Talk:CPU power dissipation

From Wikipedia, the free encyclopedia

This is the talk page for discussing improvements to the CPU power dissipation article.
Article policies

Contents

[edit] opteron request

hello... no opteron data? couldn't find some. but i would be interested how much power my two dual opterons (265) uses - i fear its near to a electric heating... (luckily i still can't put them to full load with my normal usage :-) 194.138.12.144 15:46, 3 May 2007 (UTC)


opterons in my experience run surprisingly cool - mainly because they are very good in the idle state. Speculatrix (talk)

[edit] Improvements

This article is useful but what's pretty horrible is that so many of the tables are layed out differently with different columns often in different orders, units are mixed up (should we stick with MHz throughout?), some have process geometry some not. arrggghhhh! Speculatrix 21:45, 27 January 2008 (UTC)

This really is an ugly looking article. As earlier comments here have pointed out, no mention as to the operational conditions for the given power consumption figures are given. Are they maximum power dissipation? Mean? Typical? Minimum? The statistics in this article are next to worthless without this. Additionally, the GPU section needs to go away. GPUs are certainly NOT "special-purpose CPUs" despite the unfortunate similarity of the acronym. GPUs are much more akin to stream processors or DSPs. If nobody else wants to pick up the task of cleaning this up some, I'll do it after I finish work on the CPU article. -- uberpenguin 13:05, 20 October 2005 (UTC)

Also, the article should probably be moved to "CPU power consumption." Electricity isn't "consumed" but electrical power is dissipated due to the flow of electrical energy. -- uberpenguin 13:26, 20 October 2005 (UTC)

(I agree if it is really the same thing it would be best there - I thought this was related to heat dissipation of CPUs when I first got here - unless someone really knows what TDP is I would guess the would make the same initial assumption - which is not helpful. -- MatthewKarlsen 11:28, 7 March 2007 (UTC) )

It would also be good to find some power dissipation figures for earlier discrete component transistorized computers like the PDPs. What would be cool is to redo the power consumption graph to show a broad range of CPUs' power consumption over a number of years, to show the exponential decrease. -- uberpenguin 13:39, 20 October 2005 (UTC)

The multicolored bars in the graph need work. For me, the big problem is that the coloring of the top of the bar doesn't match the more dominant coloring along the side of the bar nor help in any way to give a 3-D feel to the graph. Visually, it appears more to be two side by side bars than a single bar. -- KarlHallowell 18:02, 15 November 2005 (UTC)

Are the wattage numbers intended to be typical or maximum ones? The entry should make this clear. Tannin 22:55 Apr 17, 2003 (UTC)

That Wattage numbers are Thermal Design Power, and they have nothing to do with power consumption. They are understimated for intel CPUs, and overstimated for AMD ones. Here there are some power measures: http://www.xbitlabs.com/articles/cpu/display/athlon64-x2-3800_3.html

If there really is a difference between TDP and Power Consumption could someone spell it out clearly because atm I can't see the difference. MatthewKarlsen 11:25, 7 March 2007 (UTC)

On Thermal Design Power: Silent PC Review, shows CPU power consumed by Intel and some AMD CPUs with the PC running Seti@home software. The number-munching required by Seti or any of the other @home projects exercise the CPU, tending to push it towards conditions of maximum heat and power input. This article roughly correlates Intel CPU consumption to be around 10 to 15% higher than the TDP. It says that AMD's definition of Thermal Design Power 1. is under maximum conditions of TCASE Max, IDD Max, and VDD=VID_VDD and 2. includes all power dissipated on-die from "VDD, VDDIO, VLDT, VTT, and VDDA." None of which helps a consumer build a green PC or otherwise budget power supply unit energy. The power supply should incorporate CPU power in addition to the TDP.

What may also be of interest is what happens to CPU performance when the CPU exceeds the TDP point, since different vendors approach mitigation differently. Intel gradually lowers supply voltages, while increasing currents when TDP is reached.

On Thermal Management: The thermal management problem is exacerbated with dual and multicore processors and GPUs. Here it has been found that solutions require monitoring and control closer to the die and at different points of the die. Smaller, nano-engineered on-die film and thermal-sensing diodes are used. Novel approaches include liquid cooling and heat-conducting manifold-looking pipes that transfer the CPU sink's heat to the PC chassis directly, without air, liquids, or fans, e.g., the Zalman PC case [1].

