PC Power Management, ACPI Explained In Detail

DK writes "Computer performance has increased steadily in recent years, and unfortunately so has power consumption. An ultimate gaming system equipped with a quad-core processor, two NVIDIA GeForce 8800 Ultra, 4 sticks of DDR2 memory, and a few hard drives can easily consume 500W without doing anything! To reduce power wastage, the industry standards APM and ACPI have been developed to make our computers work more efficiently. ACPI is the successor of APM and is explained in detail in this article."

Comments

ACPI?

[crankyspice]

2002 called, it wants its Page 3 tech story back.

Not worth reading this crap

[Aranykai]

Im sick and tired of having to view 11 pages of adds to read an article that could easily fit on one. Easily 6 adds per page.

The Wikipedia ACPI article is better and doesn't shove crappy adds down your throat. http://en.wikipedia.org/wiki/ACPI

Re:Not worth reading this crap

[suv4x4]

Im sick and tired of having to view 11 pages of adds to read an article that could easily fit on one. Easily 6 adds per page.

You go there to read the article? Damn, I go there to enjoy the ads, and that little article paragraph in the middle? It's pissing me off. It's right in the middle, getting in my way, demanding attention, as if I have nothing better to do than read articles all day.

Can't there be site with just the ads and no pesky articles?

And then I found this. Best. Site. Ever.

Re:Not worth reading this crap

[slugstone]

Hell NO, the electricity might leak and run across the floor. Well that what my grampa says.

How it is supposed to work

[niceone]

TFA lists all the states and how all this power management stuff is supposed to work... what it doesn't go into is how (or if) it actually does work. My experience is that it doesn't - I press sleep on my Windows XP PC ans all I get is a message telling me that the driver of my MIDI controller keyboard will not let the machine go to sleep!

And on my (admittedly very old) Ubuntu laptop the screen just blacks out for a couple of seconds and then comes back on again. When it was running windows it used to go to sleep fine, but the wireless wouldn't work when it woke up.

I guess other people's mileage probably does vary...

Re:OS

[Whiney Mac Fanboy]

anything less then a 500w psu and it'll be under powered.

Oh Bollocks. Vista might be shit and power hungry, but many laptops with a sub 100w psu will run it just fine.

No, I didn't read the fucking article.

[adolf]

ACPI has been around for almost eleven fucking years. In-depth information about it can be had in all of the usual sources, from LKML to Wikipedia to decade-fucking-old back issues of Byte and PC Magazine.

News? Where?

500W?

[achurch]

Is that why people don't blink at PS3s and X360s that eat 150-200W when they're idle? I guess that locks me and my 100W/system power budget out of gaming . . .

Seriously, what is it that uses up so much power? I've got a pretty standard dual-core system that idles at about 65W, and I can't push it beyond 150W even when I try.

Re:OS

[donaldm]

The problem with any electronic device is they (to state the obvious) consume power so manufactureres have opted for approx 1W in standby mode. Unfortunately if you take a stereo amp plus active woofer a TV, HDD DVD recorder, set-top box (if you have one) and a least one game console (assuming they also consume 1W in standby) and you have a total of 7W consumption. Now extrapolate that to 10M people (I am being very conservative here) and that is 70MW overall consumption just for your entertainment system to do nothing.

Of course once you turn on your entertainment system the power consumption (taking the above example) can easily jump to 7GW even with fairly conservative systems. Now try the same simple maths with your fridge, microwave oven, oven clock (in fact any clock) and anything else that consumes power in standby. Add in lights even low wattage ones and your hot water heater (assume electrical off-peak not gas or solar) and the power consumption is massive. With regard to PC's and laptops consumption is dependent on what you have and can vary between 20W to over 1000W, It is possible to put a laptop in standby or sleep mode but this depends on if you are using your laptop as a standalone machine.

So what are we going to do about all that wastage? Well if you pay for your electricity and you want convenience then absolutely nothing and this is what most people will do.

ACPI is a disaster

[r00t]

We used to standardize hardware interfaces. They stood the test of time, were well supported, and were low overhead. Writing drivers, including boot code, was no serious problem. We didn't need an emulator, virtual machine, etc.

Decent standards: IDE, VGA, PC serial interface, PC parallel interface, PC keyboard interface, UHCI, OHCI, etc.

Now we standardize an interface to non-standard hardware via ACPI. The OS is supposed to run ACPI code (a script) in a complicated interpreter. ACPI code is slow and buggy, and generally gets to do whatever it wants with the hardware. It's like making BIOS calls to do everything, but without even the minor advantage of native code.

