Closed olejon closed 4 years ago
More than 195W is scarry and with 82°C, I wonder when thermal throatling is going to engage ?
I'm gonna review the driver to see if I had set a thermal sticky bit into code, which translates into HOT
being highlighted.
HOT
, but there's no way to avoid that and they're built that way, and BIOS set to throttle instead of shut down for CPU protectionThanks for your temperature calibration effort. Can't wait to see your results.
Meanwhile I've checked code and driver has no ThermalTrip query in families 17h & 18h.
Thus no HOT
will appear with Zen processors.
Perhaps kernel is logging it.
I will search for such thermal status register which may be queried onto SMU
Many bullet points. Please read them all.
k10temp
"HIGH" is 70°C, BUT forget that because:Limit 95°C
, as said before. Goes for all the Gen 3 CPUsclassroom
and the system was totally stable395W
, and 50% of that is 197.5W
as in Screenshot 2dmesg
, journalctl
, sensors
(which shows an alarm/alert for certain HW) and used grep
on ALL files in /var/log
and there was nothing matching any relevant keywordclassroom
finished faster on Windows as well, but result isn't valid as I had Ryzen Master open + I took a screenshot during the runVery interesting investigation you did, although I will go very slowly on those settings t'il we have a clear understanding on what is going on, and their side effects. I'm still not sure if hardware is well protected against over-voltage, over-consumption and so on. Don't fry your brand new Ryzen !
We find wanted registers a bit every-where, sometimes through :
ProcHot, ThermalTrip : on SMIstatus, SmiControl5 & Power Management (PM) Registers
C1e : PMx07E [CStateEn]
IdleCount (aka "idle count from the latest monitored period") at MISCx060 [IdleCntrl] (FCH::MISC::IdleCntrl)
. Is this what RM called Sleep from ?
Have a rest, you deserve it !
Don't fry your brand new Ryzen !
scenes
, more stress regular usage will ever do78-85°C
173-203
victor
took very long to load ("Warm up"), it seemed like it tried several times so I was on alert, but CoreFreq showed something was going on at least, and it did load and logical as it was the definitely longest one, but Watts during dropped to the lowest during that at times, to 173 Watts
203 Watts
only seen during the first 2 ones: bmw27
and classroom
, but never above 85°C
during those86-87°C
it only took 1-2 seconds to drop back down to 85°C
pavillion_barcelona
- I think, if not koro
- it managed to push it to 87-88°C
, but not above. Watts was well under 200 then85°C
and around 200 Watts
, so I suspect it was because they load quickly and maybe didn't allow the CPU to cool down a little during meanwhile, maybeGear Down
which should be totally safe, and another Power Saving setting for RAM which is a sub-menu under that part againD.C.O.P. = Enabled
or what?This does NOT seem to discuss that setting but rather manual overclocking in the AI Tweaker
part of the UEFI, although maybe interesting for you? Remember to see the 2 screenshots below too.
"OC mode" is automatically engaged when P0 frequency is raised above base frequency. In OC mode, the SMU inside Ryzen CPU is disabled, which means sophisticated power management and power saving features are not functional. The presense of OC mode can be verified by calling GetCurrentOCMode() via the RyzenMaster SDK or code 0C via a Port 80 debugger card. HWiNFO has 3 voltage readings related to CPU. If it's individual core VID or motherboard Vcore that is scaling down to 0.2 Volts, then that's expected. What's important is "CPU Core Voltage (SVI2 TFN)". I think this value can tell whether the CPU has gone into "OC mode" or not. If this value is not scaling down, then it most likely means OC mode is engaged
- Read the above first
I did but I don't feel comfortable with the power consumed.
173-203
are far too high to my taste
See my way in next post
- I can only find ASRock documenting "Uncore OC Mode" and it seems to have to do with RAM and Infinity Fabric only
I'm not familiar with the Zen Uncore to tweak it yet. Especially its voltage limit.
- I've not touched any of these manual settings, just turned off
Gear Down
which should be totally safe, and another Power Saving setting for RAM which is a sub-menu under that part again
I've also noticed this one and I don't know what's behind the hood. For now I have stabilized my DRAM at its OC factory mode and it will be easier to understand those extra settings when I will have a clear understanding of how the Ryzen Memory controller manages the [sub]-timings . So far, I have no specs how to query the configured DDR speed, geometry, latency, and so on
- Still system performs better. So is it simply the fact that it now actually uses my RAM to the max as
D.C.O.P. = Enabled
or what?- All people discussing this setting also do manual overclocking (entering values manually) but I don't
I have chosen D.C.O.P(XMP) to auto configure the G.Skill kit.
