Slashdot Mirror
Pentium 4 631 Overclocked to 8 GHz
Andreas writes "There are always those who are willing to take things one step further than others. A group of guys known as OC Team Italy is one of them. They recently pushed an Intel Pentium 4 631 to over 8000MHz using an ASUS P5B with modified voltage regulation and liquid nitrogen. Overclocking is cool and all, but this extends beyond what some would perhaps call useful. Still a milestone though."
All the trouble those Italians do to cook sausage without burning it.
To save thoughs who just want to see the setup pictures
Heil....HeilHitlerHe.. #In Memory of Adolf Hitler# ..Heil....HeilHitlerHe ....Heil....ilHitlerHeil .....Heil....Heil........ ......Heil....Heil........ ....Heil....Heil........ ...HeilHitlerHeilHitler .HeilHitlerHeilHitler
........Heil....Heil..... name will be vanquished. ..........Heil....Heil
........Heil....Heil.... You were a true patriot ............Heil....Heil
........Heil....Heil.... and a lover of all men, ............Heil....Heil ...HeilHitlerHe....Heil ..ilHitlerHeil....Heil
Heil....ilHitlerHeil.... We will always remember
Heil....Heil............. and cherish you. Your
Heil....Heil............. acts of selflessness
Heil....Heil........... will be passed down from
HeilHitlerHeilHitler... generation to generation.
HeilHitlerHeilHitler... The lies that dishonor your
HeilHitlerHe....Heil... all races, all religions.
ilHitlerHeil....Heil.. #In Memory of Adolf Hitler#
Thats just in time!
Vista is released in a couple of days, we need at least one machine up to spec.
liqbase
It's also how fast your circuits can switch, and how fast the signal can travel on the wires. The execution core of a Pentium 4 also happens to be double-pumped (i.e., it performs operations on both edges of the clock signal). Essentially, those ALUs would be switching at 16GHz ... I, personally, take this with a grain of salt.
The Raven
Get with it guys, now it is about making silent fanless but powerful systems....
Not creating a CPU that sucks down 300W+, has one core and generally sucks.
I once clocked a 286 to 30mhz! It caught fire and burned an entire city to the ground.
Indeed. Light travels just under 2 centimetres in the 16 GHz period. The Pentium 4 core is not much smaller than this... it seems like they're pushing their luck on order-of-magnitude estimates alone.
setup2
Thermometer at -192 deg.C
photo of screen at 8000.7MHz
CPU-Z verified
The extreme cooling they are doing is not just for removing the heat generated by the chip. As temperature decreases, the mobility of charge carriers increases, allowing for a faster circuit. In fact, if they were to run a supercooled chip at the nominal clock frequency, they would have hold time violations and the chip would not work. In other words, the data would propagate so quickly that it would corrupt the previous piece of data.
Is 8000 MHz supposed to sound more impressive than 8 GHz?
I'm just confused as to why it was worded so oddly.
as my core2 duo e6600 oc'd to 3.4 GHZ Big Smile.
Enjoy Every Sandwich
Funny the pack of cigarettes with the government mandatory sign: "Il fumo uccide" (smoking kills...) besides the smoking board...
just a spelling varient on "those" :p
What I'm more curious about is how the frak they managed to get a FSB of 1,5 GHZ on a Pentium II 333 MHZ
http://valid.x86-secret.com/show_oc.php?id=159352
Since when is increasing CPU OPS not useful? Can new drugs be designed on the fly now? Do we have games that have Lord of the Rings level computer graphics rendered in real time? Until they can get me a holodeck and do molecular computation for drug design in real time .. then dont tell me that 8 GHz is more CPU than we ever need.
Indeed. Light travels just under 2 centimetres in the 16 GHz period. The Pentium 4 core is not much smaller than this... it seems like they're pushing their luck on order-of-magnitude estimates alone.
Actually it travels around 18.7 meters in that 16GHz period. It'd travel just under 2 cm for a 16 THz chip.
You're making light way too slow. If it were THAT slow, then even 100 MHz chips would be impossible.
So: yea, sorry to break it for you, but a 16 THz chip maybe will never be possible (unless it's super tiny, I guess...).
Nope.
