kernel configuration on AMD FX 8350 box, cpu utilisation etc

[mikefot]

Dear Tony,

I have installed Gentoo on the AMD FX 8350 box I discussed in an earlier posting on the site where I had talked about a power supply that was intermittently cutting out etc.

Since then I have replaced the power supply.

The box now has both debian stretch (kernel 4.8.2) and Gentoo (kernel 4.9.0) installed on the same hard disk.

The graphics card is an older Nvidia engt 430 model - but it works OK.

The debian install seems pretty stable and well behaved.

Under debian running cat /proc/cpuinfo gives the following output:

[Zucca]

Hi. I have the exact same CPU. Maybe I can help.
Please pastebin your kernel .config, and output of dmesg.
_________________
..: Zucca :..
Gentoo IRC channels reside on Libera.Chat.
--

[wrc1944]

This seems like it may be releated to heat issues. What's your cpu cooling solution? That may be the whole problem.
Amd fx cpus are notorious over heaters. I have an 8320 (essentially the same cpu as your 8350).

Keep in mind that since you can only use j1 in makeopts, and any use of more cores seems to make the cpu/system shut down, that's a pretty good clue that your cpu is overheating. Also, the debian install behaving well as oposed to Gentoo is another clue.

Most cpus, systems, and also power supplies will have an auto shutdown feature in case heat reaches the danger point.
I keep a little gkrellm sensor window visable while compiling with Gentoo, showing all the temps for each of the 8 cores in real time as different packages compile.

Theres more heat fluctuation than one would think with Gentoo emerges. Large packages like libreoffice, chromium, etc. will take lots of time and heat up the cpu and system drastically more than a series of shorter emerges, where there's a little time between them when the cpu cores get a little rest, and cool down quite quickly.
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

[mikefot]

Many thanks for the replies to my posting.

​ I installed povray and gkrellm and then ran povray -benchmark - kindly recommended by a debian user.

I also checked the cpu temperature as well as the cpu core activity.

As the test ran I noticed two things:

1. All 8 cpu cores ran simultaneously at 100% capacity.

2. The cpu temperature rose gradually until the benchmark was about 85% complete - by then the temperature exceeded 84C and the cpu automatically cut out and the pc shut itself down.

So the suggestion about the cpu overheating problem was spot on here. Thanks.

When I bought the motherboard and cpu I got the retailer to switch the standard stock cooler for an Arctic Freezer 13 model.

They installed it for me and I fitted the motherboard into the PC cabinet with the cooler.

I had thought the cooler was working properly, but now I am not so sure......

The cooler has a number of fan settings apparently.

I googled around trying to figure out what the command in either debian or gentoo would be to change the fan setting to a greater speed etc but could not find an obvious instruction that......

Does anyone have any idea how to do achieve this?

I am going to open the pc cabinet and take a look at the cooler and run the test again to see what the fan does when the I run the test......

It should run faster.

I assume gkrellm could show the fan speed.

Suggestions on investigating the cooler are most welcome.

I will post the .config file shortly.

Regards and thanks

Michael Fothergill

[wrc1944]

OK- good work. From the cpu shutdown with more than j1, I was pretty sure temps were the problem. I've built lots of systems myself, and seen it before.

84C is dangerously high for amd cpu's, no matter what amd says. There are a few things you can do right away.

There MUST be good air flow in and hot air exhaust out from inside the case. In fact, I just keep my case side wide open, and have at times when compiling libreoffice or @world rebuilds needed a little desktop fan blowing directly into the side of the case.

The ambient room temps will also play a big role in how well the cpu and case cooling systems actually function.

Is this the cpu cooler you have? http://www.newegg.com/Product/Product.aspx?Item=N82E16835186039 If so, IMO it's not really up to really handling an AM3+ FX 8350 being asked to run Gentoo or do heavy gaming, etc, unless all other cooling factors are maxed out. The specs don't mention AM3+, only up to AM3.

Any Pre-applied thermal compound between the cpu and the cooler itself can and will eventually dry out, or not have been applied correctly. If nothing else helps with the cooling very much, this could be the problem. Pre-applied thermal compound is usually junk anyway- I always use Arctic silver 5 or better. Having proper thermal compound properly applied is really a big deal. No cooler will do it's job without it.

Again, all thermal compound can and will dry out in time from heat, so it's a good idea to know how to do this correctly yourself, or you'll have to pay a tech person. There can be serious problems removing/re-installing the cooler and cleaning the old off both the cooler and cpu, and replying a very thin coat of new compound, and remounting the cooler, If you've never done it before, to avoid serious cpu or motherboard damge you should research and/or look at youtube videos of the proper procedure. google youtube apply cpu thermal compound for lots of them.

