CyrIng

In the Log there is no evidence linked to CoreFreq but I would suggest you plug a monitor on your server VGA and post that screen photo when server becomes stuck.

@ich777 Hello, Using different command options or key=value attributes found in the manual of ttyd v1.7.4 ; running corefreq-cli in various browsers chromium, brave, firefox, Edge from local ArchLinux or remote Windows-10 ; I am not able to reproduce the UI glitch issue found by @wuudogg Edit: seen also in screenshots by @KaitsuN6

I'm trying from a Telnet session under Putty for Windows-10 to an ArchLinux Telnet Server. Whatever the mode is, UTF or ASCII, I can't reproduce the UI issue; no glitch in the CLI @ich777 Could you please tell what is the underlying software stack of the Unraid Terminal GUI ?

Thanks. I'm thinking about ordering one of those. Your NAS is highly tuned. I keep those in mind as I have to migrate an old Windows 7 HTPC to a Linux mixed solution: Koodi + Raid-5 Tricky part is to mount the Intel RST Raid disks from Linux. About the UI glitches, I don't catch how to reproduce them. My understanding is that Windows and Virtualization are the starting points of the issue.

@wuudogg There's an odd issue with the UI I'm noticing in your previous screenshot: the borders of the Settings window are missing on the sides and get width extended on the above. I can't tell yet if it is a `corefreq-cli` bug or some side effects of the employed screenshot software (SW) tool ? Are those issues visibly permanent on your screen ? By which mean are you displaying the UI ? ssh or other remote SW ? X or console Terminal SW ? Is the term color theme modified or default ?

That's such an improvement. How did you measure the whole power consumed ?

Short answer is you need these module parameters: Register_ClockSource=1 Register_Governor=1 Register_CPU_Idle=1 Register_CPU_Freq=1 Apply the Linux standards to add them as an `options` line into `/etc/modprobe.conf` or `/etc/modprobe.d/modprobe.conf` Somehow, you will also switch the System to CoreFreq Clock Source: this can be achieved in system `profile` or `init` file echo "corefreq_tsc" > /sys/devices/system/clocksource/clocksource0/current_clocksource Fyi, you can get all possible parameters using command: modinfo corefreqk.ko

Check BIOS if CC6 and PC6 are enabled Next Kernel should idle with an assembly instruction like `I/O-Wait` or `halt` ( `M-WAIT` being less efficient with Ryzen than Intel C-States ) This what you should get as a default, even using `amd_pstate` Here's the 3950X power consumption

@Aubury CPU-Freq driver [ amd-pstate] Governor [ conservative] CPU-Idle driver [ acpi_idle] You can see here that mainline drivers are already occupying the position You have first to unload `acpi_idle` and `amd-pstate` With you Unraid setup, you can try: modprobe -r acpi_idle modprobe -r amd-pstate But I doubt kernel will allow to unload once they have already been registered. In that case you have to blacklist them when booting Linux by adding these parameters in the boot command line (See grub; syslinux; systemd-boot) modprobe.blacklist=acpi_cpufreq amd_pstate.shared_mem=0 amd_pstate=disable idle=halt When starting CoreFreq, you should then read `Missing` like my screenshot Thus you are free to register CoreFreq in that order: Clock Source (optionally) Governor CPU-Idle driver CPU-Freq driver If successful, you will get this second screenshot

@KaitsuN6Good day Do you find now CoreFreq stable for production use ?

8.96W we are not far from the computed TDP TDP[ 10 W]

@KaitsuN6Power now appears more "realistic" Using my Conic stress functions which are pushing on FPU, you should read the max wattage. To stop my stress tools, press a few times the F10 key Edit: as a bonus, you are getting the DRAM watt

Don't worry; I have to refactor this code anyway. But it will be long to implement IMC through a series of UNRAID plugin packages. I have to experiment registers and for this I need to work directly on hardware. @ich777 Hello, At your disposal, I have committed in `develop` branch a tiny change to the Goldmont Power formula. May you please prepare a plugin for a @KaitsuN6 test.

@KaitsuN6 It's possible that the failing access to the MCHBAR (Memory Controller) lefts CoreFreq in unstable situation. This is the major difference I see with previous pluggin.

|- Units |- Power watt [ 0.000003906] |- Energy joule [ 0.000000061] |- Window second [ 0.000976562] With Goldmont I'm facing Power units unseen before

If it happens please grab the log files About Power, values shown as Pkg and Cores seem wrong; unrelated with the stressed load

This confirmes there is a single voltage sensor in this processor architecture About the Power, are you reading some meaningful values ? Example from the view "Power" :

Now you can select Single Core stress, Random or Round Robbin

@KaitsuN6You can change Voltage scope to SMT then apply individual CPU load and tell if Vcore is a discrete value or not ?

As default there is a single sensor for whole processor. You can try other Voltage scope in window "Settings" Also we have to find/work on Northbridge or System Agent or SoC voltage

@KaitsuN6I like to see this Vcore. Will study IMC data. Thank you EDIT: Unfortunately MCH failed to remap.

Please let me know when I can get the memory registers dump as I have a short delay window to program it. Sincerely

I was sure Rank is 2 I have a small C program to read/write registers: `zencli` source code here Let me know if you can compile it directly on your Linux ? Just download `zencli.c` and compile cc zencli.c -o zencli Once done, as root, you can read some Ryzen registers. Two addresses I need is about DimCfg specified above. Enter commands like this: ./zencli smu 0x50090 ./zencli smu 0x50094 ./zencli smu 0x50040 ./zencli smu 0x50044 ./zencli smu 0x50100 ./zencli smu 0x50104 ./zencli smu 0x50df0 ./zencli smu 0x50df4 After try a Memory Controller dump ./zencli umc

Can you please provide the output of dmidecode ? sudo dmidecode -t memory At least we should read from it: Total Width: ?? bits Data Width: ?? bits Size: ??? GB Locator: xxx Bank Locator: xxx Type: DDR? Speed: 0000 MT/s Part Number: xxx Rank: ?

I see the issue. I have based the Rank number on OnDimmMirror But this field only applies to DDR4 Edit: because processor is 7900X3D, maybe Dram3DS is a discriminant to guess Rank