"OC mode" is automatically engaged when P0 frequency is raised above base frequency. In OC mode, the SMU inside Ryzen CPU is disabled, which means sophisticated power management and power saving features are not functional. The presense of OC mode can be verified by calling GetCurrentOCMode() via the RyzenMaster SDK or code 0C via a Port 80 debugger card. HWiNFO has 3 voltage readings related to CPU. If it's individual core VID or motherboard Vcore that is scaling down to 0.2 Volts, then that's expected. What's important is "CPU Core Voltage (SVI2 TFN)". I think this value can tell whether the CPU has gone into "OC mode" or not. If this value is not scaling down, then it most likely means OC mode is engaged
Screenshots from 2 Manuals for ASRock MBs. Not much info:
- PS: I've NOT touched this voltage!
Those Windows tools are so lucky to get so many specs details from Manufacturers I'm trying to stick to Linux and to offer an Open Source software. Road is long.
fishy_cat
that causes the highest temperature. See if you can replicate that, no matter settingsTemperature / Watts / MHz
89°C
, BUT it never did, as in Linux, reached 87-88°C during fishy_cat
, other than that at 80-85°C3800 MHz
to a little above 4000 MHz
, not massive but clearly better resultsD.C.O.P.
and not touched any of the values it sets (UEFI shows what you have changed when you choose "Save & Reset")Precision Boost Override (PBO)
AMD Overclocking
AI Tweaker > Precision Boost Override
is what makes the differenceAuto
to Enabled
+ 100 MHz
, the same as in the other, and BTW Ryzen Master still shows PBO is 100 despite setting it both placesPPT
, TDC
and EDC
, from 142 W
to 395 W
, NOT that it uses that, it uses 49-50% of that:Auto
Enabled
and 100MHz
Auto
as in Screenshot 1, a Blender Benchmark causes PPT
, TDC
and EDC
to go to Orange Color, almost 100%victor
as it is the longest running one, but also the one with Lowest Temperature, actually:[AUTO]
beside a fewD.C.O.P
standard with XMP
but Memory Frequency set to 3600MHz
Applying a voltage offset decrease to use the minimum juice when CPU is idle but 4.7GHz
single Core still capable
As automatically set by XMP, DRAM voltage is kept at 1.35V
Using Conic Compute... > Hyperboloid of two sheets
which is the most computing algorithm, I'm getting ~ 145W
Package with a max Temp of 67°C
at 1.3V
BIOS: PBO disabled, XFR disabled, CPB enabled
SMT available
DDR timings: Beside Manufacturer's 16-16-16-36 values, all set to [AUTO]
CPU Power Thermal Control set by first POST. I don't know what's the purpose of this setting ?
4.7GHz
at 1.5V
1V
That came quick! Remember to see the one I just commented.
Then we only have 2 differences in UEFI:
CPU Core Voltage Offset
. No manually entered value at all100 MHz
both placesBoost Override CPU = 100
and OC Mode = Auto Overclocking
, which sets the UEFI AMD Overclocking
setting automatically, but not the one in AI Tweaker
, which for you is called Extreme Tweaker
(same settings though)CoreFreq > Voltage
, while I get for all of themHyperboloid of two sheets
, while I get a slightly better for Single Core
, and both are totally stable in MHz4065 MHz
(think it varies a little depending on scene)1V
- 33°C with Ambient 27°CHow CoreFreq is reading a Vcore of 0.2V
: is it a voltage you set in BIOS ?
You mean the one at bottom right? Saying V[0.20]
. Is that bad or anything?
I'm not into Voltage tuning AT ALL.
I haven't set ANY integer or string manually, only Drop-Downs. Only the D.C.O.P. changes automatically what you already know:
Yours show V[0.91]
. What's the diff?
You mean the one at bottom right? Saying
V[0.20]
. Is that bad or anything?I'm not into Voltage tuning AT ALL.
I haven't set ANY integer or string manually, only Drop-Downs. Only the D.C.O.P. changes automatically what you already know:
- RAM voltage to 1.35
- RAM timings, for me 16-18-18-36
- RAM frequency for me 3200 MHz
Yours show
V[0.91]
. What's the diff?
OK, so it's a CoreFreq bug Adding it to the todo list
Uncore OC Mode
enabled?Hmm, look at previous IDLE Screenshot:
https://github.com/cyring/CoreFreq/issues/186#issuecomment-646917760
V[0.91]
Uncore OC Mode
+ AI Tweaker PBO
which wasn't on when that screenshot was takenVcore is restored to [AUTO] which means that Processor posts at less than 1V
BOOST
register is confirmed OFF
To answer to your previous question: Voltage Core scope is changed to Package in Settings menu
42
for all CPUs and stress my best Core
boosted
ratio of 47
4.8GHz
frequency appears very shortly but fallback immediately to 4.7GHz
Like something is monitoring against this P-State changed.