Make sure the case and fans are not clogged with dust. If so blow them out with canned air. Dust clogged cooler fans and cases are a major problem and restrict good air flow. This is also a BIG DEAL.

Gkrellm has fan speed monitoring you can enable under builtins->sensors for config settings (right click top of gkrellm for configuration access, or just hit F1 on keyboard) Your kernel also needs the proper sensor module enabled for your specific motherboard, and then loaded for each kernel in /etc/conf.d/modules. For example, mine is CONFIG_SENSORS_IT87=m in my kernel config file

Forgot to mention- all 8 cpu cores running full out when compiling is a feature, not a problem, but does mandate a good cooling solution.
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

[mikefot]

Many thanks again for your response and advice here.

Yes it is indeed the cooler device featured in your link.

The thermal paste business is pesky because it seems you periodically have to swab/baste and then screw/truss up the Xmas turkey-cpu-kitchen-heat-sink ensemble (buffoon comment?!...) and then carefully bake/sautee it to perfection......

I will need to think about this.

Regards

Michael Fothergill

Here is the config file:

[url] https://paste.debian.net/904476/
[/url

[wrc1944]

I just looked at 10-15 of those youtube videos, and a lot of them apparently don't know what they are doing, in that they apply WAY too much compound, and have it oozing off the cpu, to the point of being ridiculous, and a real hazard.

The danger over time of excess compound (which is conductive) getting off the cpu and onto the motherboard is very real, and to be avoided. You want to only apply the bare amount needed to have an extremely thin and even coating between the cpu and cooler faces, so as not to be squeezed out when pressure from the cooler is in play.

A good compound properly applied should not need to be redone for at least a year, likely longer, depending on usage and heat experienced. If you monitor your cpu temps regulary, you will know when to suspect you might need to remove and re-apply compound. If you notice heat reading getting higher, check for dust build-up first, then suspect the cpu compound.

Your config file has no Native drivers for sensors set (as modules) What brand/model motherboard do you have? They need to have a module set depending on what sensor chip the manufacturer used.

What does lsmod report what's being auto loaded at boot?

You should have at least the default
AMD CONFIG_SENSORS_K10TEMP=m
CONFIG_SENSORS_FAM15H_POWER=m (the amd bulldozer family, i.e. 8320 and 8350)

and possibly more in your kernel config file, and to auto-load at boo listed in /etc/conf.d/modules.

For example, in my /etc/conf.d/modules file is:

[mikefot]

Many thanks for taking interest in the problem I currently have.

The motherboard is an ASUS Sabertooth FX 990 R2.0.

I will run lsmod and post the result on the site here.

Regards

Michael Fothergill

[Tony0945]

Mike, I've refrained from commenting because I don't have an FX CPU, but I concur with what has been said. I suspected a dirty heat sink but didn't know how old the system was.
Standard advice is to use canned air but I find it leaves a wet residue so make sure it's dry before you fire up the box. I had unsatisfactory results from the expensive cans so I dragged the computers into the garage where I have an air compressor set at 40 psi for pumping up car tires. I attached the air gun and followed advice to block the cpu fan with a pencil so it doesn't spin to death (actually used a broken piece of wooden dowel) and blew it out. An unbelievable amount of dust came out. I had to back flush it several times before it came clean. Don't use a vacuum cleaner as I've read that you can kill the mobo with static. Don't set the pressure real high. 40 psi was OK, I wouldn't go any higher. Lower may work depending on how packed the dust is. Try not to blow on small components. You just want to get the dust out of the heat sink fins.

I concur with the advice about Arctic Sliver, it's the best, but would be hesitant to take off the cooler and reapply. I've only done that once and it was because the cooler had come unlatched. If you do this, however, clean the cpu and the cooler face with rubbing alcohol and dry thoroughly before applying the paste. I used ordinary kitchen paper towels for application of the alcohol. Don't drench it. I've read (and it makes sense) that it's important not to leave any lint, so I wouldn't use cotton balls as some do. I'm assuming that this heat is a recent thing so the original paste was applied correctly. Blow the dust first.

At work we had magnetic filters that we slapped across the front of the metal cases and replaced weekly. I'm still investigating what to do for a home computer because the cases are different from an industrial computer. Thankfully the environment is not so extreme.

it87 has worked for all my gigabyte AMD mobos, including the FM2. Don't know about Asus, never had one. Probably both companies work from an AMD reference build.
EDIT: Some are f71882fg, sensors-detect from the lm_sensors package will find the right one for you.