16 GHz = 16 billion Hz = 16 billion cycles per second.
((1 second) / 16 billion) * c = 1.87370286 centimeters
One foot per nanosecond is the famous rule of thumb, and 16ghz is 16 cycles per nanosecond. (One nanosecond = one billionth of a second, using US terminology. 10**-9.)
f = 16 GHz = 16 × 10^9 1/s gives a period t = 1/f = 0.0625 × 10^(–9) s. Distance x = ct = 3.00 × 10^8 m/s × 0.0625 × 10^(–9) s = 0.019 m. But yes, a THz chip would be seriously up-fucked.
Your wise and impartial comments are always appreciated!
Obligatory Google Calculator link
Everyone knows bigger MHZ is still king. It's just scale. How much you can do with the given clock cycle. I think they meant that it is not practical to run a processor at 3x it's normal rating using Liquid N as a coolant. It's only useful for the duration of the Liquid N supply, and that is a small Finite amount of time. Secondly, they overclocked the CPU but not Ram(according to CPUV which showed the ram @ 533 mhz) So we have the old bottleneck situation again...
The real question here is "Does MC Lag during battle?"
How much is your data worth? Back it up now.
We were supposed to have those 10GHz Pentium 4s last year. Well, it's a start.
I might be wrong, but for all I know, light has nothing to do with it. We're not talking photon-computing, right?
How fast electrical signals travel through the wires is depending on the material the wires are made of. Light has nothing to do with it. I highly doubt that the conductors used are up to the 16GHz challenge, but they might be at those temperatures.
Free beer is never free as in speech. Free speech is always free as in beer.
First of all, core 2 duo would be two cores at 3.4Ghz so even in a totally perfect world, you're at best getting a collective 6.8Ghz. Furthermore, it isn't a perfect world. There's a certain amount of overhead in distributing tasks to each processor. Furthermore, the performance is limited based on how well threaded the applications are that you're running. A badly threaded application will likely never be faster than that 3.4Ghz on one core.
This sig has been temporarily disconnected or is no longer in service
8GHz and 640kB are all anyone will ever need. (Yes, I've read Snopes about this)
Tm
Support TBI Research: http://www.raisinhope.org
I notice that they use CPU-Z to monitor this CPU. Seems like a pretty good tool to monitor the CPU. Get a copy here http://www.cpuid.org/
And as a harware engineer: As long as you dont boost the voltage too much (Which these guys prpbably did), you can not damage anything, so go for it.
don't cut it off www.mgmbill.org
As I understand, semiconductor conductivity is dominated by the number of charge carriers, not by their mobility (as in metals), and the number of charge carriers generally decreases with temperature due to lack of thermal exitation. Does this becomes unimportant when doping is used to control the number of charge carriers? And either way, isn't the speed of transmission fairly constant?
8000 mhz should be enough for anyone...
does it make sense ? and power usage ? Just some more Watts ? I think that "smart" way is the right one.
While light itself may not have anything to do with it, the speed of light c most definitely has. It's the upper speed limit for, well, everything. Including propagation of signals.
Tedious Bloggy Stuff - hooray?
Why are there no useful benchmarks? Danger of cooling system dying off?
"Champagne for my real friends - and real pain for my sham friends!" http://ericblade.postalboard.com/
Yep, funny stuff happens when you try to move things faster than the speed of light. Gotta start talking about Quantum Mechanics and Einstien's theory of relativity to make sense of it all. :P
I have thought a bit about this before. Actually, electromagnetic waves travel even slower than the speed of light in silicon because the relative permittivity of about 10 limits the speed to around a third of it.
The conclusions I reached are that CPU design has to take this speed into account and that this is probably one of the reasons why big chips need some kind of staged pipeline system.
The speed of light in a vacuum (c) is the absolute maximum speed at which information can travel. It doesn't matter how much you cool the chip or what materials you make it out of, given our current understanding of physics* you can't push anything through it faster than 3*10**8 m/s. That gives you an absolute cannot-be-bettered upper limit for the distance that your signal can move in one cycle.
(* which might be wrong, but no-one's managed to prove it wrong yet)
*Light* has nothing to do with it, it's relativity and the *speed* of light in a vacuum that's important.
It's official. Most of you are morons.
Isn't that the supposed upper speed limit for objects? Electric signals in a wire can go faster than the electrons are moving, or so I think (electrons "hopping" to atoms, as such not travelling the width of the atom, etcetc... might be terribly wrong here). If that is true, electrical signals can in theory travel faster than light.