Interesting isn't it !I suppose :man_shrugging: :smile:
0.91V
as yours before enabling Uncore OC Mode
0.2V
, which Value in Ryzen Master should I look at? Tell me the name from the Screenshots of Ryzen MasterDifferences:
Uncore OC Mode Enabled
to get same Blender Benchmark results)I suppose
- You're using the CoreFreq CPU driver, no?
Yes all P-State tweaks done with CoreFreq. But I've not programmed a governor for AMD yet. Thus OSPM is under the control of the Linux ACPI module.
- You have CPB 2 places in your UEFI? Maybe I do too but haven't looked, just found it 1 place, but I know it hasn't been touched (tried disabling it once as said and everything went slower indeed)
Yes, the Crosshair is also showing this CPB issue, 2 places for the same name. It's confusing.
- BTW, as said my Voltage at Idle was
0.91V
as yours before enablingUncore OC Mode
- When it now says
0.2V
, which Value in Ryzen Master should I look at? Tell me the name from the Screenshots of Ryzen Master
I believe it is called CPU Voltage under Voltage Control
Differences:
- You have faster RAM
- Your MB is same brand but costs twice as much so it would be weird if you don't get better results than me with the same settings (you do, I must have
Uncore OC Mode Enabled
to get same Blender Benchmark results)- Newer kernel (my experience = better performance with every version for Ryzen Gen 3)
- As said you use the CoreFreq CPU driver(?)
Perhaps but the Processor P-States remain what they are, whatever the bells and whistles of the motherboard. My goal here is to control the P-States: next steps will be to set some manual frequency ratios in BIOS then dump the registers changes within CoreFreq driver to understand how BIOS is stabilizing frequencies...
You mean the one at bottom right? Saying
V[0.20]
. Is that bad or anything? I'm not into Voltage tuning AT ALL. I haven't set ANY integer or string manually, only Drop-Downs. Only the D.C.O.P. changes automatically what you already know:
- RAM voltage to 1.35
- RAM timings, for me 16-18-18-36
- RAM frequency for me 3200 MHz
Yours show
V[0.91]
. What's the diff?OK, so it's a CoreFreq bug Adding it to the todo list
Checking code, threshold is 0.15V
and it applies only to the lowest voltage when reading falls down to zero.
Also threshold is not applied to current value (middle column)
So it might not a bug and the voltage has been read at 0.2
for some reasons.
https://github.com/cyring/CoreFreq/blob/ab750ba6df881c4a8e959797a23689800f54ce19/corefreq.h#L492
Uncore OC Mode
and the AI Tweaker > PBO
option and recheck RM and CoreFreq (surely goes back to 0.91V at idle)best turbo/overclocking headroom
, using only 1 of the fans0.2V
in when Idle, although, since it's Windows, it goes much more up and down, but definitely in the 0.2V
area many times during 1 minuteUncore OC Mode = Disabled
and AI Tweaker > PBO = All Auto
, CoreFreq again shows V[0.91]
Conic Compute > Hyperboloid of two sheets
doesn't even manage to bring the CPU Temperature to 70°C (stays in the 60's)The 0.20V
screenshot was a "lucky one":
0.91V
2 0.96V
3 0.20V
So it's kind of like Ryzen Master, only that RM shows way more levels between 1-3 (and many more decimals, not that that matters).
I'm back to the OC mode. One can set a maximum temperature in AI Tweaker > Precision Boost Override
too say avoid it never reaches 90°C, which I guess is when the Orange Color shows, but I have never reached above 89°C, as said, and that was just briefly before down again to 87-88, and only during Blender Benchmark fishy_cat
.
I tend to check the ASUS website if there is a BIOS update. Saw there was one from June 17 published. Probably one for you too?
journalctl
shows the IRQ
stuff in Redjuni 24 00:52:21 tux kernel: do_IRQ: 1.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #2
juni 24 00:52:21 tux kernel: do_IRQ: 2.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #3
juni 24 00:52:21 tux kernel: do_IRQ: 3.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #4
juni 24 00:52:21 tux kernel: do_IRQ: 4.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #5
juni 24 00:52:21 tux kernel: do_IRQ: 5.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #6
juni 24 00:52:21 tux kernel: do_IRQ: 6.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #7
juni 24 00:52:21 tux kernel: do_IRQ: 7.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #8
juni 24 00:52:21 tux kernel: do_IRQ: 8.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #9
juni 24 00:52:21 tux kernel: do_IRQ: 9.55 No irq handler for vector
juni 24 00:52:21 tux kernel: #10
juni 24 00:52:21 tux kernel: do_IRQ: 10.55 No irq handler for vector
...
juni 24 00:52:21 tux kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PCI0.GPP8._DSM], AE_ALREADY_EXISTS (20190703/dswload2-327)
juni 24 00:52:21 tux kernel: ACPI Error: AE_ALREADY_EXISTS, During name lookup/catalog (20190703/psobject-221)
PRIME X570-PRO BIOS 2203
Update AMD AM4 AGESA PI 1.0.0.1
I tend to check the ASUS website if there is a BIOS update. Saw there was one from June 17 published. Probably one for you too?