Are you running lm_sensors? If you are running Gnome or Mate and maybe others there is an applet wherein you can watch the temperature from lm_sensors in real time.
I had the same problem in an X3, temps built up close to 100 C then the CPU just crashed. Sometimes you can "feather" the temperature by suspending the emerge with CTRL-S and restoring with CTRL-Q when the temperature drops. For me it is building Wine and Gcc that are the big problems. Best to use -j1 though.

[wrc1944]

The Asus Sabertooth FX 990 R2.0 mobo is definitely it87, k10temp, and fam15h_power. http://www.spinics.net/lists/lm-sensors/msg43352.html,

Just to check, you might try a "modprobe it87" (no quotes) as root, and then see if lsmod says it87 is loaded. probably not- but worth a shot.

You will need to enable the it87, fam15h, and k10temp in your kernel config, and put it87 and k10temp in /etc/conf.d/modules as I mentioned above..

Forgot to mention. Important Your body static electricity can and will fry computer components, including cpu, ram, or motherboard items.
You need to always discharge yourself before doing any touching of components, either with a grounded wrist strap, or before and perodically touching the bare metal chasis of your plugged in computer ( the rear will suffice. This is no joke- countless cpus and ram sticks have been fried by overlooking this point. In most cases you wouldn't even notice you were statically charged, and then wonder why your cpu or ram suddenly was dead.

It's more likely to happen in winter when the average household air is very dry from heating, and a charge easily builds up as you move around.

I compile gkrellm without the lmsensors USE flag, so I've never needed to mess with it for many years. The gkrellm defaults usually don't require any tweaking for several years. Also, the it87 sensor kernel module needs to be loaded BEFORE Gkrellm starts, or else Gkrellm doesn't detect it. You'd then need to modprobe it87 as root, and restart Gkrellm.
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

[Tony0945]

[Anon-E-moose]

I run an fx8320 overclocked to 4.0, the temps never run over 60C even during heavy compiles.

It does sound like the heatsink either isn't situated properly or is dirty OR as suggested, the compound has been applied wrong and/or needs changing.

I have filters on the computer intake vents on mine to keep dust bunnies from collecting inside the computer, and at that I have to clean the filters at least once a month. Yes, that much stuff can get into the system.
And even with that, very small particles get into the system and the cpu heatsink.

Turn the system off and use compressed air to blow out the stuff from the heatsink (from several angles)
AND blow the dust from all of the components around the cpu, northbridge, southbridge, etc.


Edit to add: on my system (Asus m5a99fx) I set the fan to run full speed, in the bios (turn off bios control)
and have it set to run with fancontrol from lm_sensors. Along with linux idle control to throttle the cpu when not needed.

Edit to add2: I don't use gkrellim (I use conky) but they both should offer to show what the fan speed is and if it is working (whether bios or other software) you should see it change depending on the heat of the cpu.
_________________
PRIME x570-pro, 3700x, 6.1 zen kernel
gcc 13, profile 17.0 (custom bare multilib), openrc, wayland

[mikefot]

Thanks a lot for the extra help again.

I am starting to believe that the cooler I have can be made to work more effectively.....

There is probably something fairly simple that is wrong with it at prtesent.

I have been researching different types of thermal paste etc. Arctic MX-4 sounds like it could be helpful here.

I have a colleague who repairs PCs and is good at cooling problems - I am going to contact him about this problem.

Regards

Michael Fothergill

[mikefot]

Dear Folks,

Here is some output from the Debian OS (I will post some from the Gentoo install later).

[wrc1944]

Yes- looks like your debian install default kernel already had the it87 module enabled in it's kernel. Gentoo kernels won't by default, unless you ran genkernel and enabled everything,or manually configured it in yourself when you installed Gentoo.

IMO, enabling everything kind of defeats the purpose of running Gentoo, where build your own very lean and fast kernel for your own hardware specs.
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

[mikefot]

I am posting the dmesg output from gentoo:

https://paste.pound-python.org/show/tYpiailkFwqhdYAdpj9E/

Regards

Michael Fothergill

[Anon-E-moose]

[mikefot]

Dear All,

I ran lsmod on the gentoo install.

It lists the realtex r8168 module that I installed and that is all.....

I ran modprobe it87 and it gave an error "FATAL: Module it87 not found in directory /lib/modules/4,9,0-gentoo".

So it can't be installed at present as was suggested.

I will fix that problem.

I also installed gentoo on the A3400 amd box I have and it seems to run OK.

Same with the kaveri box but I need to get the sound working properly.

I will focus on the heating problem on the FX 8350 install until I get it sorted.

Thanks a lot for all the help.

Regards

Michael Fothergill

[mikefot]

Dear Anon-e-moose,

I was wondering - did you post what kind of cooler you have attached to your fx cpu?

