Allegro.cc - Online Community

Allegro.cc Forums » Off-Topic Ordeals » Turned off AMD "Cool 'n Quiet", Idle Temps Drop 10c

This thread is locked; no one can reply to it. rss feed Print
 1   2 
Turned off AMD "Cool 'n Quiet", Idle Temps Drop 10c
Kris Asick
Member #1,424
July 2001

...yeah. The special AMD processor feature which allows the multiplier to be adjusted thousands of times a second in order to reduce power consumption when raw processing power isn't needed was raising my idle temps by 10c. Before now, my idle temps were 40c. Now, after disabling this feature in the BIOS, they're 30c. :o

Considering heat is a factor of how much power is being utilized, I think it's safe to say that this "power saving" feature wasn't exactly doing as such on my new Windows 8 Pro x64 system for whatever reason, even though various utilities were reporting fluctuating CPU speeds down to under half of the 4 GHz my AMD FX-8350 is capable of. With this feature off, the speed is now constantly slightly over 4 GHz, usually 4.09, though it goes up to 4.13 under load. (And yes, it's set to 4000 MHz in the BIOS with all overclocking features disabled, so it must be the CPU itself deciding to go a little higher.)

Disabling this feature also improves performance with certain applications and games, while everything that was working fine prior is continuing to work the same as before.

*shrugs* :P

--- Kris Asick (Gemini)
--- http://www.pixelships.com

Arthur Kalliokoski
Second in Command
February 2005
avatar

I don't notice any difference in temp, I was assuming that the sleep cycle was shorter when the frequency was down. It still has to do a certain amount of work per second, after all, unless a CPU hog is taking all it can get.

They all watch too much MSNBC... they get ideas.

Kris Asick
Member #1,424
July 2001

*shrugs again*

I'm going to guess it has something to do with my specific combination of hardware and software. Still, disabling it did improve performance with various things that didn't normally demand heavy amounts of CPU power, especially Flash-based stuff.

Plus, I also enabled something in the BIOS called "High Performance Computing" or something like that, which also boosted overall performance but this time for things that SHOULD have been getting maximum power. For instance, my framerate in Minecraft became a more steady 60 FPS after enabling this, whereas beforehand, it would still normally be 60 FPS but would fluctuate down to 58/59 at random moments that weren't chunk-loading related.

I tried disabling C1E as well, but this didn't really seem to do much of anything and from what I understand I really shouldn't be messing with the C-states if I intend to allow my computer to drift into sleep mode, so I turned it back on.

--- Kris Asick (Gemini)
--- http://www.pixelships.com

Thomas Fjellstrom
Member #476
June 2000
avatar

The HPC option is supposed to crank everything up to 11. No CnQ, no down clocking, no idle sleeping. Just pure unadulterated MegaHurtz(™)

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

no idle sleeping

What? It's running on DOS?

They all watch too much MSNBC... they get ideas.

Thomas Fjellstrom
Member #476
June 2000
avatar

What? It's running on DOS?

Eseentially thats what HPC people want. They want every single clock cycle they can get. Low latency is important as well, and those sleep functions really mess up latency.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

I was under the impression that the task switch took control via an IRQ tied to the system timer, and the "hlt" instruction returns the cpu upon receipt of an IRQ? If it's virtualized it might take 300 cycles or so every task switch, assuming the scheduler switches 100 times a second, that'd be 30k cycles out of 2 or 3 billion, hardly worth worrying about.

They all watch too much MSNBC... they get ideas.

Thomas Fjellstrom
Member #476
June 2000
avatar

I'm saying the CPU doesn't sleep, not that individual processes don't sleep.

Except that with HPC clusters the processes most likely won't ever want to sleep.

There are some very new linux kernel features that allow you to run a single process on a single cpu, and turn off the OS timer on that cpu, so theres absolutely no scheduler interrupts or any kind of OS jitter.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

I'm saying the CPU doesn't sleep, not that individual processes don't sleep.

So what happens when no process needs the CPU? It gave the text editor a slice, but the editor is waiting on a keystroke, the database is waiting on the hard disk, the web page is waiting on packets, so the OS spins in a busyloop?

{"name":"607765","src":"\/\/djungxnpq2nug.cloudfront.net\/image\/cache\/2\/6\/26a6343b801f34e2cccfba4132c01793.png","w":810,"h":624,"tn":"\/\/djungxnpq2nug.cloudfront.net\/image\/cache\/2\/6\/26a6343b801f34e2cccfba4132c01793"}607765

They all watch too much MSNBC... they get ideas.

Elias
Member #358
May 2000

Under Windows the process list also has a "System Idle Process" which would use up all the CPU not used by anyone else.

--
"Either help out or stop whining" - Evert

Arthur Kalliokoski
Second in Command
February 2005
avatar

But since the temperature goes down when System Idle Process is predominant, it must be halting the CPU. I remember reading about Win95 being slammed for not using the hlt instruction, and the explanation was that a major motherboard manufacturer product would crash when the hlt instruction was encountered. Funny nobody else had trouble with it.

[EDIT]

Quote:

But since the temperature goes down when System Idle Process is predominant, it must be halting the CPU.

Also when you're watching Resource Meter in Windows in a VM while you're watching top or whatever in the host OS.

They all watch too much MSNBC... they get ideas.

Thomas Fjellstrom
Member #476
June 2000
avatar

So what happens when no process needs the CPU? It gave the text editor a slice, but the editor is waiting on a keystroke, the database is waiting on the hard disk, the web page is waiting on packets, so the OS spins in a busyloop?

My guess is the CPU is still spun up, but doing nothing. Meaning its not in any sleep state. So it'll be ready for any code right away.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

My guess is the CPU is still spun up, but doing nothing.