Thanks for the tips
- After installing it, Linux boots fine, but see these and
journalctl
shows theIRQ
stuff in Redjuni 24 00:52:21 tux kernel: do_IRQ: 1.55 No irq handler for vector juni 24 00:52:21 tux kernel: #2 juni 24 00:52:21 tux kernel: do_IRQ: 2.55 No irq handler for vector juni 24 00:52:21 tux kernel: #3 juni 24 00:52:21 tux kernel: do_IRQ: 3.55 No irq handler for vector juni 24 00:52:21 tux kernel: #4 juni 24 00:52:21 tux kernel: do_IRQ: 4.55 No irq handler for vector juni 24 00:52:21 tux kernel: #5 juni 24 00:52:21 tux kernel: do_IRQ: 5.55 No irq handler for vector juni 24 00:52:21 tux kernel: #6 juni 24 00:52:21 tux kernel: do_IRQ: 6.55 No irq handler for vector juni 24 00:52:21 tux kernel: #7 juni 24 00:52:21 tux kernel: do_IRQ: 7.55 No irq handler for vector juni 24 00:52:21 tux kernel: #8 juni 24 00:52:21 tux kernel: do_IRQ: 8.55 No irq handler for vector juni 24 00:52:21 tux kernel: #9 juni 24 00:52:21 tux kernel: do_IRQ: 9.55 No irq handler for vector juni 24 00:52:21 tux kernel: #10 juni 24 00:52:21 tux kernel: do_IRQ: 10.55 No irq handler for vector ... juni 24 00:52:21 tux kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PCI0.GPP8._DSM], AE_ALREADY_EXISTS (20190703/dswload2-327) juni 24 00:52:21 tux kernel: ACPI Error: AE_ALREADY_EXISTS, During name lookup/catalog (20190703/psobject-221)
- Any ideas?
- The BIOS update seems to be AMD only stuff. You know what this is?
- After install it said it was updating the "Led firmware" and nothing more before booting
- Changelog:
PRIME X570-PRO BIOS 2203 Update AMD AM4 AGESA PI 1.0.0.1
Have you clear the CMOS right after the update ? If not, you have garbage values in the nvram
The BIOS always resets to default values after an update and I go through all the settings again and set them as before (and check if new ones), instead of loading a saved profile.
Hm. Maybe a total power off then (PSU too). If not, does a hard CMOS reset ("short circuit" pins) do more than resetting the BIOS in UEFI, really? Fixing NVRAM shit is typically a Mac thing. Think many knows Option + Command + P + R
lol (and the SMC reset one).
Had something similar, don't remember when, if was switching to Ryzen or the AMD GPU, PCIe3 to 4 NVMes... That was fixed quickly by an Ubuntu update.
The BIOS always resets to default values after an update and I go through all the settings again and set them as before (and check if new ones), instead of loading a saved profile.
Hm. Maybe a total power off then (PSU too). If not, does a hard CMOS reset ("short circuit" pins) do more than resetting the BIOS in UEFI, really? Fixing NVRAM shit is typically a Mac thing. Think many knows
Option + Command + P + R
lol (and the SMC reset one).Had something similar, don't remember when, if was switching to Ryzen or the AMD GPU, PCIe3 to 4 NVMes... That was fixed quickly by an Ubuntu update.
It still written today in the ROG manual to clear the BIOS RAM after an update. To my knowledge, nothing is automated. A jumper should be on board ?
Load optimized defaults (F5)
Update successful. System will Reset!
(and must set all settings again, or load previous profile but I never use older profiles on new BIOS)juni 24 04:02:08 tux kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PCI0.GPP8._DSM], AE_ALREADY_EXISTS (20190703/dswload2-327)
juni 24 04:02:08 tux kernel: ACPI: [Firmware Bug]: BIOS _OSI(Linux) query ignored
Some ACPI Device Specific Method _DSM
linked to GPP8
in System Bus _SB
Search for GPP8
in whole /sys/devices/LNXSYSTM
Mine is located at :
cat /sys/devices/LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/device:0f/path
\_SB_.PCI0.GPP8
Nothing...
/sys/devices/LNXSYSTM:00# find ./ -name "*GPP8*"
Suppose there's no point in reinstalling the linux-firmware amd-microcode
packages. Will check with Ubuntu 20.04 Live USB if also there. Just in case.
