Talk:How To Assemble A Desktop PC/Overclocking

Spelling
The text under "Undervolting" includes the text "ususually". Maybe ... improvable spelling? Should it be "usually" or "unusually"? Feel free to delete this "discussion" once assessing (and, probably, correcting!)

Larkloon (discuss • contribs) 02:06, 7 October 2019 (UTC)

Just as a question, does anyone think that this is completely relevant to the actual building process? I have nothing against overclocking, but I wonder whether it's suited to the target audience... --Noogz 07:08, 24 Mar 2005 (UTC)


 * Yes, because it's something you do with your new computer. And a lot of over clockers are doing so with computers they built. Reub2000 07:36, 24 Mar 2005 (UTC)
 * It's not a very good thing to do with your new computer. The speed increase could be had by waiting a few months instead. That's all you get, just a few months worth of advantage. Overclocking an obsolete computer makes more sense, because it's already out of warranty and you'd be buying a new system anyway if the old one won't go faster. AlbertCahalan 01:43, 4 May 2005 (UTC)


 * Overclocking has little to do with building a new computer. Cooling is related, but that should not go here because all computers need some form of cooling. AlbertCahalan 01:43, 4 May 2005 (UTC)


 * I take it that you don't spend much time around computer hardware forums, where they like to overclock anything they can get their hands on. It's relevent because I belive that most overclocking takes place on self built computers. OEM computers usually can't be overclocked. And many computer enthusests (people who build comps) also overclock. Besides, OCing a Radeon X850XT PE (new hardware) is the only way to get the fastest video card available. Reub2000 06:35, 4 May 2005 (UTC)

I agree with AlbertCahalan. Building a computer always involves cooling, but overclocking is optional. If there is no objection, I'm going to move stuff about making computers quieter (is there a name for that ?), making a CPU go faster (overclocking), RAID, putting pretty lights on the case, etc. into an appendix, to emphasize that they are optional. --DavidCary 17:14, 21 May 2005 (UTC)

I moved the "quiet PC" stuff that was on "Optimizing and overclocking" to How To Build A Computer/Quiet PC --DavidCary 02:53, 14 September 2006 (UTC)

No heat increase with speed increase?
A voltage increase will definitely cause more heat, but with current CMOS technology speed will also affect heat output. Think about it for a moment: during HLT (i.e. in idle CPU cycles) the CPU is effectively stopped, but running at a voltage. But, the idler the CPU, the cooler it is. This is because CMOS technology only consumes significant amounts of power during switching. With CPUs some is added to this due to leakage and tunneling, but an important part of the power consumed is still due to switching. Thus, AFAIK a speed increase can and will increase heat output, although less noticeably than a voltage increase. Marcan 02:12, 11 August 2005 (UTC)


 * Interesting. Some people (including myself, who knows a little about AC circuits) thought it would either reduce the heat output or keep it the same). In theory, increasing the frequency prevents the circuit Vmax reaching its peak and sink values, which would seem to reduce power consumption and heat output. However, this measure does not include the current, and switching would increase it assuming that the idle state is 'off'. Is this right?

Websites are typically quite ambigious on this subject. In particular, it is hardly ever made clear if overclocking or the temperature itself causes malfunctions. I think this one should make it clear.--ChrisJMoor 02:09, 7 December 2005 (UTC)


 * CPUs are CMOS circuits, usually, iirc. I was under the impression that higher frequency = higher current in this case, and since i*v=p, more power. --Anaraug 09:59, 3 February 2006 (UTC)

I agree with ChrisJMoor -- this book should make things clear. The CMOS article mentions that the biggest source of power consumption in digital CMOS devices is
 * "... the characteristic switching power dissipated by a CMOS device: $$ P = C V^2 f $$."

Can we explain this in a way that is not too complicated? But also not oversimplified?

My understanding is that *both* overclocking *and* too-high temperature cause different sorts of malfunctions. Here are 2 kinds (are there others I am forgetting?)
 * overclocking error: the combinatorial logic from one latch to the next latch takes a certain amount of time to settle. If the destination latch is clocked too soon after the source latch was clocked, there is not enough time for the logic to settle, and the "wrong" values get fed into the destination latch. That's the whole point of long pipelines -- they split up the logic into many thin layers, and insert latches between each layer, improving the worst-case settling-time (through the thickest layer). Clocking a device faster than this worst-case settling time causes occasional "wrong" values to be latched.
 * temperature damage: In a CMOS device, electrical energy is converted to thermal energy almost entirely in the transistors. If the transistors get too hot, the carefully placed patterns of diffusion can blur and melt, permanently damaging the chip. If they blur into a short-circuit, then the chip rapidly self-destructs.

My understanding is that older microcoded processors showed obvious settling-time errors at clock rates well below the temperature-damage clock rate, even without heat sinks. But now that most chip designers have figured out better ways to avoid such settling-time errors (pipelines, transistor sizing, buffering long lines, low-skew clock-tree networks, etc.), well-designed chips simply don't have settling-time errors at clock rates from DC all the way up to the clock rate that causes thermal self-destruction.

--DavidCary 02:53, 14 September 2006 (UTC)

Things that can't be overclocked
"(But you wouldn't be reading this guide if you're using an OEM computer, would you?)" I think this is a completely unnecessary thing to say, and looks like a bad attempt at humor. It should be removed.