Intel has an article on thermal solutions with its dual core mobile Centrino CPU in its Intel Technology Journal. At the risk of totally widening a discussion of CPU power dissipation, an integrated systems approach is valid. So managing system power with Advanced Configuration Power Interface using the OS, BIOS, and motherboard will indirectly lower CPU dissipation through peripheral power management, while directly controlling the CPU's sleep and hibernation states.

DonL 22:15, 9 December 2006 (UTC)

Any thoughts on how to improve the formatting of this article? I've added a bit, and have a lot more to do since this article is very lacking in completeness, but before going much further I think it really needs to be cleaned up with consistent formatting of the data, and a better way to display the information instead of such a list-like format, the article will grow to be very "long" once updated. Ideas on how to condense it a bit, but make it complete and keep it functional?

Another formatting issue to me, is how to deal with a CPU that has multiple speeds available, but different voltages and respective power consumption. Example:
CPU 100 MHz, 2 V, 10 W
CPU 150 MHz, 2 V, 15 W
CPU 150 MHz, 1.7 V, 14 W
CPU 200 MHz, 2 V, 20 W
When the second 150 MHz is put in, and there are many CPU's like this, it seems to break up the predictable flow of information, how can we deal with this in a more appealing way? -- MisterTroy 16:58, 12 February 2007 (UTC)

[edit] Content out-of-scope

I've read through this article, and it seems that some of the content is just outside of its scope. I mean, this article is about CPU electrical consumption; why, then, is there any mention of peripherals? I say we delete the headings for Hard Disks, PSUs, and Peripherals, to start. The GPU headings should stay, since, as the article mentions, they really are specialized CPUs. -- EagleOne 02:04, Nov 1, 2004 (UTC)

[edit] ATI's Radeon 9700 Pro

I've found some info about ATI GPUs, added to GPU section. The source is:

http://www.challenge.ru/video/HIS9600vs9800Pro.html (in Russian)

[edit] Graph

I have my doubts about the graph at the bottom, or at least about the descriptions. Does a "Pentium 4 Mobile 715" processor exist at all? AFAIK the Pentium M is based on the P3-design and also the Pentium M 715 has a clock frequency of 1.5GHz (as mentioned on this same page), not 2.25GHz.

It is indeed incorrect, but with no citation or source there's no way to fix it (short of generating a new one). I say scrap it, it's outdated by over two years now. (No X2, no turion, no Core/Core 2, no opteron or xeon. Several of those cpus aren't even available anymore.) Nor does it mention minimum/idle/peak as some review sites do. It's effectively worthless today.

[edit] Add MORE

This article should include much more -- pre-Pentium x86, and current $100-laptop candidate cpus. Current content should be consolidated. 69.87.203.215 21:49, 22 November 2006 (UTC)

[edit] 8088-80486 missing

This article has gotten bogged down in current details, and is missing the whole history of 8088-80486, covering about 15 years! And, there is no "Computer power consumption" article it seems. So this article should be expanded to cover year introduced, clockspeed, cpu wattage, throughput, and typical total system wattage -- or we need to start another article to cover the bigger view. At least, add links here to such info!-69.87.202.60 13:19, 17 May 2007 (UTC)

For earlier CPUs, see List of Intel microprocessors. All the list material is already covered elsewhere for the CPU. The larger view is covered somewhat in computer power supply, quiet pc, and articles on cooling and power management. (Which doesn't mean that establishing one place to tie the power consumption aspects together is out of place.) Zodon (talk) 08:19, 10 May 2008 (UTC)

[edit] 'Efficiency' (Mhz/Watt) values removed

I have removed a column from the pentium M table that gives the 'efficiency' of the chips in MHz/Watt. From an engineering standpoint, the efficiency of a CPU is zero because 100% of its power consumed turns into waste heat. My apologies to the person that did the work putting that in but megahertz/watt is not a measure of efficiency in any sense.--ChrisJMoor 23:08, 9 June 2007 (UTC)

Based on your explanation, it sounds like you did a bad thing. MHz/Watt is obviously a useful, though crude way of characterizing cpu chips: Fast ones use more power. But do they use proportionally more power? Does it make sense to use more, slower processors? You have applied your own personal meaning of the word "efficiency" as a justification for deleting useful information. You should have renamed the parameter if the title bothers you, not removed good data. Someone should revert your change.-69.87.199.53 22:57, 12 June 2007 (UTC)

I'll redo it without using 'efficiency' --MatthewKarlsen 21:14, 26 August 2007 (UTC)

[edit] Total system power consumption

This article has very good data on typical total power consumption of modern computer system subsystems under varying conditions:

  • Saving Power on idle PCs Idle represents 69 to 97% of total annual energy use, even if power management is enabled. Recommendations for Tier I ENERGY STAR Computer Specification, Natural Resources Defense Council. David Mathog and Caltech 2005, 2006, 2007

-69.87.199.53 23:04, 12 June 2007 (UTC)

[edit] Boldface?