Think my MB has BIOS Flashback to revert, but I've never used such a feature. Next time I'll google properly before updating. The April BIOS was good.
Did you understand the changelog? Have you installed it successfully?
Nothing...
/sys/devices/LNXSYSTM:00# find ./ -name "*GPP8*"
Suppose there's no point in reinstalling the
linux-firmware amd-microcode
packages. Will check with Ubuntu 20.04 Live USB if also there. Just in case.Think my MB has BIOS Flashback to revert, but I've never used such a feature. Next time I'll google properly before updating. The April BIOS was good.
Did you understand the changelog? Have you installed it successfully?
I won't install any BIOS. All features are working fine
Since Zen has been introduced, ACPI tables have been an issue. See the kernel history....
ROG CROSSHAIR VIII HERO had been released with:
BIOS 1201 11/18/2019
but it has however a remaining log trace ...
ACPI: [Firmware Bug]: BIOS _OSI(Linux) query ignored
ACPI BIOS Error (bug): AE_AML_PACKAGE_LIMIT, Index (0x000000007) is beyond end of object (length 0x6) (20200326/exoparg2-393)
ccp 0000:0c:00.1: ccp: unable to access the device: you might be running a broken BIOS.
... which is fine for CoreFreq because I'm working on my own [P][C]-States sub-drivers for Zen
My EFI entry tries to blacklist any of them:
title Arch Linux
linux /EFI/Linux/vmlinuz-linux
initrd /EFI/Linux/amd-ucode.img
initrd /EFI/Linux/initramfs-linux.img
options root=/dev/disk/by-label/root rw quiet break=n add_efi_memmap nmi_watchdog=0 selinux=0 loglevel=3 rd.systemd.show_status=auto rd.udev.log-priority=3 consoleblank=0 vt.color=0x03 modprobe.blacklist=pcspkr,nouveau,k10temp,acpi_cpufreq idle=halt cpu0_hotplug audit=0
which results in CoreFreq:
Linux:
|- Release [5.7.4-arch1-1]
|- Version [#1 SMP PREEMPT Thu, 18 Jun 2020 16:01:07 +0000]
|- Machine [x86_64]
Memory:
|- Total RAM 32856504 KB
|- Shared RAM 117756 KB
|- Free RAM 30792460 KB
|- Buffer RAM 79832 KB
|- Total High 0 KB
|- Free High 0 KB
CPU-Freq driver [ Missing]
Governor [ Missing]
CPU-Idle driver [ Missing]
Missing
being the sign that nothing is in control, letting me choose Target P-States
For example, decrease from default 35
to 22
but put back CPU #12
to 35
But target ratios are still moving on their own, resulting in
CPU #12
= 36
and its SMT CPU #28
jumping between 43
, 36
other CPUs are also not fully in control.
I believe its part of the collaborative stuff between Processor and some software, feeding back to it the requested Target. So far studying those entry functions: acpi_cpufreq_fast_switch , acpi_idle_enter and cppc_cpufreq_set_target
I'm noticing this strange behavior
with acpi_cpufreq
kmod blacklist and thus no cppc enabled (to my understanding)
CoreFreq started with AutoClock=0
argument to read a BCLK of 100MHz
based on the latest develop
of CoreFreq which has been changed to display 3 items in the Target selector:
The frequencies plan as follow:
|- Base Clock [100.000]
|- Frequency (MHz) Ratio
Min 400.00 < 4 >
Max 3500.00 < 35 >
|- Factory [100.000]
3500 [ 35 ]
|- Performance
|- OSPM
TGT 3500.00 < 35 >
|- Turbo Boost [ UNLOCK]
XFR 4700.00 [ 47 ]
CPB 4600.00 [ 46 ]
1C 2800.00 < 28 >
2C 2200.00 < 22 >
>>
meaning 3 enabled Ratio P-States: P0 = 35
, P1 = 28
and P2 = 22
Now stressing one Core, at least one physical, preferably the best one which is CPU#0
P2 = 22
CPU#16
, to ratio P1 = 28
>>
the obtained frequency is 2.8GHz
CPU#16
to P0 = 35
>>
it triggers Turbo Boost ! It's a strange behavior which mimics the Turbo conditions found in the Intel Nehalem architecture
Luckily it was easy to revert back to BIOS 1407
(April). It did enhance performance somewhat (not reverting but upgrading to that one from the previous).
Load optimized defaults (F5) in UEFI
and save.CAP
fileUpdate successful, System will Reset!
