Talk:Overclocking

From Wikipedia, the free encyclopedia

This is the talk page for discussing improvements to the Overclocking article. This is not a forum for general discussion about the article's subject.
Please sign and date your posts by typing four tildes (`~~~~`). Put new text under old text. Click here to start a new topic. New to Wikipedia? Welcome! Ask questions, get answers.	Be polite Assume good faith No personal attacks Be welcoming	Article policies No original research Neutral point of view Verifiability

漢

▶

A version of this article in Italian has been listed at Wikipedia:Translation_into_English/Italian#:it:Overclocking. If you'd like to translate it, sign your name there, or, if you'd also like the article to be translated, sign your name there under "Supported:".

[edit] external link suggestion

< cleaned up a bit, see history for details >

[edit] Article suggestion about the disadvantage of overclocking

Somebody should edit the "Disadvantages of Overclocking" section to include the categories "Disadvantages of Incompetent Overclocking" and "Overall Disadvantages of Overclocking". For example, incompetent overclockers may damage their drives, blow up CPUs, and render their systems unstable. This is not true for people who know what they're doing, as their overclocked machines are GUARANTEED to be 100% stable due to the stress testing whereas stock machines aren't. In this respect, overclocked machines are more stable than stock machines. Also the "60 to 66 frame/s" comment presents a bit of a fallacy, as you do not generally overclocking your "computer", but rather individual parts. A little bit faster here + a little bit faster there will end up giving you a significant boost.

On the other hand, disadvantages such as higher power draw apply to both categories whether you're competent or not. I really think we need to split this into 2 sections.

I agree. I'm working on that now.

Pueywei 02:52, 18 March 2006 (UTC)

I have edited and rewritten some points to make them clearer. Some new subsections have been created too, for some points listed are not really disadvantages, but more of limitations of overclocking. Please review and make any changes as necessary. :)

No, I believe that the disadvantages section is blown out of all proportion, and merely needs some parts edited, others deleted, and all of it compacted into a few well chosen sub-topics.

DOTANDCOM 18:07, 1 June 2006 (UTC)

[edit] Disadvantages of overclocking

Many of the disadvantages of overclocking can be mitigated or reduced in severity by skilled overclockers. However, novice overclockers may make mistakes while overclocking which introduces many avoidable drawbacks.

[edit] General Disadvantages of overclocking

These disadvantages are unavoidable by both novices or veterans.

The lifespan of a processor is negatively impacted by higher operation frequencies, increased voltages and heat. However, the effect of operation frequencies and voltages has not been proven to be a major factor in processor lifespan. An experienced overclocker would have taken steps to ensure operation within safe temperature ranges. Morever, with the rapid obseletence of processors coupled with the long life of solid state microprocessors (10 years or more), it is argued that the processor will be replaced before any threat of failure.

--Comment - See http://www.google.com/search?q=cache:El49Z0vfWS8J:www.techspot.com/vb/all/windows/t-5191-How-often-do-you-shutdown-your-computer.html+electron+migration&hl=en&ct=clnk&cd=10&client=opera Quote " the main cause of failure for semiconductor devices (assuming that they are operating within range while in use) is electron migration (much as for lightbulbs that are kept ON). Electron migration is effected by current flowing through a device, and will happen more quickly when the current is high. I know this to be a fact, as I used to carry out defect investigation of electronic components when I worked in the defence industry. I examined hundreds of components, and each time it was components that carried most current that failed first, even if they were not operating anywhere near their limit.

All things wear out, even electronic components, but it takes a very long time for these to fail due to wear. Also, you might wish to consider how CPUs and even Computer PSUs work. By their very nature these devices are constantly being switched ON/OFF (transistors) billions of times each second (in some cases) and yet they don't fail. Ok so they aren't being heated and cooled constantly, but really, semiconductors are not lightbulbs and they expire mainly due to usage (electron migration), and operation at high temperatures (cpu - an increase of 10 deg C will cut component life in half - approximation, it varies) rather than hot/cold cycles."

Among other sources.

Increased clock speeds and sometimes voltages results in higher power consumption and a higher power bill.

- Comment : More serious in hot regions where air conditioners (for cooling) are commonly used. It is generally accepted (sources from data center managers) that the increased thermal load in watts x 3 = total increase in power consumption due to cooling system ineffencies.

