Hi,

I was just wondering. If we build a CPU twice the actual size, will it produce less heat? Or what does produce the heat and how to reduce it?

I still remember my 060 which didn't need a cooling system and all the problems I have because of the G4 "overheating". (I'm now fixing this)

_________________
Philippe 'Elwood' Ferrucci
Sam460 1.10 Ghz
AmigaOS 4 betatester
Amiga Translator Organisation

@elwood

A smaller die size(transistor) determines the amount of power required therefore a smaller chip will generally be cooler.

@EgoX

but then, the transistors will also be tightly confined, and will invariably be switching faster.

basically: size of die(which is the actual lump of silicon, not the transistors themselves!), size of transistors, transistor density, core clock rate / transistor switching speed, current and voltage levels (Power = Current x Voltage, Power roughly equates to heat) are all factors.

Last edited by Agafaster on 13-Jun-2005 at 10:23 AM.

_________________

@elwood

If your chip draws 20w, it means it'll generate 20w of heat. The G4/G3 as you know, have a very small junction up towards the heatsink. It's generally 1x1cm or smaller. 20w on an area that small means alot of heat concentrated on that spot. If that junction area were to be expanded to the entire upper side of the processor, it would be alot cooler. It would still generate 20w of heat, but spread out across 3x3cm (or something).

So yes, making it larger, or atleast the junction area up towards the heatsink would be a very good thing, it would be alot easier to cool off, for one.

_________________
It is not the critic who counts: not the man who points out how the strong man stumbles or where the doer of deeds could have done better. The credit belongs to the man who is actually in the arena." Theodore Roosevelt, speech in Paris, 1910

@GuruMeditation

But then it would require more power to perform the same, which negates the reason for making it larger.

@GuruMeditation

Okiee.
Maybe the Cell doesn't need a cooling system. (dreaming)

_________________
Philippe 'Elwood' Ferrucci
Sam460 1.10 Ghz
AmigaOS 4 betatester
Amiga Translator Organisation

@elwood

Quote:

elwood wrote: @GuruMeditation Okiee. Maybe the Cell doesn't need a cooling system. (dreaming)

CELL on the left of the pic...



Heatsyncs for the CELL

_________________
Dark Lord Design Wicked Solutions For Damned Problems

@treblesix

Just as a note. That's a test mobo, with overdimentioned heathsinks designed to take the heath that's generated when you overclock the chips to "see what they can do".

So while Cell chips (running about 3Ghz) will propably need air cooling, it will propably not be as bad as that shown in the pictures.

@Eric_S

Quote:

Eric_S wrote: @treblesix Just as a note. That's a test mobo, with overdimentioned heathsinks designed to take the heath that's generated when you overclock the chips to "see what they can do". So while Cell chips (running about 3Ghz) will propably need air cooling, it will propably not be as bad as that shown in the pictures.

and just for note, heat, Heaths are anthropogenic habitats found primarily in northern and western Europe, where they have been created by thousands of years of human clearance of natural forest vegetation by grazing and burning on mainly infertile acidic soils. They subdivide into two broad categories depending on climate, with true heathland developing in warm, dry conditions, and moorland developing in cooler, wetter conditions.


Wouldn't want one of them in my computer

Last edited by treblesix on 13-Jun-2005 at 12:16 PM.

_________________
Dark Lord Design Wicked Solutions For Damned Problems

@EgoX

You can make it bigger without adding more transistors and etc to it. You don't have to add to the complexity, just make the junction larger, and with a material that transfers heat well. I believe the intel processors has a large "shield" covering their cores, that acts as both protection and heat transferrer over to the heatsink.

That requires no power.

_________________
It is not the critic who counts: not the man who points out how the strong man stumbles or where the doer of deeds could have done better. The credit belongs to the man who is actually in the arena." Theodore Roosevelt, speech in Paris, 1910

@treblesix



or...

a 1970's Conservative Prime Minister...

nope, gotta agree, wouldnt work well in an A1 !!

Last edited by Agafaster on 13-Jun-2005 at 02:12 PM.

_________________

@elwood

Size and make up of chip has a relationship to the removal of heat on the CPU itself. If you have a 20Watt processor at 3 square inches vs 20 Watt processor at 1 square inch. It should, in theory, be easier to cool the bigger processor.

@GuruMeditation

Sure improving the material would solve the problem but this would make the product more expensive which would make it difficult to market.

bigger junctions alone wouldn't cut it because that would cause more resistance meaning you would require more power.


Interesting article for anyone who's interested

@treblesix

So, are you trying to say it'll only need a 1/4 sized massive heatsink as compared to a PPC?

_________________
Tony T.

People who generalize are always wrong.


