A question for scientific bods

In the 1870s, Joseph William Swan solved the problem of the carbon filaments of electric light bulbs being oxidised by the intense heat of the current passing few them by placing the filament in a vacuum tube. Similarly, across the Atlantic, Henry Woodward and Matthew Evans made advances by displacing the air in the glass tube with nitrogen.

What strikes me about this story is that, when faced by similar problems of heat in the world of overclocking – rather than neatly sidestepping the issue, we instead opt for costly cooling devices. Why is this?

Unlike electric light bulbs, CPU designing and equipments cost thousands (if not millions) times more. It would be cheaper to get better cooling than to design a new CPU.

Going dual core is already some kind of sidestepping, is it not?

Ah, but that's an issue of practicality. Has this simply not been solved because of a lack of financial incentive or is it a physical impossibility?

The oxygen was removed to make the fillament burn out slower so the bulb would last, not to cool it.

^^ beaten too it

Have a cookie. They didn't bother to cool their filaments, they found a way around the problem. Why can't this be done for our silicon chips?

AFAIK CPUs don't break because they start burning... they just can't work when they get too hot, so the the light bulb scenario cannot be placed alongside the CPU temp scenario.

Quote:

---

Originally Posted by n3wt

Ah, but that's an issue of practicality. Has this simply not been solved because of a lack of financial incentive or is it a physical impossibility?

---

Physics is also a problem for processors to go that hot. More transistors = more heat. Transistors run faster = produce more heat. Lowering voltage and process are not enough to counteract that two increases.

Intel or AMD could make some more efficient processors (may be IA64 kind), but what is the point making a processor that no OS or program support.

It is like if you make a SATA10-30000 harddrive with 100000TB capacity, but no motherboard supports SATA10, nor Windows XP supports 100000TB capacity. No one is going to use it, no one going to buy it. Since no one will buy it your won't make it, or even design it.

Its not a cheapo bulb you are talking about, and the "improved" bulb can be used in the same socket as the old one, use the same AC power, and even emitting the same white light.

I doubt anyone want to scrape his system, buy all the new softwares and learn how to use them if they want a faster and cooler system.

We all know oil is going to run out, but what percentage of people actually switching to renewable energy? what percentage of people bother to buy energy-saving lamps.

Quote:

---

Originally Posted by Kez

AFAIK CPUs don't break because they start burning... they just can't work when they get too hot, so the the light bulb scenario cannot be placed alongside the CPU temp scenario.

---

I realise the problems are more related to resistance and the like. I merely offer the light bulb as a parallel. What they did was make it so that, while light bulbs still produced heat, this heat was no longer a problem. Wheras we waste energy trying to dispel this heat. How could we change things so heat would no longer be a problem for chips?

arthurleung - your comment, which perhaps runs parallel to the question of how could we solve this - is that it probably could be solved, only there is no financial incentive to do so as it would entail complete redesign of the computer system? Bear in mind that, if such a new technology could be devised, it would stretch beyond CPUs - memory, northbridge, GPU, etc etc could all be replaced.

Maybe edit physics.c so that silicon doesn't become more resistive as temperature increase? :P

One thing that should NOT be used to avoid the problem is power saving options. I'm seeing Intel use SpeedStep on desktop boards and while I welcome it to a degree, it worries me. It's supposed to cut power consumption, not heat output. What happens when you run the CPU at 100% for an hour and your cooling solution can't keep up?

Quote:

---

Originally Posted by Kez

Maybe edit physics.c so that silicon doesn't become more resistive as temperature increase? :P

---

Or maybe use something other than silicon...?

One word: Superconductivity

The first person to create a superconductor at room temperature will be a multi-trillionaire.

Quote:

---

Originally Posted by n3wt

I realise the problems are more related to resistance and the like. I merely offer the light bulb as a parallel. What they did was make it so that, while light bulbs still produced heat, this heat was no longer a problem. Wheras we waste energy trying to dispel this heat. How could we change things so heat would no longer be a problem for chips?

---

You could get 40 500Mhz xscale processors and they can process information faster than a P4 4Ghz, but they can't play game (even if they do, games will be slow like h3ll since games are not multi-threaded)

You could use 100 LEDs to light your room, they produce more light and of coz more energy efficience, but how many people will bother to use LEDs to light rooms. (Technologyly possible, cheap, and most importantly it works), just because you already have a lamp socket and you wouldn't bother to modify your socket so it can take the LED light.

Quote:

---

Originally Posted by Kez

Maybe edit physics.c so that silicon doesn't become more resistive as temperature increase? :P

One thing that should NOT be used to avoid the problem is power saving options. I'm seeing Intel use SpeedStep on desktop boards and while I welcome it to a degree, it worries me. It's supposed to cut power consumption, not heat output. What happens when you run the CPU at 100% for an hour and your cooling solution can't keep up?

---

I wonder what will businesses say if Intel sell its Xeon processors with a label like this: "Intel Xeon processors are not designed to run at full load continously, continous usage may cause system failure, blah blah"

Quote:

---

Originally Posted by arthurleung

You could get 40 500Mhz xscale processors and they can process information faster than a P4 4Ghz, but they can't play game (even if they do, games will be slow like h3ll since games are not multi-threaded)

You could use 100 LEDs to light your room, they produce more light and of coz more energy efficience, but how many people will bother to use LEDs to light rooms. (Technologyly possible, cheap, and most importantly it works), just because you already have a lamp socket and you wouldn't bother to modify your socket so it can take the LED light.

---

At a local major DIY retailer near you
http://www.diy.com/diy/jsp/bq/produc...PRODID=6680003

You can also get LED bulb arrays to fit in regular low voltage spotlight fittings

Quote:

---

Originally Posted by arthurleung

I wonder what will businesses say if Intel sell its Xeon processors with a label like this: "Intel Xeon processors are not designed to run at full load continously, continous usage may cause system failure, blah blah"

---

Well servers are a little different. They don't need to be fairly quiet and sit in a home office/living room. Fair point though.

Quote:

---

Originally Posted by SiC

At a local major DIY retailer near you
http://www.diy.com/diy/jsp/bq/produc...PRODID=6680003

You can also get LED bulb arrays to fit in regular low voltage spotlight fittings

---

I know those exists, but I personally never seen anyone using those. I should go and buy some myself, as an example :P

EDIT: I looked into that thing a little bit, its not as strong as those 100W filament lamps, I meant those which is close to that level of brightness :P

LEDs aren't actually quite as efficient as gas discharge tubes.

Anyway, the whole point of a filament bulb is that the filament gets hot, it doesn't produce light otherwise.

Interesting thread, but I am confused how light bulbs can be compared to cpus, as Rave said: light bulbs need heat to work.

I cant see any way in which a cpu would need heat to work. There is a new technology that uses a different material other than silicone that produces less heat and can withstand higher temperatures. But still even these cpus really don't need or want heat.

Heat is a by product of cpu's working, due to resistivity. You can reduce this heat but only by applying less voltage or using different architecture or different material for chips.

Light bulbs do indeed need heat to work, that is why their resistance is such. The reason why light bulbs have a vacuum is simply to stop them oxidising to form oxides of the filiment wire and therefore require increasing current to make them glow, which leads to overload and failure. Oxides tend to be more brittle.

i.e. Vacuums or low oxygen atmospheres cannot be associated with removal of heat.

Radiation will travel through a vacuum.

Go out on a hot sunny day and bask in the heat. Where does it come from? The sun; 95 million (vacuous) miles away.

If the world/universe were at 0°K then matter would cease to exist and nothing would work (as we know it).

I think the point in this thread is "How could we change things so heat would no longer be a problem for chips?" Not how could we reduce the heat output from chips/cool them better.

Although this would mean either using a different material than silicon to make the chips, or using a completly differnet process, maybe not using electricity at all... (no electrical resistance = no heat)

Quote:

---

Originally Posted by Mblaster

I think the point in this thread is "How could we change things so heat would no longer be a problem for chips?" Not how could we reduce the heat output from chips/cool them better.

Although this would mean either using a different material than silicon to make the chips, or using a completly differnet process, maybe not using electricity at all... (no electrical resistance = no heat)

---

Down to the basics of computing, you need 1s and 0s which is expressed by voltage difference. Even your brain use electric pulses to send information. I very doubt a processor can work without electricity.

May be optical electronics (or whatever it is called)?

Well, come to think of it they are working on chips that can work at much higher temperatures- the plan is to use diamond rather than silicon. There's a company out there that already has the capability to grow flawless diamonds from carbon plasma, but ATM they're not big enough to use for chips. When they refine the technology enough to produce 200mm diamond wafers, we could be in for some proper hot chips (and some very cheap bling;)). Thing is though, do you want a computer that uses, say, 500W to power the processor alone? That's about pound a day in electricity to run the thing if it's on 24 hours a day, not to mention that it'll heat your room up to a fairly unpleasant degree in the summer....

From Tomshardware

"As we mentioned above, perhaps the most interesting aspect of the comparison between the Pentium 4 with the Pentium M is related to thermal design power (TDP). The fastest P4 processors are specified for up to 115 or 84 Watts in Socket 775, depending on the classification and validation of each part after production. For Socket 478, the maximum TDP is 103 W for the 3.4 GHz part. In contrast, the Pentium M numbers are much, much lower. Looking at Dothan, the maximum TDP is only 21 watts, while Banias consumes a maximum of 22 watts."

So architecture is certainly one way.

Quote:

---

Originally Posted by arthurleung

Down to the basics of computing, you need 1s and 0s which is expressed by voltage difference. Even your brain use electric pulses to send information. I very doubt a processor can work without electricity.