Unpossible. It's either in a halt state, or executing code of some kind. I suppose it could be slowed down to 33Mhz and put into a busyloop or something, but I really doubt it.

They all watch too much MSNBC... they get ideas.

Thomas Fjellstrom
Member #476
June 2000
avatar

I meant its not in any kind of C sleep state. a HLT doesn't really save any power. It won't put any of the silicon to sleep. It's just idling.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

Thomas Fjellstrom
Member #476
June 2000
avatar

http://en.wikipedia.org/wiki/C_state#Processor_states

And one thing I didn't know till just now, is that HLT/Idle is a sleep state, C1. Interesting.

Anyhow, I think C1 is about the lowest HPC people would like it to go. And I think it may be possible to even disable C1. Just because the CPU has nothing to execute, doesn't mean it's powered anything down.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

I never heard of those Cx states until just now, but I'd assume C2 and C3 are associated with "entire computer goes to sleep" type deals.

But how could a CPU "not execute anything" without being powered down? (or stopping its internal clock, which should be the equivalent). I ought to see what happens to the RDTSC count when halt is executed on DOS. Of course the 55 millisecond DOS timer interrupt will stop the hlt 18 times a second, so I'd have to have a loop with a hlt instruction in it between the start and stop times.

They all watch too much MSNBC... they get ideas.

Thomas Fjellstrom
Member #476
June 2000
avatar

No, I don't think C2 is part of "entire computer goes to sleep" type deals. C3 looks like it's used in Suspend to RAM though, but I don't think its required that you tie a C state to a specific sleep type.

Each sleep state just takes longer and is more complex to come back out of.

But how could a CPU "not execute anything" without being powered down?

I imagine it'd behave like a pipeline stall. There's no new instructions to decode or run, so it just sits there. Probably can put itself into C1 at that point, if allowed.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

Wikipedia said:

The execution stage of the pipeline must always be performing an action. A bubble is represented in the execution stage as a NOP instruction, which has no effect other than to stall the instructions being executed in the pipeline.

Emphasis mine, although I've never gotten this far into it before.

They all watch too much MSNBC... they get ideas.

Thomas Fjellstrom
Member #476
June 2000
avatar

I honestly don't know how its actually implemented. But hey, the C1 state exists, and at that point nothing is running.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Arthur Kalliokoski
Second in Command
February 2005
avatar

The way I understand it, a CPU has a clock (the RDTSC cycle count clock), and its purpose is to "halt" everything at certain nanosecond intervals to allow the electrical impulses to stabilize to high and low states before the next sequence of microcode starts a new wave of activity to continue processing. I'd guess that a HLT instruction stops this clock, but the assembly language NOP instruction is simply an assembly language alias for XCHG AX,AX. I don't think that's what the pipeline stall NOP is, but I can't see how circuits could easily be set to ignore the "start" impulse from the clock.

They all watch too much MSNBC... they get ideas.

Kris Asick
Member #1,424
July 2001

Wow, I really got you two going. ::)

There's actually more C-states than just 0 through 3, however those first four are the most important ones. Some BIOSes allow each state from 1 onward to be enabled or disabled. My BIOS only allows the enabling or disabling of C1 and C6. I believe C6 is specifically for sleeping individual cores when they're not in use, but my cores are constantly all in use so I doubt it's actually doing anything on my system. Plus, don't quote me on that one as I'm not 100% sure.

Some people swear that you need to disable the C-states for performance, others say that you should leave them on since sleep mode won't function properly if you turn them off. Enabling or disabling them is more important for overclockers in order to keep voltage changes from being as dramatic or as frequent, but for typical users there's no reason to turn them off at all, unless you intentionally want to handicap your system. ;)

The point to this whole thread though is that my CPU idles noticably cooler yet performs better than it did before by turning OFF AMD's trademarked power saving feature and I still find that ironic and stupid. :P

--- Kris Asick (Gemini)
--- http://www.pixelships.com

Thomas Fjellstrom
Member #476
June 2000
avatar

The point to this whole thread though is that my CPU idles noticably cooler yet performs better than it did before by turning OFF AMD's trademarked power saving feature and I still find that ironic and stupid. :P

Could be anything really. CPU glitch, BIOS glitch (most likely).

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

Kris Asick
Member #1,424
July 2001

Oh wow... this just got EVEN BETTER! :D

I just started doing some video conversions for the next episode of my web show. They're processing at the usual speed... but using a THIRD of the amount of CPU power! :o

So... that actually has me partly concerned. It means my codecs aren't configured for proper multi-threading, since only one core is being utilized for the conversion right now, but at the same time, when I had AMD Cool 'n Quiet enabled, ALL cores would be used, CPU usage would hit an overall 67% instead of the current 22% overall, yet would process at the same speed...

Well, the good news is that now that I know this is going on, I can attempt to find the right settings to utilize more cores and get more speed out of my video conversions. The bad news is that it further proves how power-hungry AMD Cool 'n Quiet really was. :P

--- Kris Asick (Gemini)
--- http://www.pixelships.com

Thomas Fjellstrom
Member #476
June 2000
avatar

The bad news is that it further proves how power-hungry AMD Cool 'n Quiet really was. :P

It really doesn't. I have never heard of this sort of thing ever happening. I'd have expected something as annoying as this would have blown up all over the internet by now.

--
Thomas Fjellstrom - [website] - [email] - [Allegro Wiki] - [Allegro TODO]
"If you can't think of a better solution, don't try to make a better solution." -- weapon_S
"The less evidence we have for what we believe is certain, the more violently we defend beliefs against those who don't agree" -- https://twitter.com/neiltyson/status/592870205409353730

 1   2 


Go to: