You are not logged in.

#51 2015-02-28 02:29:45

graysky
Wiki Maintainer
From: :wq
Registered: 2008-12-01
Posts: 10,595
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

dragonn wrote:
graysky wrote:

...but does the tick rate or delay help with your freq?

For me no hmm I have tried default kernel with HZ=1000Hz (not BLD this time) it didn't help. Setting the delay to 20 on HZ=1000Hz have helped a little bit - the CPU goes some time to ~2GHz. With GPU do you have?

Using the onboard GPU here.

max.bra wrote:

tested fedora config: frequency downscale correctly

If you boot into this kernel. Leave the machine idle and do this for 4-5 min.  Post the result.

Last edited by graysky (2015-02-28 02:31:23)


CPU-optimized Linux-ck packages @ Repo-ck  • AUR packagesZsh and other configs

Offline

#52 2015-02-28 11:06:12

dragonn
Member
Registered: 2012-05-04
Posts: 65

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

graysky wrote:
dragonn wrote:
graysky wrote:

...but does the tick rate or delay help with your freq?

For me no hmm I have tried default kernel with HZ=1000Hz (not BLD this time) it didn't help. Setting the delay to 20 on HZ=1000Hz have helped a little bit - the CPU goes some time to ~2GHz. With GPU do you have?

Using the onboard GPU here.

max.bra wrote:

tested fedora config: frequency downscale correctly

If you boot into this kernel. Leave the machine idle and do this for 4-5 min.  Post the result.

Maybe it is the difference because most of us use Nvidia GPU as I see in this thread? Hard to see, maybe some one will test it and change for a test to the onboard GPU.

Offline

#53 2015-02-28 11:44:42

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

dragonn wrote:

Maybe it is the difference because most of us use Nvidia GPU as I see in this thread? Hard to see, maybe some one will test it and change for a test to the onboard GPU.

already done, stock kernel + nouveau + xorg session. no difference with 10 and 19ms sample. 2-3 mins log with ~3840Mhz median.
the way was good but nothing to do...

Edit:

graysky wrote:

If you boot into this kernel. Leave the machine idle and do this for 4-5 min.  Post the result.

tomorrow i will do it for sure

Last edited by max.bra (2015-02-28 11:46:50)

Offline

#54 2015-03-01 12:55:13

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

more time, more tests and better report

kernel                                 median  counts
Arch Stock, 10ms sample, nvidia xorg   3785    1260
Arch Stock, 10ms sample, nouveau xorg  3856    1200
Arch Stock, 10ms sample, tty           3623    1388

Arch Stock, 19ms sample, nvidia xorg   3606    1536
Arch Stock, 19ms sample, nouveau xorg  3795    1216
Arch Stock, 19ms sample, tty           1296    1124

Arch 1000Hz, nvidia xorg               3484    1344
Arch 1000Hz, nouveau xorg              3767    1568
Arch 1000Hz, tty                       900     1788

Fedora config, nvidia xorg             2502    1480
Fedora config, nouveau xorg            3320    1328
Fedora config, tty                     846     2332

Last edited by max.bra (2015-03-01 16:12:24)

Offline

#55 2015-03-01 13:14:08

graysky
Wiki Maintainer
From: :wq
Registered: 2008-12-01
Posts: 10,595
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

That is interesting... there seems to be something else in the fedora config helping your nouveau idle.  Also, it is clear that the nvidia drivers exacerbate the problem under either config.  My machine uses the onboard GPU (the Intel 4600 HD).  So I can't comment about the nvidia effect.


CPU-optimized Linux-ck packages @ Repo-ck  • AUR packagesZsh and other configs

Offline

#56 2015-03-01 14:35:27

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

sorry my bad. terrible copy paste.
last fedora was tty. corrected.

Offline

#57 2015-03-01 14:49:24

graysky
Wiki Maintainer
From: :wq
Registered: 2008-12-01
Posts: 10,595
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

max.bra wrote:

sorry my bad. terrible copy paste.
last fedora was tty. corrected.

...what about nouveau + fedora config?


CPU-optimized Linux-ck packages @ Repo-ck  • AUR packagesZsh and other configs

Offline

#58 2015-03-01 14:56:52

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

and yes, nvidia (proprietary or free) seems to be involved.
tty 1000hz and 19ms stock behaves like yours.

i'm looking for some about firmware settings or xdamage related.
using nvidia blob, i've never seen that powertop register nouveau as the first cpu wakener.

Offline

#59 2015-03-01 14:59:19

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

graysky wrote:
max.bra wrote:

sorry my bad. terrible copy paste.
last fedora was tty. corrected.

...what about nouveau + fedora config?

do you think is needed?

Offline

#60 2015-03-01 15:03:25

graysky
Wiki Maintainer
From: :wq
Registered: 2008-12-01
Posts: 10,595
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

@max.bra - Maybe not... just trying to understand it.  For my system, the default 10 ms using the modified tick rate of 1000 sets it straight.

stats.sh mylog.txt
median: 931.41
count : 1052

Again, I think there's more going on with your system.  Hopefully, the kernel devs will take this on based on the bugzilla.


CPU-optimized Linux-ck packages @ Repo-ck  • AUR packagesZsh and other configs

Offline

#61 2015-03-01 15:06:26

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

time to compile and log...

Offline

#62 2015-03-01 16:12:53

max.bra
Member
From: Bologna - Italy
Registered: 2008-06-02
Posts: 93

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

added 'fedora nouveau xorg'
nothing new...

Offline

#63 2015-03-23 01:08:07

RankoKohime
Member
Registered: 2014-01-08
Posts: 87
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

bmanuel wrote:

@graysky, is it normal for this CPU to reach 93°C (gkrellm/lmsensors) while compiling the linux kernel? I've the stock cooler and the CPU is not overclocked o_O

Intel specifies 74°C for the Tcase, which is described as the temperature at the heat spreader, which is going to be different than the core temperature.  I can't find a listing for the actual core temp, but I understand Haswell (and Refresh) desktop CPU's to throttle around 95-100°C.  Running the max-heat torture test in Prime95 can get my 4790k up to low 90's on a 120mm AIO.

Either way, I'm not personally comfortable with temperatures that high, and with the computer itself drawing >200 watts from the wall, up from ~50 watts at idle, I think the CPU is blowing well past it's TDP rating when running in turbo mode.

For this reason I like to disable turbo when running some non-time-critical process, such as kernel compilation.  It does assume you're running pstate, of course.

echo 1 > /sys/devices/system/cpu/intel_pstate/no_turbo

Last edited by RankoKohime (2015-03-23 01:09:16)

Offline

#64 2015-03-23 08:55:43

graysky
Wiki Maintainer
From: :wq
Registered: 2008-12-01
Posts: 10,595
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

Ranko - consider lowing your vcore or upgrading your cooling solution particularly if your cpu is not overclocked.


CPU-optimized Linux-ck packages @ Repo-ck  • AUR packagesZsh and other configs

Offline

#65 2015-03-23 15:17:52

bmanuel
Member
Registered: 2011-10-06
Posts: 106

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

@RankoKohime thank you for getting back with your insights: yeah, i'm pretty sure it was something related to the very old thermal paste i stupidly used. Now that i switched to an ARCTIC Freezer 13 i rarely see even 75°C..

Offline

#66 2015-03-24 01:54:11

RankoKohime
Member
Registered: 2014-01-08
Posts: 87
Website

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

graysky wrote:

Ranko - consider lowing your vcore or upgrading your cooling solution particularly if your cpu is not overclocked.

Cooling upgrade is in the budget queue.  The temperatures are a more manageable <=70°c when running anything other than Prime95, (even with turbo enabled) it's just the algorithms in Prime95 can melt Haswell chips.  (And I believe Ivy chips as well?)

Offline

#67 2015-12-08 14:23:10

Arup
Member
From: Earth
Registered: 2014-02-20
Posts: 95

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

https://bbs.archlinux.org/viewtopic.php?id=173180

Seems this issue has been going on for long but there are different school of conclusion for it. For some its normal but for others its not.

Offline

#68 2016-06-17 10:30:05

Arup
Member
From: Earth
Registered: 2014-02-20
Posts: 95

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

CONFIG_FAST_NO_HZ=y in stock Ubuntu kernels also could be the reasons for lower temps in Ubuntu vs Arch apart from Hz=300

The latest 4.6.2.1 does improve over the older Arch kernel as now the CPU scales rather than get stuck at turbo speed but temps are still up over Ubuntu 16.04 4.4 kernel.

Offline

#69 2016-06-26 13:41:03

Arup
Member
From: Earth
Registered: 2014-02-20
Posts: 95

Re: Arch specific bug-high idle CPU freq for Haswell Refresh CPUs

Just bumping this thread, with kernel 4.6 the cpu frequency on my Intel Haswell 4790 now scales regularly. However temps have gone down but are about 2 degrees more than Ubuntu 16.04 with kernel 4.4, whats others take on this.

Offline

Board footer

Powered by FluxBB