1005
first, to fix PBO really not doing anything on many Ryzen 3rd Gen CPUsk10temp
reading the temperatures, but that I don't know as that update came more or less at the same time as a BIOS updatesystemd-fstrim
working finally, after not working the first half year or so (and can't be trimmed on Windows as no NTFS)1404 & 1405
Changelogs for those:
1404
(2019):
1405
(2019):
1407
(April 2020 one I run now):
radeon
driver (IDK after system or BIOS update), but as said that was fixed within days by Ubuntu kernel and/or grahpic stack devs2203
update was Ryzen Master actually showing all cores sleeping when system is idle, so some CPU stuff in it2203
. Only thing is Update AMD AM4 AGESA PI
. Staying on 1407
until I see some more interesting changelog. Not installing it if it says one cant downgrade after updating. Also waiting at least 1 month and google first (Linux + 3950X)Glad you successfully fallback to previous firmware and thank you for this detailed procedure. As I told you I'm not in a hurry to upgrade BIOS. There are still so many bits I have to discover and to program into CoreFreq
Using the latest version, are you also noticing than for a standard target performance ratio of 35
, processors answers with a value of 36
?
To monitor this, press key !
to toggle to absolute frequency
Are you also observing that 36
is the magic value which triggers Turbo Boost ?
libc
not shown here, which may have a fix for the bug :man_shrugging: 5.3.18
, but these came:linux-image-5.3.0-61-generic linux-modules-extra-5.3.0-61-generic linux-headers-5.3.0-61-generic linux-headers-5.3.0-61 linux-modules-5.3.0-61-generic
1407
, they're gone now. BTW only thing showing in plain text after that update was also Updating Led Firmware
, but that might have been really a downgrade from the June onefreq
BASH alias I have on all my devices, which runs this command:watch -n 2 cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq
!
toggle since fall 2019 and DON'T remember how it was back thenHere already running one of the latest kernel version 5.7.5 but still having ACPI issue. My BIOS might be not ACPI tables compliant or the Kernel still don't know how to process some of them.
Your P-State ratio is showing 22
because, I believe, your processor is under the control of acpi_cpufreq
.
In my use case, I have blacklist any kernel modules, to study P-States in a deterministic way
Without CPUFreq drivers (confirmed by Missing
) I can program things these ways:
22
then you stress CPU and its frequency ratio is capped to22
(as expected)28
35
, idle shows 36
and full load goes up to 47
To summarize: I don't get the 47
ratio boost with P-States 22
and 28
but only with the P0 P-State of factory value 35
This is what we get from the Rocket SSD
From Tom's Hardware:
ATTO is a simple and free application that SSD vendors commonly use to assign sequential performance specifications to their products. It also gives us insight into how the device handles different file sizes
Straight from Corsair:
ATTO Disk Benchmark has long been an industry staple for measuring drive performance
/home
NVMe. Don't remember if it's the CPU bound or Chipset bound, but they get the same speed in Linux so...I fear that WSL won't offer more than the TSC WiP #191
Oh lol, I was about to try CoreFreq on the Windows Linux subsystem one day, but assumed it wouldn't work, but worth a shot since it's easy enough to try. That if I could actually install the build-essential
metapackage. Otherwise I would assume not. Also noticed screen
only works as root/sudo. Don't like to run e.g. apt
(or any package manager) to install/update many packages without screen
or similar, on any system.
Kind of odd Microsoft has added that option. First they kind of had their subsystem where you could run BASH directly from the Start Menu (through search only I think, was a hidden link) or Shift + Right Click in a folder in Windows Explorer. But they removed that without saying really. BASH just exited.
But was no good anyways when actually worked. Lacked so much. Ubuntu from the Microsoft Store runs the latest LTS, or you can choose a compatible LTS. Then it works, and has more or less what macOS has by default. Easier to install more, though. Not Homebrew and Apple making that harder or screwing up installed packages for every version.
It's weird to have cmd.exe
which is horrible but can do much more than one might think. Many driver packages, some software and enterprise AD use it for setup still... Has tons of AD/networking commands but very undocumented. Unless you've taken a MS certification course or something, it's hard to work with.
Powershell I played with a little. Long story short, even our Powershell expert couldn't make it do what a BASH script I wrote in 30 seconds did (read webcam barcode value and open browser to X URL with Y parameters).
So now Window's got 3! They gotta keep cmd for drivers and certain software, but hide it and just use BASH... IMHO.
Pretty nicely integrated with your /
Linux folder right there in Windows Explorer and everything. Little like Chromebooks but they also run GUI software just fine like GIMP etc.
Like... Have you tried just to get the MD5 value of a file on Windows without a 3rd party program? Found some official Microsoft package to add an insanely long command to cmd to do checksums but turned out to be deprecated, without saying any option... Now that's easy. And much more basics that a freakin' OS should be able to. Lots of Windows packages does "Verifying download/install" but IDK what mechanism. Apple has md5 (not called md5sum) and shasum like Linux, but doesn't use BASH by default anymore.
