Can't get Offset working as desired

[Lazar1981]

Hello,

For a few weeks now I'm an owner of the ROG STRIX X470-F with a R5 2600X. I realized the amount of voltage that is used for the CPU being too high. I tried some BIOS settings to reduce that. My goal would be to have this managed by negative offset. For idle I got it down to 0.75 - for load I would like to have it not higher than 1.375 - but that seem to be Impossible. A max negative offset of 0.875 results more than 1.375 with load. Setting it higher makes the system unbootable.
On the other hand - when I set a fixed voltage of 1.375 everything runs fine and stable. So why can't I get the max voltage of 1.375 working with a negative offset? Maybe I forgot to set a needed BIOS option?

Hope someone can help me with this

Greets

[RedSector73]

Unsubscribe from source material.
Use AUTO, ignore what you think about voltage the chip is using being too high (it's wrong) and problem is solved.

[Lazar1981]

Thanks for your reply and sry
Where or why should I unsubscribe? I don't understand what you are saying.

[RedSector73]

That whatever or whoever told you your CPU is running too high is incorrect.

[Lazar1981]

Can you explain that more in detail? Or can you give me some sources, where I can do some more research?
Get me right... I really want to understand, what's making such a voltage necessary and in result the higher temperature especially when a lower voltage for load doesn't affect stability. I had Intel platforms for years and switched over to AMD for the first time. That's new for me

[RedSector73]

Ryzen does not work like Intel chipset. I don't know what you're trying to achieve, that is not clear from your posts, which is why not really investing into an answer.

All going to say without you showing your benchmark baseline, then how do you know that your offset isnt just gimping your CPU for no real reason. In that, Voltage tends to go higher when there isn't any load. It does not mean it is out of specification. Higher temperature (but below throttling) vs something lower will not make any difference to the life of your CPU. It never has for Intel or AMD.

[Lazar1981]

RedSector73 wrote:
Ryzen does not work like Intel chipset. I don't know what you're trying to achieve, that is not clear from your posts, which is why not really investing into an answer.

All going to say without you showing your benchmark baseline, then how your offset isnt just gimping your CPU for no real reason. Is that, Voltage tends to go higher when there isn't any load. It does not mean it is out of specification. Higher temperature (but below throttling) vs something lower will not make any difference to the life of your CPU. It never has for Intel or AMD.

Ah well... Didn't realize it is that incomprehensible. I want to get temperatures as low as possible. And in fact I got them down with lower voltages. But I wasn't aware of having no effect on lifetime of the CPU with lower temperatures. So in summary: I have cearly NO advantage by undervolting the CPU?

[AfterShock]

Lazar1981 wrote:
Ah well... Didn't realize it is that incomprehensible. I want to get temperatures as low as possible. And in fact I got them down with lower voltages. But I wasn't aware of having no effect on lifetime of the CPU with lower temperatures. So in summary: I have cearly NO advantage by undervolting the CPU?

what program are you measuring temps and voltage with and what are you using to load your cpu? I have a 3800xt that hangs around 1.45 to 1.5v under light loads like browsing forums my temps are around 50c, when I run cinebench my voltage drops to around 1.3 to 1.35 and stays in that range.

When I got my 1st ryzen cpu which was an 1800x I was worried about how high the cpu voltage was and messed around with offsets and undervolting but with ryzen chips that just leads to stability issues. 2 things I've learned over the years is that ryzen is perfectly capable of handing it's own voltage and it's own overclocking. Just set it to auto and let it do it's thing.

[RedSector73]

AfterShock wrote:
When I got my 1st ryzen cpu which was an 1800x I was worried about how high the cpu voltage was and messed around with offsets and undervolting but with ryzen chips that just leads to stability issues. 2 things I've learned over the years is that ryzen is perfectly capable of handing it's own voltage and it's own overclocking. Just set it to auto and let it do it's thing.

Yes, that is my learned experience as well. From 1800x, 2700x, 3700x, 3900x & 5950x.

