Overclockers Top 6GHz With A 3.6GHz-Rated P4

sH4RD writes "The 6GHz barrier has been broken by two guys, a little LN2 (liquid nitrogen for those not as chemistry inclined), and an Intel Pentium 4 (Prescott) 3.60GHz. Check out some icing and some proof of speed. Better yet take a look at how fast it calculates pi. Also be sure to check out the original announcement."

Don't forget the dual clocked ALU

[prestwich]

So P4's double clock their ALUs - that means that ALU is shifting at > 12GHz.

Welcome to measuring your operations in picoseconds.

Re:Only 5.4GHz

[mobby_6kl]

As the AC already said, it's not stable at 6Ghz, from the article:

Not bench stable - just a screenshot record :)

The CPU powersupply seems to require quite a bit of modding in order to bench past 5.4GHz.

LN2 ? Try some LHe!

[cyberfunk2]

What's this LN2 stuff... everyone knows REAL overclockers use liquid helium surrounded by a vaccum flask with another LN2 flask outside of it. (P.S. That's the setup they use to cool NMR machines for chemistry that have superconducting magnets in them)

Heh.

[SatanicPuppy]

It would take a HUGE fan to keep it from overheating and causing a board shutdown or a processor meltdown.

I've got a 3.02 ghz, mildly overclocked, and the fan shutting down and the board automatically shutting down due to high heat are nearly simultaneous.

Re:Erm...

[_Pablo]

Clock for clock, the final P3 (the Pentium III S) would smoke the original P4 (Willamete) no questions asked. So it's quite possible dual P3/500s could have beaten a 1.5Ghz P4 on many benchmarks.

Re:Erm...

[EinarH]

There is no *way* that a dual P3/500 system will match an 1 GHz P4, not to speak of your claims of beating an 1.5 GHz P4.

Sure it is.
The first Pentium 4 CPU was slow compared with a P3 1 GHz. One would belive that a 1.5GHz CPU would beat the last generations 1 GHz CPU, but in many tasks the P3 was faster.
-The P3 pipeline had 12 stages the P4 had 20.
-The P3 Katmai had 512k L2 cache, the P4 had only 256k. I remember some MySQL benchmarks showing a single P3 500 MHz Katmai beating a P4 1400 MHz in some tasks.

So even with all the IDE stuff enabled a Dual P3 could be faster than a P4 in Gentooing.

Re:Hang one outside the ISS !!

[Holi]

Actually no, It would overheat quite quickly as a vaccuum is a very good insulator (heh some would say almost perfect).

Re:Erm...

[berkut7]

The reason why they run with so little memory is the same why they are using a cheap video card: there is a chance they might kill it. The other more important reason is that they can reach higher FSB clocks with less memory sticks. I fthey had two or more memory sticks they would be able to reach same FSB speeds, an in turn, same CPU clock speeds.

+7.3, informationativinal

[Anonymous Coward]

Looks like one of these doodads to me:

http://www.xbitlabs.com/articles/memory/display/oc z-booster.html [xbitlabs.com]

Re:liquid nitrogen sucks

[Big Bob the Finder]

Liquid propane boils at -42.1 degrees C. It goes from the solid to the liquid at -187.7 degrees C, which is not important for this, but read on below. It also has an explosive range from 2.8% to 9.5% in air- a little lower than natural gas.

Liquid nitrogen boils at -195.8 degrees C, which is cold enough to freeze propane into a solid (there's a fun experiment for you). Further, liquid nitrogen is not flammable, and presents no hazards other than asphyxiation and freeze damage. Nitrogen already makes up 80% of the air we breathe, so unless one works in an enclosed space with plenty of NL2 boiling off, it's tough to die from asphyxiation.

In other words, LN2 is colder, and won't blow up on you. I've used it for years, and have yet to get hurt by it. A little respect goes a long way.

The Calculation

[div_B]

Radiation is not nearly as efficient as conductiion hence the need for so many fins on a heat sink. So I still think it would overheat extremely fast.

I think you'll find that the fins are to increase surface area for the purposes of convection. Convection of course dominates radiative transfers in a fluid like air.

As for radiative cooling in space, a quick ball park calculation is quite educational:

Objects emit radiation depending on their temperature, according to Stefan's law. They also absorb radiation from their surroundings according to the same equation, hence we can express the following formula for net power emitted as

P_net = {sigma}*A*e*(T^4 - T_0^4)

Here {sigma} is the stefan-boltzmann constant, 5.67e-8 W/(M^2*K^4), A the surface area of the object. T is the temperature of the object, and T_0 that of the background. e is the emissivity of the object, which we will assume to be 1 (perfect blackbody).

I saw a photo of the thermometer displaying -46 deg C(=227 K), and standard Pentium 4 3GHz apparently consume about 80 watts of power. We'll therefore assume that the madly overclocked P4 produces 200W of heat. The question is then, what area of radiator is required to maintain the chip's temperature, given that the temperature of deep space is about 3K (cosmic background radiation)?

A = P_net / ( {sigma} *(T^4 - T_0^4) )
= 200 / (5.67e-8 * (230^4 - 3^4) )
~= 1.3 m^2

An area of 1.3 m^2 corresponds to a sphere of radius 30cm. Conclusion: Put the chip in good thermal contact with a well-emitting sphere big enough to contain the chip and motherboard, and it'll probably be fine.

Re:Alienware's next product.

[notanatheist]

Yeah, it's not the preppies buying them. It's the lazy kid's parents buying them! The preppies are more concerned with how they look, OTOH the rich little snob kid likes how pretty his machine is and blames his gaming skill on lag.

Re:What's that plugged in the DIMM slot?

[KarateBob]

I think it *might* be the OCZ DDR boster [ocztechnology.com], that lets you increase the RAM voltage to an insanely high 3.9v

Nah, LH makes for some serious problems.

[caveat]

I wrote a paper on Type I superconductivity (appears in metals when cooled to a few K of zero; ceramics are a totally different beastie) in school and got diverted into reading up on ultracryogenics for a few weeks - apparently at temps that low, you get all sorts of problems like extreme brittleness and differing rates of thermal expansion, the latter being a fairly major issue in designing an ultracryogenic system. There's a good chance the CPU die, wires, and case would all tear away from each other and destroy the thing. Not to mention that lead superconducts at 7.196K [superconductors.org]; i wonder what resistanceless solder would do to a mobo...

Re:Just in time...

[tzanger]

Out in the boonies it's very common to run oil for heat (in addition to wood).

Re:Erm...

[ionpro]

That's 333Mhz BEFORE clock-quadding. So it's actually an effective FSB of 1332Mhz. This probably isn't the highest FSB overclock achieved so far, but it is unusual to get such a high overclock on the top of the line (i.e. highest multiplier) chip.

Re:Even colder...

[The Master Control P]

The problem with liquid helium (This made MRI scanners horribly impractical for a LOT of years) is that it has 1/20 the heat capacity of nitrogen, and you have to suck a thousand times the power to get down to helium temperatures compared to nitrogen. There would also be no quantum anomalies with silicon. It can only be compelled to a superconductive state under Extreme pressure [superconductors.org].

BTW, which is it... are we mounting it in a vaccuum or under liquid helium :)

Re:6 GHz is not that impressive.

[Anonymous Coward]

Well they wouldn't get very far with FORTRAN, as OFW uses Forth :)

Re:calculate pi...

[Anonymous Coward]

It looks like a geometric series, with |r| less than 1, so its going to converge. There's a closed formula for it, like a*[1/(1-r)] -- or something like that.

Re:Pi has been slashdotted

[bairy]

Yeah like that's impressive compared to this [1415926535...944592.com]

Re:Just in time...

[Xabraxas]

Boonies? I live in the city and we have oil heat. It's quite common in the northeast.

Re:Don't forget the dual clocked ALU

[Aadain2001]

Self resetting domino logic has been known about for a while now (I think since mid 90's), but Intel was the first company to actually do something useful with it in a real product. Synchronizing it to the rest of this chip shouldn't be that hard if you understand domino logic and use memory cells at the input and output ports of the ALU. I personally don't know exactly HOW they do that, but that's my best guess.

Re:Mod parent as clueless

[einhverfr]

I do know enough about number systems to derive my own conversion mechanisms. So yes, I know what base 16 digits are, thank you very much.

It is possible that we are talking about different things. I think of a digit extractor as something which will let you calculate the nth digit without having to calculate any other digits.

Mathworld mistakenly called this equasion a "digit extraction algorythm" and this is a mistake IMO because the equasion still requires that you calculate every preceding digit. This is because the portion of the equasion before the (1/(16^x)) will not produce whole numbers, even in hex. The numbers are less than 0 and therefore must be calculated before the next digit. Does this make sense?

The problem is that for any value of x where x > 0, 4/(8x+1) - 2/(8x+4) - 1/(8x+5) - 1/(8x+6) always yields a fraction whose denominator is *not* a power of 2 and hence not a power of 16. So you end up with a value which does not exactly correspond to any single digit in pi (even in hex).

Now this equasion is very helpful because I could write a program to calculate n digits of pi in hex and speed things up by shifting values off when they are no longer needed for active calculation. But it will *not* allow me to calculate the thirty-four billion 396 millionth decimal of pi without calculating all prior places.