Could be mitigated by clock throttling - see: http://www.google.com/search?q=cache:p1mEUrOBRVoJ:www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/Power_Cooling_Datacenter_34146A.pdf+datacenter+increase+thermal+load+increase+power+watts&hl=en&ct=clnk&cd=3&client=opera

AMD PowerNow! technology enhances the energy consumption performance of these devices with multiple levels of lower clock speed and voltage, reducing power consumption by as much as 43 percent while under an approximate 60 percent load and as much as 75 percent during idle times

The clock rates may be dynamically adjusted via utilities like RMClock, which detects system load and adjustes the multiplier and Vcore accordingly. Example: I'm running an Opteron 146 at 1175Mhz@0.825v now. This would consume less power and produce less heat than its stock 2Ghz@1.35v. I can ramp it up to a full overclocked speed of 2.95Ghz@1.425v when gaming. However -> More work, requires stability test for all permutations/steps.

While systems may be throughly tested for stability before usage, stability problems may surface after prolonged usage due to an unusual workload or untested portions of the processor core. Although this is rare, such incidents may result in data loss.

Not every component of a computer is overclockable; for example, hard drive platters can have an increased rpm rate by overvolting the drive, but this often leads to the destruction of the drive; or hard drive corruption. Where hard drive read/write rate is the bottleneck, as it often is, overclocking brings little or no speed advantage. There are also cases in which overclocking is possible, but the risk of doing so is unacceptable (as in the hard drive example, where any possibility of data loss cannot be tolerated).

-- Changed to drive motor - the drive would likely has integrated regulators to cater to varying voltages that various psus supply.

[edit] Disadvantages of Overclocking by Novices

Increasing the operation frequency of a component increase its thermal output in a linear fashion, while an increase in voltage causes an expodential increase. Furthermore, physics dictate that higher temperatures results in a higher resistance. Improperly managed, a novice may cause chip temperatures to rise so quickly that a permament decrease in life expectancy or irreverisible damage is caused to the chip.
With the advent of ever wider ranges of voltage options on motherboards, the risk of fire or burns is not tiny. The chip itself, or power mofsets may burn and capacitors may burst.

--Comment For example some a64 motherboards include the option of vCore of 2v, HTT of 400, vMem of 4v, etc.

202.156.6.54 04:04, 18 March 2006 (UTC)

[edit] More changes

The article was:

The main aspect of overclocking is the need for more effective cooling than that of the inefficient air-based cooling systems which come packaged with typical CPUs and graphics cards. High-end, specially-designed, copper heatsinks are often used with powerful fans for better cooling. Liquid (usually water) cooling is another popular method, which uses liquid as a coolant because it is more efficient conductor of heat than air is. Instability is a major danger of overclocking, although most overclockers take some precautions to check their systems for stability. The process of stress-testing a system is often called burn-in, and it is common to run several applications simultaneously or special burn-in application that place a high load on the component being tested.

I have changed the items in bold.

Firstly, "the inefficient air-based cooling systems" is borderline non-NPOV, but more importantly, it's unclear what is meant by 'efficient' in this context. Stock air cooling uses very little electricity, which is what I would consider efficient, in the same way that a small, unpowerful car is efficient.

Secondly, solids are a much better conductor of heat than liquids, but this is besides the point - air and liquid are used as convectants, not conductors. Liquid cooling removes more heat because it can move the heat to larger heatsink which is usually air cooled. The water is used primarily to transport the heat. 62.252.192.7 12:15, 26 Dec 2004 (UTC)

[edit] Request to add more advantages

Could someone add some more advantages? This article is starting to be a little one-sided! 62.252.192.7 12:29, 26 Dec 2004 (UTC)

[edit] Sysextreme link spamming

I've reverted the edit that put sysextreme.com in the major forums list.

Alexa ranks this site at about 4,500,000th most visited website, while others in this section are about 16,000 to 40,000th. With less than 1% of the visits the real major forums get, Sysextreme is not big enough to be listed here. WikianJim 16:46, 26 Apr 2005 (UTC)

- They now are using "XMS.MS" to try to promote their site.

[edit] XS Spam

I've reverted the edits by "FUGGER" because of blatant spam. He owns XtremeSystems (not to be confused with sysextreme.com). Putting "These and more top names can be found at XtremeSystems.org (linked below)." in the intro of the article is spam. Furthermore, adding the link to the XtremeSystems homepage as an overclocking resource when the home page has very little (or none at all?) original articles about overclocking and contains only links to other sites or the XtremeSystems forums. Therefore, having a link to XtremeSystems in both the links of resources and forums is redundant. Also, removing XtremeResources.org from the forum listings is debatable as the site has multiple members in the top 10 in various versions of 3dMark and the forum provides a number of overclocking tips. I just wanted to clarify why I reverted the edits, in case people were wondering. - PS2pcGAMER

I have been talking with "FUGGER" via Talk pages and email, and I we have both agreed on some ideas... Notable names in overclocking SHOULD stay at the end of the article, and FUGGER is no exception. I suggested that he give points at the end of the links, and cite particular web articles instead of google searches, to which he also agreed. Lastly, I suggested that he make an article that curtails Overclocking breakthroughs and their record owners, and wiki-link that instead of just posting notable names at the end of the article. I haven't heard from him on the last bit, but this is all being cleared up, so no worries. This guy was kinda "spamming", but he was simply trying to put content back into the article that was originally there.