1989 - 500 / 1991 - 3000 / 1997 - Genesis Flyer 1200T / 2003 - A1XE

@treblesix
Hmm, who's Cell blade? The connector doesn't look like a normal IBM Blade connector?

Whoever made the comment RE: heat sink size is certainly right...some of the first prototype IBM blades of different models took up 2-3 blade slots, final fit into one. The above heat sinks could likely be shrunk by going to all copper as well...

_________________
Are we not done with the same silly arguments and flames yet??!

@treblesix

Can you imagine the noise the PS3 is gonna make when it's on.

_________________
- Ronald

All beer tastes bad.

@elwood

Heat in todays chips mostly comes from the electric current required to charge or discharge a capacitor. CMOS transistors each act like a tiny capacitor as you have to charge up the gate or discharge it to change the on/off state.

The larger the transistor, the more charge/discharge needs done, the hotter it gets.

The faster the on/off requency, the more often you chare or discharge it, the hotter it gets.

The more transistors you have, the more capacitors you have to charge and discharge, the hotter it gets.

The higher the voltage, the more you have to charge or discharge each transistor, the hotter it gets.


So, the ways to reduce heat output are these:

Reduce the number of transistors

Reduce the size of each transistor

Reduce the voltage

Reduce the speed of the circuit


You can't simply take N transistors of Y size and spread them out over a larger surface area to reduce heat, as the number, size, speed, and voltage are all the same. This may actually increase power dissipation, as the wiring is now larger, and capacitance between wires will thus be larger than if it was packed smaller to use shorter wiring. Longer wires also have higher resistance which increases heat.

As not many people would want a slower computer, the key is smaller, smaller, smaller to reduce heat generated. This is one part of why chip technology keeps shrinking. You wouldn't be able to operate a Pentium4 at today's clock rates with 0.5 microns running at 5volts. That would literally melt itself rather quickly. Smaller size and lower voltages are required to make faster possible by reducing heat output to a managable level those big heatsinks are able to cope with.

Smaller is also less expensive, and not many would be willing to make something cost more like this. Smaller is cheaper because of how many good chips come off a single wafer. If a wafer holding 600 chips has 70% yield, a wafer of the same overall size, with smaller technology holding 1000 chips with the same 70% yield, overall production has gone up quite a bit. As it generally costs approximately the same to manufacture a wafer of one technology to the next in terms of resists, etching, exposing masks, and growing the next layer, you can recover the wafer cost with a lower end price from the 1000 chip wafer than from the 600 chip wafer.

_________________
All glory to the Hypnotoad!

@billt

Quote:

Reduce the size of each transistor

Unless you're Intel where the reduced 90nm size increased the heat generated.

@BrianK

Quote:

>Reduce the size of each transistor Unless you're Intel where the reduced 90nm size increased the heat generated.

Surely they also increased the number of transistors enough or upped the clock speed enough (or all ov the above) to cause an overall increase in full-chip power?

Remember, decreasing one of these items does not guarantee overall chip power goes down, only that this one aspect goes down. If you cut your transistor size in half but triple the number, the overall chip power will still be higher. It's a big balancing act with the second motivation to reduce transistor size being the desire to put more of them in there to add new features or more internal cache memory or a longer pipeline with smaller steps per stage to be able to crank the clock up even higher. If you keep things in balance, then overall power can go down. per transistor power is easy to lower, but if other aspects are given as much leeway as possible they can easily outpace the savings you've made.

I understand the Pentium-M ULV is rather impressive on performance to heat dissipation compared to the Pentium-4, and I'm guessing that is the chip that our buddy Mr. Jobs is so excited about in his performance/watt speach.

Last edited by billt on 14-Jun-2005 at 04:37 AM.

_________________
All glory to the Hypnotoad!

@billt

Quote:

Surely they also increased the number of transistors enough or upped the clock speed enough (or all ov the above) to cause an overall increase in full-chip power?

Yes the architecture of the Prescott 90nm PIV is different then the Northwood 130nm PIV. However, Intel made a big mistake. They increased transitor count, increased heat, made it smaller, but at the same Mhz decreased performance. Intel cancelled the 4Ghz part and is doing Dual core and focusing on the Pentium-M, which is a different architecture. Intel simply knows this was a mistake and appears to be killing off the Prescott line.

Quote:

I understand the Pentium-M ULV is rather impressive on performance to heat dissipation compared to the Pentium-4, and I'm guessing that is the chip that our buddy Mr. Jobs is so excited about in his performance/watt speach.

Perhaps. Remember Jobs said the 'roadmap' with Intel was where they wanted to be. Additionally consumer machines would come fall of 06. By this time Dual Core desktops should pretty much be the standard. There's probably a future chip, perhaps based off of the Pentium-M, that Jobs is looking at.