Is Hyper-V the Hypervisor of WSL-2 ? Booting the CoreFreq ISO into a Hyper-V VM behaves the same as within WSL. Thus is WSL a particular VM of Hyper-V, but without showing up into the VM manager.
Many MSR can be queried. Although they return zero, they don't crash. They are thus not pathrough, but trapped.
But APERF/MPERF counters, which are similar to MSR, crash. I feel they are trapped by Hyper-V, but something is missing. Probably, some settings are missing in my WSL configuration, such as Xen, KVM and VBOX let you define rules when a MSR is called.
Why Hyper-V makes possible a virtualized TSC, and not APERF /MPERF ? Especially when: virtualizing the TSC, the hardest part, has been achieved.
Well one search and...
Does WSL 2 use Hyper-V? Will it be available on Windows 10 Home? WSL 2 will be available on all SKUs where WSL is currently available, including Windows 10 Home. The newest version of WSL uses Hyper-V architecture to enable its virtualization. This architecture will be available in the 'Virtual Machine Platform' optional component. This optional component will be available on all SKUs. You can expect to see more details about this experience soon as we get closer to the WSL 2 release.
Definitively it's the same Hypervisor: CoreFreq is running in an ArchLinux VM and we're reading the same Vendor ID
It's been so hot in Europe these last days, I tuned my CPUs to cool down: No OC, especially PBO
Yeah, also very hot In Norway, for weeks :sun_with_face:
classroom
+ Ryzen Master to monitorclassroom
+ CoreFreq to monitorD.C.O.P = Enabled
to actually use my RAM chips' speedCSM = Enabled
to be able to boot Windows, since legacy bootNOTE: I divide this into 2 Categories: AMD OC
found under AMD Overclocking
and Manual OC
found under AI Tweaker
, or for you Extreme Tweaker
. From what I see they have the same Settings. One (fancy manual) setting was added in a later BIOS version than yours. I've never touched it.
OC Mode
set to Auto Overclocking
and RebootAMD OC
onlyfrom 3850 to 3920 MHz
70 MHz
Gear Down Mode = Disabled
AMD OC
onlyEnabled
GearDownEn = Enabled/Disabled
Power Down Enable = Enabled
AMD OC
onlyAuto
, may be affected by Gear Down Mode = Disabled
, I did not check, waste of timeManual OC
"Power Down Enable = Auto
AI Tweaker
, set Precision Boost Override = Enabled
Auto
AMD OC
Built-In PBO SettingsPBO
to a hypothetical 100 MHz
Overclock = Enabled
while the rest stays at Auto
70 MHz
higher on WindowsAMD OC
Built-In RAM SettingsGear Down Mode = Disabled
gave better Blender Benchmark resultPower Down Enable = Enabled
default is Auto
, as said may be affected by Gear Down Mode
Manual OC
SettingsAI Tweaker
, and don't plan to mess with othersUncore OC Mode
(my bad) which is under AMD OC
SettingsPrecision Boost Override = Enabled
and leave rest at Auto
100 MHz
here too200 MHz
on both Windows and Linuxclassroom
OC Mode = Default
then PPT = 95-97% of 142 Watts
OC Mode = Auto Overclocking
then PPT = 100% of 142 Watts
, showing Orange Color, but since 197 Watts+ is apparently no problem, it's just an indicator for that PPT142 Watts
tooPPT
, TDC
or EDC
reach 100%, the CPU does not give more performance65-67°C
, but NOT valid, fans barely spin up at that with my current UEFI fans config64-66°C
, but NOT valid, fans barely spin up at that with my current UEFI fans configOC Mode
Default
and Auto Overclocking
Gear Down Mode = Disabled
did not affect Temperature either. This was as said with OC Mode = Auto Overclocking
so PPT = 100% of 142 Watts
, still the better resultOC Mode = Auto Overclocking
is as said a Peak MHz of approx. 70 MHz
higher, but it is higher on Windows actuallyOC Mode = Auto Overclocking
causes 3850 MHz
to 3920 MHz
(more or less)OC Mode = Auto Overclocking
causes 3835 MHz
to 3905 MHz
(more or less)Conic Compute > Hyperboloid of two sheets
= 1.09 V
Precision Boost Override = Enabled
under Manual OC
OC Mode = Auto Overclocking
and Gear Down Mode = Disabled
for best resultsPPT
changes from 142 Watts
to 395 Watts
TDC
changes from 95 A
to 255 A
, but it never reaches that of course, always stays at Green ColorEDC
changes from 140 A
to 255 A
, but it never reaches that of course, always stays at Green ColorPPT = 50% of 395 Watts = 197.5 Watts
, and totally stable at that193-203 Watts
, but not constantly, so an average approx. Windows203 W
on Linux with this particular OC, while Windows doesn't want to go over 197.5 W
, it doesn't go below either, while Linux does (some balancing?)85-86°C
84-85°C
fishy_cat
in this mode on Windows, MHz and everything totally stable, but Watts is lower with that sceneConic Compute > Hyperboloid of two sheets
= 1.28 V
Awesome.