Fair enough. I just thought his additions were not in good taste and they weren't really beneficial to the article. I will also re-add the link to XtremeResources to bring things back to how they should be. Hopefully that will be the end of this. - PS2pcGAMER

[edit] External Links

I've removed several external links. Some of these links in Overclocking are getting ridiculous. There was even one to a forum that required registration prior to viewing. I don't think there's need for links to more than one or two overclocking fora. This is NOT a directory of overclocking forums. Excercise some restraint in your self-promotion, please. PManson 03 Sep 2005

More link SPAM today. Xtremesystems.org seems intent on spamming on this and other pages (like FSB). They are doing it in multiple IPs and going back for quite a while through the history. Each time someone removes the link a new IP pops up and repaces it. This page - and other OC related pages like FSB, water cooling etc - need even more careful watching. PManson 04 Sep 2005

I have removed the notable overclockers links. Yes, notable overclockers should be mentioned somewhere in this article, just not under this section. Linking to a google seach in the "External Links" section is completely pointless. The first few results aren't even related to their own personal overclocking achievements. Example, the first link for "FUGGER" on google links to Ricky's 4ghz overclock. Besides, anyone can use google. Think before you add links. PS2pcGAMER Sept 04, 2004.

1. I've removed the overclocking targeted manufacturers. Almost all mobo manufacturers now cater for OC, as do a million other companies from Zalman to AMD itself. There's no need to isolate two mobo makers for listing.

2. This page is getting more and more in need of a clean up as it says at the top of the article. I'm tempted to take the job on and, if I do, a lot of stuff is going to get chucked out. Including some of remaining ext links.

3. I'm glad the links to SERPS results are out but am not sure that nicks of specific overclockers need a mention in the article. Any names mentioned here could quite easily be gone tomorrow with someone else taking their place. I mean, we aren't talking Nelson Mandela/George Washington kind of eternal fame. I'll let them stay for now but any cleanup - whether done by me or someone else - is likely to see those names all removed. This is the kind of thing that's lowering the professionalism of the article.

PManson 05 Sep 2005

I learned overclocking by reading the stickied guides in the Overclocking section of the OC Forums. They're excellent guides and far more relevant than some of the other links that are on there...

The external links section is getting swamped again. I am going to remove all general computing sites and only list sites that are devoted to overclocking. I have also removed a few dedicated overclocking sites: OCAU because they require registration, Hardware Asylum because it is relatively small and has lots of crossover with XS, OverclockersOnline because it doesn't have the depth as some of the other sites, Mac Xtreme because it is non-english (an English mac overclocking site would be great though), EOC and overclock.net because they are similar to ocforums, but not as big. Basically there is one of the following, a general overclocking website (OCForums) which has great depth on air, water, and the "less" extreme overclocking methods, an overclocking news site and this is kind of debatable if it should be left (OCWorkBench), a site with a few guides (OCInside.de) and finally a site devoted to the more extreme aspects of overclocking (XtremeSystems). Wikipedia is not a website for promoting your own personal website. IMO, the external links should be directly relevant to overclocking and not general computing sites with just some overclocking content. PS2pcGAMER 03:55, 29 November 2005 (UTC)

[edit] External links, Round 2

Anonymous IPs keep adding and removing external links without explaning why. This is getting tiresome as it is just the same links that are added and removed.

