
Most of them look savable. But yeah, it's kind of inconvenient when you have to work against hardware damage and firmware incompatability at the same time.
It'd be really interesting to investigate the overclocking headroom as well. With those chips set for 2 GHz turbo vs. 3.5+ for the production version. I wonder how many clock improvements they actually find between these ES chips and the final design.
 

nero243

Active Member
Oct 28, 2018
108
76
28
It'd be really interesting to investigate the overclocking headroom as well. With those chips set for 2 GHz turbo vs. 3.5+ for the production version. I wonder how many clock improvements they actually find between these ES chips and the final design.
I actually don't think there's alot to be gained from a 64 core Rome CPU with the VRM you get on Supermicro boards. While the VRM can theoretically deliver enough power, the vdroop is just insane. You'd have to set your voltage to 1.4v (which has to be applied in idle) to get 1.2V under load. And a jumping voltage like this is not really a recipe for a stable system.
 

diogin

Member
Mar 28, 2018
32
5
8
Beijing, China
I actually don't think there's alot to be gained from a 64 core Rome CPU with the VRM you get on Supermicro boards. While the VRM can theoretically deliver enough power, the vdroop is just insane. You'd have to set your voltage to 1.4v (which has to be applied in idle) to get 1.2V under load. And a jumping voltage like this is not really a recipe for a stable system.
Bios on Intel consumer motherboards often provide a setting called load-line calibration to prevent vdroop, does this setting exist (but hidden) on Epyc motherboards?
 

nero243

Active Member
Oct 28, 2018
108
76
28
Has anyone been able to OC a Rome cpu yet?
I don't think so. There is no MSR documentation and no one seems to have unlocked the CBS menu yet. Maybe we'll have to wait for the new threadrippers, hope that these use the same MSR and some insider makes a tool for it. Maybe someone can monitor the MSR with RW-everything on a ryzen 3000 platform while ocing with the ryzenmaster software. I think the MSR has changed for the new generation Zen because they introduced the CCX overclocking options, which is substantially diffrent from Zen and Zen+.


Bios on Intel consumer motherboards often provide a setting called load-line calibration to prevent vdroop, does this setting exist (but hidden) on Epyc motherboards?
No, there are no LLC options in the bios and i don't remember to have found them in the IR controller software.
 

RageBone

Active Member
Jul 11, 2017
291
77
28
Yea, the IR Software is messy and gives you different options depending on the controller.

But there is at least a tool to do basic loadline calculation and I assume that there is also a setting inside.
I'm not sure thoughe, if there is a PMBus command related to loadline.
 

Quade

New Member
Nov 13, 2019
6
0
1
Holy cow, I just stumbled upon this thread, very interesting.
There are SM boards up for grabs locally for like 400 bucks new bulk packaging, but finding a CPU is an absolute headache.

Seeing this japanese dude has been putting batches of questionable SKUs for sale, compatibility is dodgy, and these might have been fried already by seeing it doesn't boot on their testing SM board.
Sell As-Is! AMD EPYC 7551 ES 2S1451A4VIHE4 1.44GHz 32C Socket SP3 14nm | eBay
Sell As-Is! AMD EPYC 7551 ES 2S1905A4VIHF4 1.9GHz 32C Socket SP3 | eBay

I wonder has anybody tried their luck with those?