Lazar1981 wrote:
Ah well... Didn't realize it is that incomprehensible. I want to get temperatures as low as possible. And in fact I got them down with lower voltages. But I wasn't aware of having no effect on lifetime of the CPU with lower temperatures. So in summary: I have cearly NO advantage by undervolting the CPU?

Then buy AIO. And leave the CPU in AUTO.

Not only a case of, no advantage, if you run CB20 test you will see the performance degradation it is causing.

[Lazar1981]

RedSector73 wrote:
Then buy AIO. And leave the CPU in AUTO.

Not only a case of, no advantage, if you run CB20 test you will see the performance degradation it is causing.

I agree, the stock cooler isn't that effective. Due to being noctua addict I ordered a NH-U12A. That will do a great job.

Concerning CB20 - I ran the test twice (with fixed 1.375 and with auto settings). Due to lack of time it was a somehow hasty process. I did it early in the morning just before going to work. The first run with 1.375v was directly after reactivating the PC from sleep mode. The second run with auto settings was after a reboot. So I don't trust the results that much. But for now it tells me, that it has 2953 points with auto settings and 2987 points with fixed voltage (not a mentionable difference but that wasn't my goal). At this point I can't see performance degredation. I will rerun this tests tonight when I'm back from work to prove the reliability.

AfterShock wrote:
what program are you measuring temps and voltage with and what are you using to load your cpu? I have a 3800xt that hangs around 1.45 to 1.5v under light loads like browsing forums my temps are around 50c, when I run cinebench my voltage drops to around 1.3 to 1.35 and stays in that range

I am using hwinfo, CPU-Z, core temp and NZXT CAM (for my grid fan control)
I can confirm the 1.3 to 1.35 volts under load with CB in auto settings. And also seeing the fast rising temps staying around the 50s as you described. So if this normal then I won't worry anymore.

Kelutrel wrote:
I also used to keep my 5900X undervolted on my C8F for the same reasons, thermals and a slightly higher top performance in MT.

The fact that some CPUs can't be undervolted even by 0.01v, as they immediately become unstable, is due to binning. I got lucky somehow and my 5900X can be undervolted by a negative 0.05v offset without issues. Depending on the load, my cpu core voltage would vary between 1.15v in idle to a peak of 1.45v during CBR20. I also usually keep all the CO offsets to -30 but the two best cores at -16.

Your desired max voltage of 1.375 seems pretty low to me for a multithreaded usage, did you verify that it was stable at that voltage under load ?

Additionally, you should consider that a negative voltage offset is applied to both the peak voltage and the lower deep sleep voltages of each core, and this is probably why your machine can't boot if you go too low with the voltage offset.
If you offset of negative 0.1v your cpu may attempt to go into a C6 ultra-low power sleep state using 0.1v instead of the normal 0.2v and be unable to bring a core back alive, while this would not happen if you manually set the CPU to a fixed 1.375v as it will never go under that voltage even during low power sleep states.
You may overcome this issue a bit by disabling the "PSS Support" and "Global C-state Control" and "DF Cstates" options that you should be able to find in your BIOS.

Also, a negative offset of 0.875 that you mentioned would make no sense because if you undervolt by 0.875v a cpu that usually runs at 1.5v you would be trying to run it at 0.625v and that looks quite improbable.

However, I have to confirm that with the latest BIOS versions the AGESA management of the voltages improved quite a lot, and currently with the latest BIOS I get no differences in thermals or performances by udervolting the CPU, it improved so much that AGESA seems to be now able to keep the voltages naturally lower on my CPU without any BIOS setting.

Thank you very much - that's the explanation I was seeking for.
I did some stability tests (3DMark, Cinebench, Prime95) - everything ran fine - but maybe I didn't test enough


To all: thx again. And again get me right - it is not that I don't believe what you are saying, but I have to prove it. I am new with AMD (used a K6 years ago - that was my last contact with AMD). I want to understand what I have here and what's the best way to handle everything. So thanks for all your replies. I will dig deeper into that and hope to get a little more knowledge about this. 🙂