Is it time to be more strict about this sort of thing? On the Sudoku article, there is an explicit note that all external links must be discussed on the talk page, or it will be reverted, becuase of all of the spam. I feel that we are on the verge of having to follow a similar policy for this page. Do we need to come up with a short list of external links that should remain and any new ones most be discussed on the talk page first, or should we remove all of the forums from the list? --PS2pcGAMER (talk) 06:33, 15 January 2006 (UTC)

[edit] External links

I think the external link section of this page needs some cleaning. It is my understanding that linking discussion forums are generally discouraged in Wikipedia, and even if they are not, there are too many forums listed. This is not a web directory. --BorgQueen 15:56, 19 March 2006 (UTC)

I agree completely. I say remove them all. Unless there is something especially noteworthy about the link, it doesn't belong there. If we have just one forum listed for no perticular reason, it will just encourage people to add other forums and we'll get a web directory again. --PS2pcGAMER (talk) 22:28, 19 March 2006 (UTC)

Done. Now let's find some truly informative sites which we can use to improve this article :-) --BorgQueen 22:34, 19 March 2006 (UTC)

I think it make sense to keep some good pages like OCworkbench (most OC Mobo Reviews, etc), OCinside.de (interactive guides like pinmod, etc) and OCIA.net (overclock guides).

why is OCIA.net singled out to be one of the few "OC related sites" to be added, there are 100 of sites like it which offer OC guides, would be better to link to guides which cover in-depth most aspects and were used to support the article. e.g. : AMD Athlon 64 Overclocking Guide , OCIA lists OC experiences on how they build a system uses some components and shows photos & specs; there are 1000 of such experience articles everywhere, a guide should cover the basics, don't limit to one product or brand and should be able to be used with no matter what HW combination the guide is written for. my 2 cents. --Jmke 12:56, 27 July 2006 (UTC)

But it doesn't make sense to keep non overclocking pages like Ars Technica (move them to the CPU pages), or pages like the new electromigration one.

I would suggest two pages: OverClocked inside and OCWorkbench because the offer the most overclocking basics.

But it also make sense to keep the "Overclocking/Benchmark databases" Futuremark Benchmark Database and Comprehensive CPU OC Database because these are the most known databases.

[edit] Factual inaccuracies, general poor quality.

This article is the worst i've read yet, i've corrected one mistake but i'm finding so many little things and improper writing, this article needs improvement which is beyond my skill. It has the quality of an average post on a hardware forum.

Highlight: "hard drive platters can have an increased rpm rate by overvolting the drive motor" Simply not true, harddrives use synchonous motors.

On the bright side, it seems fairly balanced on the pros and cons.—The preceding unsigned comment was added by 82.156.167.207 (talk • contribs).

I agree, the article really does need a makeover. Maybe it should be scrapped entirely and started over? Whatever is decided, we need someone with the knowledge and time to take this challenge on. By the way, to sign your comments, you can just put ~~~~ and the wiki software will automatically insert your name (or IP in this case) and the date. Cheers, PS2pcGAMER (talk) 02:11, 21 March 2006 (UTC)

Yes, I do agree. However, scrapping it and starting over seems a little excessive. Just junk the unverified or dubious stuff? Btw, what is the target audience of this article? Pueywei 10:52, 31 March 2006 (UTC)

I suggest bringing in the Overclock.net crew to do a major re-write lol. OmniAngel 14:59, 22 March 2006 (UTC)

I think that the article as it is really isn't that encyclopedic -- it delves down too much into specifics of things. I'd suggest removing most of the hardware-specific bits -- e.g., the last third or so of "Measuring effects of overclocking" -- in favor of a general overview of the subject. Any hardware-specific information is quickly going to become dated. Efindel 16:34, 8 May 2006 (UTC)

I agree that this article needs a lot of work to bring up its quality. I'll take a stab at this later today. --Zsinj^Talk 15:43, 19 May 2006 (UTC)

I corrected a couple of blatent factual errors. The tone of the article is still bad, though. It doesn't even bother to explain WHY heat can lead to instability (think: carrier density in semiconductors versus temperature, and more importantly, temperature-sensative resistors). It also doesn't make mention of the fact that clock periods cannot be decreased beyond the point of worst-case delay in an IC. This is rather fundamental since it partially explains sporadic malfunctions in synchronous ICs run outside spec. -- uberpenguin @ 2006-06-02 16:06Z

I copy edited the article a bit to improve some blantantly bad grammar errors along with removing the list-like feel to the article. In one case I deleted an entire paragraph (details were getting far too specific in terms of which programs to use). This article definatley needs an expert or if not an expert someone who has the time to find a ton of sources for the information given. I would agree with Uberpenguin that a scientific approach needs to be taken in at least one section of the article. As of now it seems to be more of a guide to overclocking and good programs you might use to do so, rather than an encyclopedic article.--SomeStranger (T | C) 02:02, 4 June 2006 (UTC)

[edit] structure

I've taken another cleanup pass. I resolved the {{expert}} tag, and restructured the article. I'd like to pull out the section I called "root cause" and move it somewhere more introductory, but I stopped short of that.

I can't figure out why the "The difference between"Unlocking" and "Flashing" a video card" section is in this article, since it's got nothing to do with overclocking. Isn't there a better home for this text? -- Mikeblas 10:07, 18 July 2006 (UTC)

[edit] Regarding the source of instability in hot processors

I notice that there's been some disagreement about the source of instability in processors which are overclocked and running hot. Conventional wisdom passed down to me has always been that transistors run slower when hot - at least in my process corner simulations I've always found that regardless of layout effects related to wires, the transistors themselves switch more slowly when the temperature is increased. As I understood it this was a materials concern - when at higher temperatures, the channel resistance increases and the current transmitted decreases. Here's a link which seems to back that up. I have read that in recent microprocessors the overall logic delay is reaching 80% due to wires, with only 20% due to transistors. However it was still my understanding that the majority of the delay change due to temperature was due to devices and not wires. Is that the question which we're in disagreement about or can someone rephrase? - McCart42 22:54, 13 September 2006 (UTC)

You're more or less addressing the issue. It is definitely true that the channel resistance and saturation current change with temperature (mostly due to threshold voltage changes). My contention has been that I cannot identify any reason why a MOSFET should slow down significantly in the small temperature ranges we're talking about. I'll admit that I'm definitely not a digital logic person. My circuit experience is with analog electronics and my primary semiconductor physics experience is with heterojunction bipolar devices. I could very well be overlooking some important effect that would change the time it takes to drive an independent MOS system from weak to strong inversion as a function of temperature.

That being said, so far I've seen no compelling reason to believe that this is the case. I've been presented with SPICE circuit simulations that indeed demonstrate the dependence of logic delay on temperature. However, I feel it's a logical leap to suggest that this circuit behavior necessarily indicates that individual transistors' "switching" speed is slowing down. While that is one possible explanation, I don't think it's the case for reasons I've described (weak parasitic capacitance temperature dependence, significant temperature-dependent threshold voltage effects). There are still a lot of factors like channel resistance, saturation current, and especially interconnect transmission line effects that could equally well describe the end result of increased logic delay.

I am interested to know if anyone can find a good explanation of logic delay's temperature dependance, and in some of the preliminary reading I've done, channel resistance and saturation current do seem to play some role. I'm still not entirely sure why this is the case since well designed MOS systems require trivial amounts of gate input current, but a couple of papers I've skimmed that address temperature effects on highly scaled CMOS talk a lot about saturation current. It also would make perfect sense if interconnects are playing a large role since they are non-trivial sources of complex impedence at the frequencies common in a modern microprocessor. However, I really don't desire to make a huge deal out of what amounts to one sentence. I changed the phrasing to something that I think we can all agree upon. It expresses the original point for the purposes of this article, but doesn't suggest that individual FET switching speed is necessarily the cause. -- mattb @ 2006-09-13T23:28Z

For what it's worth, I am a digital logic person.

So we're all in agreement that some part of transistors is causing transistors to slow down, right? It sounds to me like the original wording that specifically blames the FETs can go back, since it provides more detail than just "propagation". If someone said, "the sun is bright", that would be an accurate statement even if the person doesn't know the reason.

Back to my spice circuits, the only objects in the ring oscillator were transistors. There wasn't anything else. Therefore, the transistors are getting slower for definitions of "slower" used in ordinary digital logic designer discussion (specifically, that given an input transition, the output node crosses Vdd/2 later, whether this is due to the transistor not turning fully on as quickly or whether it's due to reduced Ids). As McCart42 said, channel resistance increases, so even if all of the capacitances stay the same, it will take longer for the gate cap of the load devices (i.e. inputs of the next level of logic in a path) to charge and discharge. As I've said before, there are no transmission lines in this circuit simulation. However, I think I now understand where you're coming from.

As to "why this is the case since ... MOS systems require trivial amounts of gate input current", the answer is simple. MOSFETs switch based on gate voltage, rather than base current. There is some gate capacitance in the devices, and this capacitance has to be charged or discharged past the threshold voltage for the device to switch. As Ids falls, the first logic gate in a chain charges the input caps of the second logic gate in the chain (plus its own drain diffusion caps and any parasitics in the circuit) more slowly, so the gate voltage at the second gate changes more slowly... it takes longer to cross the threshold voltage, resulting in the second gate switching later. Now, the "switching point" for an actual gate (e.g. an inverter) actually depends on the beta ratio and all that fun stuff, because there's a region where both the pfet and nfet are on, and fight each other, but hopefully it makes sense that if the transistors making up the gate switch later, the whole gate gets slower.

Coming from the analog world dealing with BJTs, you're used to transistors that are actually controlled by the current, and even a 30% drop in saturation current is not going to matter too much since you only need like 1% of that current going into a base to switch the next transistor in your circuits.

I drew a picture showing a very simplified MOSFET-based inverter chain to hopefully clear up how I'm saying these things work: http://ctho.ath.cx/pics/mosfetInverterChainDelay.png The switches open and close based on the voltage across the capacitors. As the resistances increase, it will take longer for the first transistor to charge cap c2, and thus take longer to cross the threshold of the second transistor's FETs.

I hope this all makes sense.

One last thing I'd like to touch on: McCart42 mentioned that a lot of delay is in the wires nowadays. That's true for some paths, but not true for a very large number of paths. When it comes to temperature and overclocking, I'd actually expect the wire-dominated paths to be less of a problem than the gate-dominated paths, since the gate delay is going to increase so much more than the wire delay. --CTho 03:18, 14 September 2006 (UTC)

Okay, now I too understand where you're coming from. However, I still totally disagree with adding text that claims that the FETs are going "slower". As you just explained, the circuit slowdown is due to increased resistance in the channel of the previous device, which causes the gate cap to charge more slowly. That means that the delay increase is as a result of the way the gates are being used in a circuit, not because the transistors themselves "slow down". I think this is merely a terminology issue because "speed" isn't something that's generally well defined for a single transistor. There are parameters like cutoff frequency and such, but those don't translate to speed. Now, your claim is that the usage of the term "speed" as you're familiar with it in this context is analogous to "digital logic propagation delay". I can accept that, but I think it's therefore most appropriate for the article to say "propagation delay", not "MOSFET speed". Propagation delay is a circuit factor and is not well-defined in the context of an individual FET. In other words, the circuit is getting slower inasmuch as delay is increasing, but the individual transistors are simply passing a little less current through their channels.

In this case, your hyperbolic "the sun is bright" simplification is probably appropriate. A real explanation of what's going on here requires a paragraph about channel resistance as a function of carrier mobility and threshold voltage, which are themselves functions of temperature. That discussion may be appropriate somewhere (like, say, a section in propagation delay), but I really don't think it's important to talk about transistor operation in an article where it's more than sufficient to say "temperature increases propagation delay." There's always the question of "how much is too much to explain". For an article on a topic that really has little to do with circuit design, I think that the very brief explanation is the best. If we were to explain mobility and threshold voltage, where do we stop? Do we explain thermal voltage? Surface band bending? Majority carrier type inversion? The quantum mechanical effects behind effective carrier mass and mobility? You see where I'm going with this?

Incidentally, you don't need to explain how FETs work in laymens' terms for my benefit (kudos for a good simple explanation, though I've never personally liked the switch simplification even for digital contexts). I've done some MOS design and fab and have worked a good bit on HEMTs (type of FET built out of heterojunctions), so I'm well aware of the physics of FET operation. I was actually thinking along different lines when I asked that question, but I expressed myself pretty badly and I can totally understand why you thought I was confused about device operation principles. Sorry for the confusion there, but I assure you that I know how MOS systems work.

P.S. - I'm interested to know what the "beta ratio" you spoke of earlier is. That doesn't sound like a transistor parameter since the closest thing in transistor-speak is bipolar transitor beta (common-emitter current amplification; a function of emitter injection efficiency and base transport factor). I assume it's a digital logic term relating to the points where the output is momentarily undefined? In any case, this has been interesting dialogue. I now have a new factoid to file away that explains why temperature increases digital propagation delay. -- mattb @ 2006-09-14T04:09Z

So propagation delay is of course what we've been referring to when we speak of "speed"; at least that was always my assumption. It's certainly now clear that this was an analog-digital disconnect. When I think of speed in a digital context I generally think of half-VDD rising on input to half-VDD falling on output (propagation delay), whereas 20% VDD rising to 80% VDD rising on output could also be considered a measure of speed. To clear up what CTho meant by beta ratio - beta ratio is the ratio of PFET resistance per square to NFET resistance per square. This winds up being the same ratio as the PFET W/L to NFET W/L for equal rise and fall time. - McCart42 05:13, 14 September 2006 (UTC)

Now that we're all on the same page, I don't really like the current wording of the paragraph, because the way I read it, it sounds more like wire delay is the main factor (since that's where I imagine signals propagating). How about this? It specifies what part of a circuit gets worse by the largest amount (the FETs), yet describes the slowness as arising at the circuit level, which fits with a definition of "slow" I think we all agree on now.

Due to changes in MOSFET device characteristics, circuits slow down at high temperatures. Since overclockers aim to operate circuits at higher frequencies, it is critical to keep delay from increasing to the point where data cannot propagate completely within a clock cycle. Wire resistance also increases slightly at higher temperatures[1]; this is a small additional factor contributing to decreased circuit performance.

--CTho 23:28, 14 September 2006 (UTC)

That verbiage is perfectly agreeable to me. Go for it. -- mattb @ 2006-09-15T02:23Z

[edit] Condradictory "Disadvantages" Section

This section repeatedly makes a negative statement about overclocking, then contradicts it with a statement about how it really isn't a problem. It really ought to be changed.

[edit] downbinning / "under rating" / "manufacturer underclocking"

I added a bunch of "verify source" tags to claims that manufacturers downbin components based on demand / yields, since I've never heard this from a reputable source. It doesn't make economic sense to me, so I'm going to remove these claims from the article if nobody can provide a trustworthy source or at least convince me that it is an economical behavior for a company. --CTho 01:15, 20 September 2006 (UTC)

I have heard of it happening, by word of mouth, though I am the farthest thing from an econ major, and couldn't imagine how it makes good business sense. If anyone feels like posting a situation in which it would be beneficial to a company to do, feel free. - McCart42 02:38, 20 September 2006 (UTC)

I used to work at Intel developing software for the Assembly and Test factory floor. I know for a fact that downbinning occured as it was a part of the business process implemented in the software. It make business sense in the following way: say Dell orders 1000 100MHz CPUs. From experience you know you get a mix of 10% 100MHz, 60% 150MHz, and 30% 200MHz. So if you don't do any downbinning in order to fill Dells order you need to start enough wafers to make 1000 100MHz CPUs, but that would also produce 6000 150MHz CPUs and 3000 200MHz CPUs. If you don't have buyers for those extra 9000 CPUs, that's time and space you are wasting. Instead, they start just enough wafers to fulfill all of the orders even if it means downbinning some CPUs. You save money by shipping and storing less product, and getting orders done on time.

Doing so makes perfect economic sense. See Price discrimination. It would be stupid for manufacturers not to downbin. Grouse 08:15, 13 October 2006 (UTC)

Why not drop the price on the 150MHz parts instead, thus making your product much more appealing than the competitor's, boosting your volume of shipments in the bins where you get the most parts? Just because that's not a concern for a monopoly? AMD is currently not a monopoly, and Intel seems unlikely to want to act like one right now given AMD's lawsuit, so would it make sense nowadays? Anyway, do you have anything we could use as a source? If it's true and someone has a source, it should certainly appear once in the article (though it doesn't need to appear 3 times as it did before) --CTho 18:48, 14 October 2006 (UTC)

The price on the more expensive chips does eventually drop, largely because of process refinement. Maintaining the various speed bins allows the manufacturer some time to get all the bugs worked out of their process. The first batch of the highest speed bin will be the most expensive because yield at that point is very low. As the process is refined a bit, yield on those faster chips increases and therefore the price can drop. Eventually the process will be at the point where it gives roughly the same yield for several speed ranges. So if the aforementioned big order comes in, it makes much more sense to downbin some of the ICs that can run faster to meet the demand rather than doing several more process runs to pick out several more lower-performing devices and thereby have an excess stock of the higher performing ones. So again, it's basically about what the demand is, because at a certain point the yield for various performance levels becomes pretty similar, and a large-scale CMOS facility builds by the wafer, not by the die. -- mattb @ 2006-10-17T19:38Z

[edit] "exploding" cpu video

I believe this video[1] is deceptive. When Tom's Hardware performed the same experiment, they merely observed extreme temperatures and smoking components. We are not actually shown the cables from the exploding system going to the monitor. PCBs are quite sturdy, so it would require significant force to blow a hole in the motherboard; with that much force I would expect more damage to the socket, and the motherboard to jump visibly. If you look carefully, you can see that the hole actually goes through the table and you can see through to the ground - any blast that blew a hole through the table would have resulted in a lot more than the CPU flying out (e.g. the whole motherboard would probably be launched). I suspect that this is staged: the motherboard and table already had holes cut in them, the system we are shown is not the one displaying images on the monitor, and a small explosive such as a firecracker was used to launch the CPU. --CTho 06:04, 11 November 2006 (UTC)

you are so thorough, I suppose rvv would explain enough--200.181.168.199 20:16, 20 November 2006 (UTC)