Why are some of the power dissipations in bold? I don't think it's clearly stated in the article. Lidnariq 19:37, 16 July 2007 (UTC)

[edit] Suggestion for Mhz/Watt

The Mhz/watt doesn't make sense when comparing dual core/quad core processors with single cores.

Maybe No of Cores * Mhz/Watt would be better? Govinddutt 08:37, 28 August 2007 (UTC)

Does it make sense at all? Where is the reference/rationalle for this ratio? Different architectures (e.g. netburst vs. pentium M) do markedly different amounts of work in one clock cycle.
For certain types of work loads, FLOPS/watt might be useful (at least it is a bogus comparison with some background to it.) Is there some similar measure for non-floating point workloads?
Zodon (talk) 10:00, 14 February 2008 (UTC)
After more reading on this, the Mhz/watt seems to be even more bogus than it at first appeared. Since the watt numbers used here are often TDP, they relate to the highest use by the highest speed member of the processor family (typically don't vary from speed grade to speed grade). So it is Mhz of this chip/maximum design power of that chip family. If it stays, needs a much better explanation/caveats/etc. Since it appears to be WP:OR, should probably just come out. Zodon (talk) 00:14, 14 May 2008 (UTC)
Nowadays CPUs integrate more and more capabilities, that are basically intendet to be part of MPUs. Because the CPU is always getting a smaller part of the processor, the power consumption of the CPU can't be described simply by the value of MHz/Watt. This value makes sense with old CPUs, but with the new ones not. The new types of CPU shoud be compared by the capabilities of the sections around the CPU. In addition to the clock frequency, the mentioned integrated L2-Cache, frequency of FSB/HT/HT3, stepping and voltage seem to be good values to decide the ratio of capability / Watt. This is based on the Wikipedia Policy "neutral point of view", "reliable sources" and "no original research". Ignoring these rules, would allow to establish a value "Benchmark XY / Watt", but it's not possible. Every benchmarking is obsolete short time after it is established. At least after two years. So, it's nothing one can count on. I'd like to keep "MHz/Watt" for the elder CPUs and the more detailed description for the newer ones. 145.254.96.198 (talk) 20:50, 27 May 2008 (UTC)
Can you provide a citation to establish that MHz/watt is not WP:OR? If a decent citation shows that then I wouldn't be particular about whether it has all the individual values. (Though as I understand it, a strict interpretation of WP:OR would require that.) Just wanting to keep something isn't enough, has to be verifiable. Zodon (talk) 06:09, 28 May 2008 (UTC)
CPUs are mainly based on the CMOS technology. CMOS technology is known for its straight line in the Diagram showing the power consumption depending on the clock frequency. Raising the frequency in a CMOS environment three times higher, produces rising power consumption by factor 3. Power consumption is equal to the power converted to heat, the power dissipation. . CPUs are a mixture of TTL and CMOS technology. The current consuming TTL technology produces a narly constant current over the frequency range, even when the clock speed is forced near zero. . . Both charateristics put together, show that, the Pentium with 200MHz (12.9 MHz/W) is more energy efficient compared to the Pentium with 75MHz (9 MHz/W). That shows the speed:power ratio column in the "Pentium" table. . . A similar expanation was given for the i486 (25 MHz, 33MHz, 40 MHz) in the c't (German magazine, issue 4/1991, page 104 and following). I will find more articles like that, but this should be an acceptable citation. . . In the reality of hardware and software combination, the ratio of "benchmark / power consumption" is reversed to "speed / power" ratio, because the technology outside the CPUs, slows down the overall performance of the system. In the "CPU power dissipation" topic, there is only taken the view on the CPU, not the complete performance of the system. . . The "clock speed / power consumption" ratio depends on information taken from the "clock speed" column and the "power" column. In my opinion, the "speed / power" ratio column with the values of MHz/watt, is not WP:OR. . Did this answer your Question? 145.254.97.85 (talk) 21:52, 28 May 2008 (UTC)
As I understand it, WP:OR prohibits even such simple reasoning as the syllogism. I am not up on my number theory to know how division is defined, but if even simple reasoning is considered original research, it seems doubtful that something as complex as division would be acceptable. So the fact that it is division does not mean it is meaningful or isn't original research. (One could just as well divide the cache size by the number of pins.)
However as I said, I think a reasonable quality citation that provides a few MHz/Watt values for CPUs would be sufficient to verify that this is a metric that is used. A citation a little more widely accessible would be nice (e.g. English WP:VUE, etc.; is the article on their website?). Zodon (talk) 23:50, 28 May 2008 (UTC)

