taki0n

Thank you for the info @I.nfraR.ed and sorry for the delay for my answer. - When I was preparing my P-State possible OC, to know what were the P0 values I made a normal BIOS OC using fixed multiplier and fixed VCore and the values were almost identical to what you show us in the Example 1 picture -> Same fid=149 and did=8 but my vid matched 1.35V. In this fixed OC BIOS state the frequency was the expected and no apparently changes of any voltage lower than vCore or frequency P1/P2 changes even if I activated C&Q and other power saving options in BIOS. The OC max values where successfully applied and I had increases in performance in many stress or bench applications. I thought at that moment that voltages didn't drop (or frequencies) because of fixed multipliers/voltage numbers, so I decided to change BIOS voltage to offset auto mode and "auto" in frequency multiplier too, and my objective was to reach the prior stable OC value in the P-0 state by software using r4m0n zenstates python scripts. - Using zenstates I made a table in wich p-0 was set to OC values 3725@1.35V, and mantaining p-1 and p-2 at stock values. The result was not as expected: The "idle" state values were apparently identical to having a fixed OC in BIOS (P-0 all the time), but while monitoring values using "CoreFreq" under load, the voltage instead of going up or maintaining at 1.35V, was going down to around 1V (very mysteriously), so I abandoned idea of OC by software and wrote this post. I never tried what you did, having a DID=6 or different to ,8 because I don't know exactly what I am touching ¿what is the purpose of DID divider? Now without any OCing or P-state changing this are the voltages of the home server in almost idle and next image is under "stress-ng -c 4": Thanks!

Hi, I already made a reply, but it seems it "vanished" or needs moderator to accept it. In the case I am writing two replies, please erase one. I attach the JSON code regarding my setup just bought. OK. My particular situation is having an 2200GE "Engineering Version" and wanting to use a higher frequency P-State 0 in Debian Linux for a machine acting as a home server 24x7. I was comfortable OCing to 3.725GHz@1.35V. With previous Python application from r4m0n I was able to set this numbrers to P-0 and keep stock P-1 and P-2 numbers in Debian. The problem I have is that the voltage doesn't drop at P1 or P2, and more strangely, it apparently drops at 100% load which is contra intuitive (using an stress-ng script, it appears to use around 1V or less, I am using some small LLC compensation, so it's not vdroop) Now I have decided until someone have a better idea to not OC at all (not by software in Linux not using bios), and keep the server at 3.4GHz enabling "turbo" frequency and dropping voltage to 1.2V. Using this, it drops nicely voltage and frequency down proportionally to the load, but it is a pity I can't use the extra punch of the 3.725GHz OC using software. I hope my JSON can contribute, thanks for your work. { "AppVersion": "1.2.5.0", "OSVersion": "Microsoft Windows 10 Enterprise LTSC", "CpuId": "810F10", "CpuCodeName": "RavenRidge", "CpuName": "AMD Eng Sample: ZD2200C6M4MFB_34/26_Y", "MbVendor": "Micro-Star International Co., Ltd.", "MbName": "B450M PRO-VDH MAX (MS-7A38)", "BiosVersion": "B.40", "SmuVersion": "30.87.00", "Mailboxes": [ { "MsgAddress": "0x03B10A20", "RspAddress": "0x03B10A80", "ArgAddress": "0x03B10A88" }, { "MsgAddress": "0x03B10A24", "RspAddress": "0x03B10A84", "ArgAddress": "0x03B10A8C" } ] }