[edit] OVerclok

I heard sum1 used a refirgerator to cool down the CPU, I woner if u can clock internet!! Realg187 21:40, 13 December 2006 (UTC)

Its not possible to overclock the internet, only computer chips. Sorry. Dekard

Too bad, how can I make my PC faster, is it in the BIOS?? Realg187 16:29, 14 December 2006 (UTC)

I'm not even sure where to start in correcting misconceptions here... My only advice is not to worry too much about overclocking. If you really want to learn something, pick up a book on microprocessor architecture. -- mattb @ 2006-12-14T17:51Z

You could start at overclockingwiki.org... read the articles on overclocking basics there and it will get you started. After you have done some basic reading on overclocking there you can hit the forums and get some personal help. Dekard 16:12, 15 December 2006 (UTC)

I like forums!!, is it different for each PC and wroth it?? Ill have to ask on the forum!

what have I done.. Dekard

[edit] OverclockingWiki.org

This site is getting a pretty considerable amount of overclocking information (143+ articles) and has an active user base. People looking for overclocking information here would also be well served with the information on www.overclockingwiki.org, which includes a user to user support system via their forums. Dekard 16:15, 14 December 2006 (UTC)

Why was the link to overclockingwiki removed without discussion? -- Mikeblas 18:09, 14 January 2007 (UTC)

Because, if I believe it's not within Wikipedia policy - and is more a directory link - then I can remove it. Just as anybody else can. MansonP 09:50, 15 January 2007 (UTC)

If you can come up with better reasoning than "because you can", then let's talk it over. Meanwhile, I've added the link to overclockingwiki.org to the external links section. Per WP:EL, the site provides a level of detail that's inappropriate for Wikipedia artcles, and is meaningful and relevant content to this article. -- Mikeblas 19:15, 16 January 2007 (UTC)

220+ pages of overclocking information is enough for me to call it details. A directory would merely link to other sites, all the article links I saw were internal with the exception of links to manufactures homepage. And the focus is pretty strict, unlike WP's overall scope. I think its a great link. Dekard 17:02, 18 January 2007 (UTC)

Mikeblas, sure there are better reasons. For one, the WP:EL guidelines clearly state that links to be avoided are "Links to open wikis, except those with a substantial history of stability and a substantial number of editors." This domain was registered at the end of last year and hardly qualifies under substantial history. Heck, it doesn't even figure in archive.org at all. We won't even get to the substantial number of editors qualification.

BTW, Mikeblas and Dekard, the two defenders of keeping this external link, are you the same Mikeblas and Dekard at who run that place? If so, is there somewhere here you've disclosed that you are connected to the site? MansonP 10:09, 19 January 2007 (UTC)

[edit] Recent reversion

The editor who added the content that I recently reverted asked me to detail my reasoning for doing so. I'll start with the incorrect explanations. First, higher transistor density does little to make "electrons travel faster". Carrier mobility (the net drift velocity response per unit electric field) is largely related to temperature and band gap energy (through its relationship with carrier effective mass) and to various scattering mechanisms (lattice/phonon, ionized impurity, etc). Making devices smaller doesn't make electrons flow faster, though it does reduce the amount of charges that must be moved around to change transistor modes (which is a large reason smaller transistors are faster). To make electrons (and holes) truly flow faster, you would need to increase the electric field. However, generally VLSI houses try to practice constant electric field scaling (that is, scale voltages with the dimensions to keep electric field roughly constant) in order to avoid hitting carrier saturation velocity too soon as well as a host of breakdown mechanisms (avalanche, tunneling, etc). In reality, constant electric field scaling isn't precisely followed due to some real-life difficulties, but that's a much more advanced discussion.

Now for the ill-drawn conclusions. I would not agree with the assessment that smaller transistors automatically imply better ability to overclock. It is a factor, but it is not nearly the only one nor the simplest one. How much one can overclock a microprocessor is really defined by how much clock speed tolerance the manufacturer sets for a given batch of dies. In other words, it's usually the economics more than the technology that determines how well a chip will overclock. Granted, some processes will facilitate higher clock speeds better than others, often as a virtue of thermal improvements (which is what I would guess is playing a large role in the Pentium D example case), but to take this and conclude that smaller transistors will make a microprocessor more overclockable is an oversimplification. -- mattb @ 2007-02-14T15:11Z

Retrieved from "http://en.wikipedia.org../../../o/v/e/Talk%7EOverclocking_9126.html"

Categories: Translation requests | Translation request