Finally: Overclocking EPYC Rome ES

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

heavyarms2112

Member
Feb 11, 2022
52
11
8
Perfect. Stupid question - how do I work out the decimal values in terms of voltage? :)
use the zenstates.py script to list out the factory p-states.
Code:
CPUs: 1
CPUID: 00A20F10
Package Type: 2
P0 - Enabled - FID = 94 - DID = 8 - VID = 48 - Ratio = 37.00 - vCore = 1.10000
P1 - Enabled - FID = 8C - DID = A - VID = 58 - Ratio = 28.00 - vCore = 1.00000
P2 - Enabled - FID = 84 - DID = C - VID = 68 - Ratio = 22.00 - vCore = 0.90000
example. VID here is 48 translates to 1.1v and VID of 58 translates to 1.0v.
hex values are 0 to 15 (16 in total) so 48 to 58 has 16 steps. So 0.1v (1.1v - 1v) is 16 steps (58 - 48) == 6.25mV/step.

So if 48 is 1.1v, 50 would be 1.1v + 0.625*2 = 1.1125v

there was also a modified python script (in this thread) that you can use which would also print out what's mV being set instead of printing out hex.
 

mirrormax

Active Member
Apr 10, 2020
225
83
28
50 is 1.05
Formula is -X/0.00625+248
Then decimal to hex.
=DEC2HEX(-X/0.00625+248)
Works in sheets
 

EagerToLearn

Active Member
Feb 25, 2020
115
25
28
Regarding the EPYC Rome overclocking: A while back, smb. said, that the shown clock frequencies are not the actuall ones. So I am wondering: When I use the overclocking tool and the taskmanager shows 3,45GHz, is that the actual frequency?
 

Spartus

Active Member
Mar 28, 2012
323
121
43
Toronto, Canada
Regarding the EPYC Rome overclocking: A while back, smb. said, that the shown clock frequencies are not the actuall ones. So I am wondering: When I use the overclocking tool and the taskmanager shows 3,45GHz, is that the actual frequency?
Task manager isn't even reliable anyways, it doesn't appreciate that cores can run at different speeds. I would use HWinfo64 to inspect individual clock speeds, but then one must also consider clock stretching.



Clock Stretching is a safety feature that is built into all AMD Ryzen CPUs. When the CPU thinks the actual voltage is too low to sustain a stable system at a given frequency, it will reduce the clock period until the voltage is back at the acceptable level.


The result of clock stretching is that while the system will report the Core Frequency as usual, the effective clock will be lower. So, since a lower amount of clock cycles occur the performance will also be lower.


Lastly there was some issue about an clock speed offset or what was it, 800 MHz?, for the 64 core SKUs with overclocking using this tool, but I actually only have 32 core models so no experience with that.
 
  • Like
Reactions: EagerToLearn

EagerToLearn

Active Member
Feb 25, 2020
115
25
28
Thank you very much, for that detailed explanation. Is there an easier way to detect Clock Stretching than suggested in the link?
 

RolloZ170

Well-Known Member
Apr 24, 2016
5,159
1,549
113
Is there an easier way to detect Clock Stretching than suggested in the link?
you have to draw a clock cb-score curve chart.
if the clock rises but the cb-score stucks, the core is dropping clocks.
this behavior was also reported by der8auer, ifsome part the ALU is not ready the core drops clocks to wait for completeness operations.
in addition to this: the core clock is made by the BLCK * multiplicator - but the multiplication unit is a complex circuit and can not generate all thinkable clocks.
 
  • Like
Reactions: heavyarms2112

mirrormax

Active Member
Apr 10, 2020
225
83
28
Yea when overclocking theres lots of wierd heaviours, like if you push it too high it starts suddenly performing worse, and only way to recover would be a reboot. displayed mhz when pushed too high does not reflect the performance, same clock usually performs better the more voltage you can manage to give it but VRM overheating will very quickly limit your OC. even with good vrm cooling you might trip the board OCP causing hard shutdowns etc

highest ive ever managed to run a dual 64c is 32.5 multi 1.18v and that got me to 100k cb23, but that was with watercooling the vrm+cpus and i think the system load was around 1000w
 
  • Like
Reactions: RolloZ170

Wilhelm

New Member
Aug 1, 2019
13
0
1
I just upgraded to 7.2 this week, and I just checked and am seeing the same issue. I don't see any errors, just nothing happens. I guess I shouldn't be quick to upgrade the other box. :(
anyone found a solution to this problem yet?
 

mirrormax

Active Member
Apr 10, 2020
225
83
28
anyone found a solution to this problem yet?
have you tried using the commands manually as root to see if they work? i think i resorted to that at one point when i couldnt get the scripts to work.(not using promox but a newer linux kernel)