PBO is indeed helping a lot in performance score. But did you mention which policy plan is in used: Windows default or AMD policy ? The last is keeping the frequency target to a high value, at the expense of temperature; whereas Windows Balanced mode is keeping frequency among the lowest P-State ratios.
Would be cool if CoreFreq could read PPT, TDC and EDC MAX values as Ryzen Master shows them. Well at least PPT since CoreFreq shows Watts. Possible?
One software on GitHub is getting those values by reversed engineering. I've not try it yet. And specs just give the basic information. Not enough to program a safe algorithm.
D.C.O.P.
OC Mode: Default
reached 97% of PPT (142 W
) as beforeBut first the Voltages:
classroom
: 1.14 - 1.16 V
0.32 V
(up some times, but for the vast majority of time)classroom
: 1.06 - 1.09 V
0.20 - 0.91 - 0.96 V
(vast majority of time at 0.91 V
, but all cores show they've been at 0.20 V
too, and occasionally goes 0.96 V
)Frequencies after having run 1 minute:
3825 - 3850 MHz
15-20 MHz
lower, never hitting 3850 MHz
again after the very beginningACTUAL RESULTS:
OC Mode: Auto Overclocking
and Reboot142 W
) as expectedGear Down Mode = Disabled
too135 Watts
132 Watts
classroom
went like this: 4m23s, 4m26s, 4m29s, 4m30s
, and then I facepalmed and gave up136 Watts
classrom
: 4m plus/minus 1 second
, every timeD.C.O.P.
POSSIBLE REASONS:
TO YOUR QUESTION:
Balanced
Power Plan as it's the only one really made for Windows 10. Changing to another removes a Performance Slider from the new Settings Appondemand
on Linux when I see the frequencies using HWMonitor (same as !
in CoreFreq)Performance
plan compared to the Ryzen Performance
plan they are pretty identicalRyzen Performance
does nothing more than make the CPU run at Base Clock MHz all the timesysfs
too... Setting governor
to performance
or a Specific Frequency for all 32 in sysfs
(think this was needed, changing governor alone didn't work, it still boosted)Power Saver
, Balanced
and Performance
. Those old ones found in the "old" Control Panel are just there for legacy reasons it seems, considering the Slider disappears if not using Balanced
as said, at least it was like that, and I'm 100% sure it disappears if you choose a 3rd Party PlanBalanced
plan to the same settings as seen in Ryzen Performance
, and IDK what the Slider is for then, maybe it disappearsIt depends on Windows' mood or time of day?
There are so many Windows service processes of any kind running that it's hard to measure and reproduce benchmark. Even RM has a 100 to 300 MHz CPU overhead for its own monitoring. Either Windows is stripped of any tasks to make it bare metal, or Linux Kernel is built with most drivers and security removed, such as page randomization. Then you will set processor registers to your will and barely get a reproductible and understandable score.
My Windows 10 is as Clean as it gets. All permanent or scheduled unnecessary services Disabled.
After boot:
cmd.exe
I did try the Ryzen Performance
Power Plan, and observed Ryzen Master for any changes:
Ryzen Performance
Power Plan is different than the standard Windows Balanced
is that it sets "Minimum CPU % = 100", which does nothing as said aboveRyzen Balanced
Power Plan is even stupider. Its only difference is that it sets "Minimum CPU % = 99". Those using that are just stupid. No difference whatsoever in the values aboveBTW: Not sure if you're ever going to find that the Cores actually Sleep. I've been watching and never seen a Core below 130 MHz - once it's below 130 MHz it says "Sleeping". So maybe "Sleeping" is just indicating very low MHz...
BTW: Not sure if you're ever going to find that the Cores actually Sleep. I've been watching and never seen a Core below 130 MHz - once it's below 130 MHz it says "Sleeping". So maybe "Sleeping" is just indicating very low MHz...
But AMD says Processor has C-States. Those are mentioned in ASUS BIOS. The Software Manual specifies the X86 MONITOR/MWAIT instructions to enter idle states. This has to exist but we don't have so far the counters addresses to monitor them.
I'm reading you that RM may implement a CPU load minimum threshold to say: "it's sleeping" But why such dummy algorithm ? RM is an AMD software: authors have full access to the NDA specifications to make an accurate software !
I updated CoreFreq a couple of days ago using my automatic script and it worked just fine.
Now I did it again, and loading the module works but the daemon does not load and the UI shows nothing.
Setup is same as before when you asked about my entire setup. I see you did some changes recently.