This is especially painful for boot loaders. You can't run an ACPI interpreter in a 512-byte boot sector. You probably can't do it in any reasonable boot loader.

This is even painful for power management. For example, OLPC wants to suspend the CPU between every keystroke; that doesn't work so well if you need to run an ACPI code script to do it.

Re:ACPI is a disaster

[Bert64]

Whats worse is that...
There is a standard for these ACPI scripts, as you pointed out it's not great but at least there is one. There's also a compiler for them, written by Intel that complies with the standard.
But most hardware makers don't use Intel's compiler that complies with standards... They use Microsoft's compiler that completely breaks the standards, thus OS authors can't just implement according to Intel's published standards, they have to reverse engineer Microsoft's unpublished variations.

Re:ACPI is a disaster

[*SECADM]

The ACPI interpreter is designed to work within the OSPM. The reason being the assumption that the OS will always be the central point that knows the most about the whole system at any given time, and therefore it is most qualified to make power management decisions.

About your point about standardizing on an abstract interface of non-standardized hardware, consider:

1. This is nothing new, and is what always happens in the computer industry. If we were so hung up on standardized hardware interfaces that are well understood by the industry, we would still be coding all in x86/68k/ia64 assembly. Thanks to a wonderful high level (read: abstract) language (C/C++) that is standardized so we can write portable code for.
   
2. ACPI hasn't replaced any of those hardware standards you mentioned. E.g. ATA drives still take the same commands as before and the standard is constantly evolving. But more importantly, do those hardware standards have the advance power management capabilities offered by ACPI? And, if they do implement it somehow in the hardware spec, how is updating each individual hardware's firmware *and* driver to include the new power management support, better than just updating the firmware AML code to expose the new capabilities to the OS in a standardized way?
   
3. You didn't mention the other half of the ACPI standard, the Configuration aspect of the spec. All those hardware interfaces that you know and love used to each expose its own resource requirement differently, unlike ACPI which standardizes resource descriptors and boot-configured requirements. How is supporting a bunch of hardware specific resource description format better than supporting one standardized format that is well understood by the OS?
   

Finally, to go back to your OLPC example. First of all I am not sure why you would want to put an ACPI interpreter into the boot loader? Is there a reason OLPC is doing this? As for you implying that you need to run ACPI code to suspend the CPU (after each keystroke), this is just not true. C-states (which i assume you are talking about) are entered by simply reading a register for the state, as described by the _CST definition. So all you need to do is parse the definition once, and remember the corresponding registers and never parse the AML code again. Same for T-states or P-states, which are entered by writing to the proper registers. There is no such running "ACPI script" overhead like you've described when handling processor power management.

Whatever

[Anonymous Coward]

Noones computer idles at 500 Watts, not even close. I wish people would check their facts before posting nonsense.

My 4 year old xenon dual processor (Thats two physical CPUs) PC with (~10 fans) with no power management support in the CPUs idles at 200 watts including powering the display and extraneous trinkets attached to the watt meter plugged into my wall.

All new PCs with multiple cores on single processors have power management features and use concideribly less power when idling.

Whats worse is the article spouts all kinds of mostly useless techno crap about power states without providing any context into what it means or useful information in terms of actual OS power settings one can configure to do something about their PCs power usage.

/. invents time-travel

[jb.cancer]

Now tell me what ISA is..

Sleep is worthless

[paul248]

Being able to put components to sleep is pretty much worthless if you want to run anything resembling a server. Hardware manufacturers need to focus less on sleep states, and more on making components consume less power while they're active.

A good first step is the 80plus initiative for power supplies. By increasing the power supply from 65-70% to 80-85% efficiency, you gain a decent amount of active power savings right off the top. If you care at all about conservation, make sure to check the efficiency rating of your next power supply.

The people at Intel and AMD have made great strides toward power efficient CPUs, which can scale back their clocks on-demand without noticeably hurting performance, but the real remaining problem areas are in video cards, RAM, and especially hard drives.

The ideal computer would consume almost zero power while sitting there doing "nothing," but be able to wake up at a moment's notice to handle requests from the user or the network. Power management should be hardware-based and completely transparent. ACPI is just a dirty hack that's becoming more useless as network accessibility becomes more important.

Obligatory linus quote

[josephdrivein]

Modern PCs are horrible. ACPI is a complete design disaster in every way. But we're kind of stuck with it. If any Intel people are listening to this and you had anything to do with ACPI, shoot yourself now, before you reproduce.

From: http://en.wikiquote.org/wiki/Linus_Torvalds