F5F Stay Refreshed Power Users Overclocking No, there is no feasible method to undervolt EPYC CPUs, especially Zen2 variants.

No, there is no feasible method to undervolt EPYC CPUs, especially Zen2 variants.

No, there is no feasible method to undervolt EPYC CPUs, especially Zen2 variants.

K
kenken0724
Member
178
12-29-2022, 10:54 AM
#1
Hello, I've put in a lot of effort to reach where I am today with these components...
- H11DSi (supermicro mobo)
- 2x AMD EPYC 7k62
- 16x 8GB RAM at 2933MHz
- A clean Linux live server OS (43470)
Working well, but I'm looking for ways to keep things short. Can I possibly lower the voltages? Not just by changing power states—I want to maintain performance close to current levels and try reducing power use. It seems a good balance between performance and efficiency is possible, but I might be misusing something. Are there other options I can explore? Since we're discussing server CPUs, I'm not sure what the experts say, so I hope you can help.
My main goal is to maximize performance per watt. If this isn't the right forum, let me know. There are many helpful discussions online, but I'm not sure where to start. Undecided
K
kenken0724
12-29-2022, 10:54 AM #1

Hello, I've put in a lot of effort to reach where I am today with these components...
- H11DSi (supermicro mobo)
- 2x AMD EPYC 7k62
- 16x 8GB RAM at 2933MHz
- A clean Linux live server OS (43470)
Working well, but I'm looking for ways to keep things short. Can I possibly lower the voltages? Not just by changing power states—I want to maintain performance close to current levels and try reducing power use. It seems a good balance between performance and efficiency is possible, but I might be misusing something. Are there other options I can explore? Since we're discussing server CPUs, I'm not sure what the experts say, so I hope you can help.
My main goal is to maximize performance per watt. If this isn't the right forum, let me know. There are many helpful discussions online, but I'm not sure where to start. Undecided

L
Leqionaire
Junior Member
4
12-30-2022, 07:23 PM
#2
Undervolting might lead to certain programs not receiving sufficient voltage to operate correctly, wouldn't you think? Additionally, these processors are already optimized for maximum efficiency, meaning even minor adjustments could trigger problems.
L
Leqionaire
12-30-2022, 07:23 PM #2

Undervolting might lead to certain programs not receiving sufficient voltage to operate correctly, wouldn't you think? Additionally, these processors are already optimized for maximum efficiency, meaning even minor adjustments could trigger problems.

B
BlackAvenuee
Junior Member
19
01-05-2023, 07:16 AM
#3
But I want to clarify that in cases where the applications are extremely non-essential—such as mining—I would like to test and try anyway.
B
BlackAvenuee
01-05-2023, 07:16 AM #3

But I want to clarify that in cases where the applications are extremely non-essential—such as mining—I would like to test and try anyway.

D
Dude76258
Member
112
01-06-2023, 08:03 AM
#4
Searching online indicates that very few, perhaps just a couple, Threadripper motherboards offer adjustments for this. A review of your manual shows it doesn’t appear to support such changes.
D
Dude76258
01-06-2023, 08:03 AM #4

Searching online indicates that very few, perhaps just a couple, Threadripper motherboards offer adjustments for this. A review of your manual shows it doesn’t appear to support such changes.

R
r4tz
Member
51
01-06-2023, 01:01 PM
#5
Thank you for avoiding the same manual repeatedly. I’m not aiming to disable cores or significantly hurt performance. Removing core perf boost proved beneficial since the energy saved was much greater than the performance drop, so I was hoping for another way to improve efficiency. And according to what I know, cTDP can only be reduced to 200 W at most, which is still higher than the power usage of these CPUs without core perf boost.
R
r4tz
01-06-2023, 01:01 PM #5

Thank you for avoiding the same manual repeatedly. I’m not aiming to disable cores or significantly hurt performance. Removing core perf boost proved beneficial since the energy saved was much greater than the performance drop, so I was hoping for another way to improve efficiency. And according to what I know, cTDP can only be reduced to 200 W at most, which is still higher than the power usage of these CPUs without core perf boost.