May be optical electronics (or whatever it is called)?

---

But you can tell me that high/low voltage is the only way to show a 1/0?It might be possible for light to be used (as you mentioned), I don't know, but I'm sure there could be other ways of doing it, even if we haven't discovered it yet.

Quote:

---

Originally Posted by Mblaster

But you can tell me that high/low voltage is the only way to show a 1/0?It might be possible for light to be used (as you mentioned), I don't know, but I'm sure there could be other ways of doing it, even if we haven't discovered it yet.

---

For example chemicals used to symbollise 1 or 0. (A different chemical for each)

I think there is research going on about how cells can be used as processors, and they are massivly qucik as they can do so many calculations per second.

^^ yeah that's right, I'm sure I read somthing about this in new scientist about 8 years ago when I was at college. 'Computers made from liquid' or somthing the article was called, was a very interesting at the time, have never heard of it again since....

personally I don't think binary encoding is the only way to go to process information, look at DNA's base 6 code for example, or perhaps look into neural structures, the human brain is the most complex information processing system known to us, granted the methods involved are not fully understood.. theres always more than one way to break and egg, as they say.

Quote:

---

Originally Posted by Timmy!!!

For example chemicals used to symbollise 1 or 0. (A different chemical for each)

I think there is research going on about how cells can be used as processors, and they are massivly qucik as they can do so many calculations per second.

---

Even chemical need energy to change from one state to another. Not to mention kinetic energy (hdd/fan vibration) may be used as input of energy therefore causing error.

Problem of using cells (I assume you are talking about biological cells) are they can die, and they will split. Not to mention they can move. How to dynamically address them is a problem. They will be damn large and heavy too. Imagine a 10L bottle filled with cells have processing power faster than a FX55. You need to sort out waste, food, temperature, sunlight and bacteria problem, etc.

Check this out - possibility of a quantum computer:

http://www.harvard-magazine.com/on-line/110157.html

further quantum computer reading:

http://computer.howstuffworks.com/quantum-computer.htm

I would have thought that a biochemical solution would be far too unstable as mentioned above.

The most promising technology I can see for the future of CPU's (i.e. post silicon) is the use of "carbon nanotubes" for transistors. Transistors can be made smaller than current or future silicon based transistors and also require drastically less voltage to operate.

I would expect to see silicon/carbon nanotube hybrids first. The technology will take a decade or two to come to fruition (he speculates!)

In the meantime expect to see "strained silicon" combined with "silicon on insulator" (SOI) technology called "Strained Silicon Directly on Insulator" (SSDOI) to either reduce the heat and/or more likely increase the number of transistors.

The most efficient number of states to have is actaully "e" (2.7 ish), obviously you can have 0.7 of a state, so in theory 3 states would be the best, but 2 is much easier to implement so is used.
Semiconductors is a massive industry.
Optoelectronics is cool...lol
Damn...was going to post a suitable link, but its intranet notes...
well. somewhere on www.ecs.soton.ac.uk I think there is interesting stuff...lol

Quote:

---

Originally Posted by Twigman

The most efficient number of states to have is actaully "e" (2.7 ish), obviously you can have 0.7 of a state, so in theory 3 states would be the best, but 2 is much easier to implement so is used.
Semiconductors is a massive industry.
Optoelectronics is cool...lol
Damn...was going to post a suitable link, but its intranet notes...
well. somewhere on www.ecs.soton.ac.uk I think there is interesting stuff...lol

---

Problem is you need extra logic to decide if its a 1 or a 2 or a 3.

Just like your LCD is 24 bit colour, giving you a screen of colour can you decide if its deep-deep blue, deep blue or black? and being correct every single time?

It is very unlikely a 2 states system will have such error (its either black or white), but more states = more likely to get the state wrong.

Which is why 2 is used :)
0v or 5V (or whatever your using) :)
The more states you have the more it becomes an analgue signal and more suseptable to noise it becomes :)

Quote:

---

Originally Posted by Twigman

Which is why 2 is used :)
0v or 5V (or whatever your using) :)
The more states you have the more it becomes an analgue signal and more suseptable to noise it becomes :)

---

If there is a two way transistor, you can actually get 3 states -5V, 0V and 5V without noise :P

then you get back to your other point then :p
If only I put this much effort into my work on transistors...lol

Quote:

---

Originally Posted by Kez

One thing that should NOT be used to avoid the problem is power saving options. I'm seeing Intel use SpeedStep on desktop boards and while I welcome it to a degree, it worries me. It's supposed to cut power consumption, not heat output. What happens when you run the CPU at 100% for an hour and your cooling solution can't keep up?

---

Then you take it back to the place you got it from and get the muppets to fit a better HSF.
I cant see Intel ever releasing a chip that cant be run at full speed all the time due to no cooling solutions being able to keep up with it.