[edit] terrible science

the second sentance needs to be a couple paragraphs: its the only scientific / non-empirical data in the whole @#$@# article. all power consumption comes from these two factors, and what they are needs to be fleshed out far better —Preceding unsigned comment added by 75.71.143.45 (talk • contribs) 05:01, December 14, 2007

[edit] Merger proposal

Most of the lists of Intel processor power consumption here are duplication of material that is in the Wikipedia lists of various Intel processors. Since those lists are more fully developed (and more informative), suggest removing the list material here, filling in any values that are missing on those lists, and just referring readers to those lists. Zodon (talk) 08:43, 10 May 2008 (UTC)

  • CPU power dissipation material would apply to non-Intel CPUs too, thus the merger is unwise. Perhaps you should move material instead and replace with references to avoid duplication. --Treekids (talk) 16:48, 14 May 2008 (UTC)
Note that the sections proposed for merger so far are just those sections that deal with Intel CPUs. However, there are also lists of AMD CPUs, and merging the list data here into those lists would also probably be appropriate. I just haven't had time to look at those lists to see if they already have TDP data.
The list portions of this article appear to mostly be duplication of effort. Duplication of information that is already covered in the lists of processors articles, as well as being available in the external reference on CPU electrical specifications. Zodon (talk) 07:35, 15 May 2008 (UTC)
I think the sections can be merged easily. Only an extra column is necessary for clock speed (MHz) per unity of power (Watt) Andries (talk) 15:29, 17 May 2008 (UTC)
Yes, seemed pretty easy. Don't even have to add the MHz/Watt, I think it should be removed because it seems pretty meaningless, and appears to be WP:OR (see #Suggestion for Mhz/Watt). Zodon (talk) 23:41, 17 May 2008 (UTC)
Well, in my opinion, the "CPU power dissipation" section provided information about the power consumption at a glance long before "green IT", "green computing" or "power saving" were topics for most customers. As far as I know, when the power consumption of the CPUs wasn't even mentioned in the detailed descriptions of CPU-Models, this section was the only source at Wiki that dealt with this big topic. . . Sure, all these information could be merged, but this is still the only overview showing different models and brands. . . In time of constantly growing prices of electricity and growing consumption of electric energy around the world, an overview like this appears to be at the right place and at the right time. As I mentioned above in the "Suggestion for Mhz/Watt" discussion, the duplication of some values seem to be necessary for the reader to understand the TDP. . . Besides, the "CPU power dissipation" section seemed to add the new value "Extended Halt Power" (EHP) at Wiki first. . . Perhaps, these arguments dosn't meet the requirements of Wiki. I haven't visited a convention. 145.254.94.208 (talk) 13:25, 28 May 2008 (UTC)
It may have been a unique resource in the past, but now it duplicates material elsewhere here on Wikipedia. If a comprehensive overview in one place is important, Processor Electrical Specifications provides it for x86 processors, and has done so for years. (An encyclopedia doesn't have to contain copies of all information, linking to external source for lists, etc. is perfectly acceptable.) There are practical limits to how large an article should be, this would be a way to bring this article down to a more reasonable size. Rather than just being long lists of data this article could describe/compare power consumption of processor families, architectures, etc.; with links to the lists of numbers for those so inclined. (i.e. it could be an encyclopedia article instead of a parts catalog.) For example, EHP numbers could be added to the list of specifications where it is appropriate, and explained/described here with actual information about its significance (rather than buried in a bunch of uninterpreted data). Zodon (talk) 23:16, 28 May 2008 (UTC)