Is it superior to the one I am using?

Regards

Michael Fothergill

[Anon-E-moose]

[mikefot]

Many thanks for your reply on this topic.

I had begun to wonder if I needed to get a more powerful cooler but as I posted earlier it does seem that the kit I currently
have will work once I have cleaned it, dusted it, rebasted it (if necessary), set the fan speed correctly and installed and configured all kernel features
plus temperature control software it should fix the cooling problem.

With the kaveri box Gentoo installation that Tony kindly help me work through, I ran it yesterday and did a sync update and then updated all the packages.

This took some time. It included a full compilation of gcc and chromium.

Makeopts was set to -j5 which is correct for the kaveri box.

It ran continuously without any shutdowns occurring.

I did not think to have e.g. gkrellm or lmsensors running to see if I was sailing close to a thermal shutdown at any point in this upgrade.

But from now on I am going to do this with every Gentoo installation I have to check on the thermal health of the cpu during @world type updates etc.

It did occur to me that if the kaveri box were clever enough it might in theory be able to throttle down the cpu in some way if it were starting to overheat and control the temperature
that way if the fan wasn't coping well enough during the most demanding compilations etc.

If that were true then it would mean that the compilations were not performed in a truly optimised manner. The machine would be underperforming in practice.

The kaveri box only has the standard cooling fan that was shipped with the motherboard.

Regards

Michael Fothergill

[Tony0945]

Assuming the upgrade was 4.9.3 to 4.94, don't forget to re-emerge libtool as discussed on the wiki here: https://wiki.gentoo.org/wiki/Upgrading_GCC If yopu upgrade from 4.9 to 5.4 then there is more. It's all in the wiki article.

BTW, my Athlon II box started overheating too. The revdep-rebuild wanted to do 122 packages and it quickly started showing over 100 C, so I thought it had to be on the edge, so I stopped it, went into the BIOS, locked CPU speed at 800MHz (the slowest available) and did the revdep-rebuild with -j1, so the build went at 1/12 speed. I started Christmas morning and it just finished late last night. lm_sensors showed 138F and the BIOS monitor showed 135F. I shut it down and this morning took it into the garage and blew it out, after removing the hard drives. Lot's of dust. I also blew out the power supply which I have never done before in the six years I've had it. UNBELIEVABLE fountains of dust!
The steel case was really cold when I brought it back in, so I'm waiting for it to return to room temperature before putting the drives back. I blew out the drive electronics with my mouth. I have asthma, so it wasn't that strong. If this hasn't helped, I'll take the cooler (stock) off and reapply paste. I hope not. I'm hoping this CPU/mobo will limp along until AMD finally starts shipping Zen.

EDIT: Oh, man! Because I removed one data drive, OpenRC wouldn't complete because it couldn't mount that drive and it's removal made another data drive no mountable as ext4 because it's really JFS. Some of those OpenRC "improvements". So I had to boot sysrescuecd (which I have permanently attached on a usb) to edit fstab. After this contretemps I booted at the normal load determined settings and found the system idling at 85 F, a 53 degree reduction. The moral here? Don't forget to clean your power supply too (unplug it from the wall first).

[Tony0945]

[Anon-E-moose]

With idle and fancontrol

at idle cpu runs at 1400 __________ full out at 4000
_____ fan runs at 850 rpm ________ full out at 2200 rpm
and the vcore varies between 0.9 and 1.33 volts.

Right now (room temp is 77F)
case is 29 C - 84F
cpu is 32 C - 89.6F
K10 is 14 C - 57F (always reads low until it hits near the 40C mark, known problem with Fx series)
AMD is notorious for the cpu and k10 sensors not matching up until it hits near the 40 C mark.

At full tilt, all 8 cpus 4000hz, and fan at 2200rpm with a heavy compile the temps
cpu runs 60-64 C - 140-147F
K10 runs 58-60 C - 136-140F

Note: From what I've seen I use the K10 sensor as it is more reliable for true cpu temp, but again that's once it hits 40 C.
I believe the cpu temp is the socket temp and the k10 is the cpu itself.
But unlike intel chips, there is only one temp on the cpu for amd, at least through the fx series.

If I compile something like firefox or gcc it will run the system full on, all 8 cores ( I use -j8 )
but I also do heavy compiles from the command line, not in a gui.
And the temps spike to highs, but as soon as the cores are not being used the temps drop quickly.

The number one key is proper airflow, I not only have the cpu fan tuned, but plenty of airflow in the case itself to vent the hot air.
_________________
PRIME x570-pro, 3700x, 6.1 zen kernel
gcc 13, profile 17.0 (custom bare multilib), openrc, wayland

[Tony0945]