I once heard of a experiment at CERN (I think) where they managed to push a particle beyond the speed of light in that enviroment, creating a light-wave very much akin to an ultrasonic boom. Don't know the details.
Free beer is never free as in speech. Free speech is always free as in beer.
... a nice stable 4.0 GHz on simple air cooling out of a P4 and also without having to overvoltage it too much. I think that would be just about as much processor horsepower as you could actually utilize in an x86 platform machine for practical general use and gaming too. I have a 3.4GHz P4 (a socket 478 at that) that'll do a good stable 3.8GHz on air cooling with a big copper Thermalright heatsink and a panaflo 92mm fan that doesn't sound too much like a vacuum cleaner running. It does put out an incredible amount of heat however, and I can run it at 4.0GHz on this same air cooling setup but only for very short period of time (maybe 1 to 2 minutes max before it overpowers the heatsink's capacity to dissipate the heat. I'm sure it would be plenty stable if I used water cooling, but that's a hassle and I don't want to risk a leak damaging my machine. Lately I've just been running it at the stock 3.4GHz speed because it just makes too much heat at 3.8 GHz though it does definitely give me a gaming advantage in UT2004 at the higher speeds.
> Isn't that the supposed upper speed limit for objects? Electric signals in a wire can go faster than the electrons are moving, or so I think (electrons "hopping" to atoms, as such not travelling the width of the atom, etcetc... might be terribly wrong here). If that is true, electrical signals can in theory travel faster than light.
Despite all this, in theory, electrical signals do not travel faster than light.
> I once heard of a experiment at CERN (I think) where they managed to push a particle beyond the speed of light in that enviroment, creating a light-wave very much akin to an ultrasonic boom. Don't know the details.
Look up cherenkov radiation. The particle in question might have gone faster than c in that medium, but it didn't manage to go faster than the speed of light in vacuum.
Yep, funny stuff happens when you try to move things faster than the speed of light. Gotta start talking about [...] Einstien's theory of relativity
I wouldn't consider "it's impossible" funny stuff. You either get a divide by zero or a divide by an imaginary number.
The electron pressure pulse (read:signal travels very close .99 ?? to the speed of light)
The electrons themselves would be lucky to move a single millimeter down a wire every 10 seconds (includes wires in your home power system (DC and AC but in AC they just go back and forth at 50Hz), etc) thats why were using the speed of light here
Light has everything to do with it. Light is a wave in the elctromagnetic field, changes at which propogate at c in a vacuum. Signals travel down "wires" at somewhat less than c (typically 65-80% * c), dependant on the dielectric constant of the insulator around the wire as well as the geometry of the conductor. In general, the electrons move much slower than this, with instantaneous velocities of a small fraction of c and drift velocities on the order of a few cm/s.
HYDROGEN! At a boiling point of -252C, they should be able to get about 60 degrees cooler. They should be able to run even faster. I can't imagine any other concerns.
>> As long as you dont boost the voltage too much (Which these guys prpbably did), you can not damage anything, so go for it.
The thermal stress caused by varying rates of thermal expansion for silicon, the resin underfill and the package puts a a lot of stress of the flip-chip bumps cycling between "room temp" and cryogenic temperatures. I'm not so sure that I'd say this isn't going to damage anything.
There'd like be no problem if you do it a couple of times, but over more thermal cycles, I'm certain the some of the bumps would start to shear from metal fatigue. There's been studies of flip-chip designs on Mars examining long-term reliability of the bump solder connections thermally cycled from day to night, and most conventional design last less than 100 cycles. If you design for this, you can fix it but I am pretty confident that the Pentium 4 was not designed for cryogenic cycling.
Swedish plasma phys. PhD student; MSc EE; knows maths, programming, electronics; finance interest; seeks opportunities
Even then it doesn't really make sense most of the time.
As the Americans learned so painfully in Earth's final century,free flow of information is the only safeguard against...
Gate and metal delays will both be important for this processor, but I would guess that the total delay between registers would still be dominated by gate delays (about 70-80%). You are right that drift velocity has a linear correlation with dopant concentration. However, this linearity ends when the doping level is too high or the electric field is too strong; such a situation is called "velocity saturation" and occurs at about 10^7cm/s. Otherwise, you could just increase your doping level (or voltage) and get an infinitely fast transistor!
/.
That was an interesting point you brought up about how charge carriers vary with temperature. As long as the temperature remains above a certain point, there is enough thermal energy for ALL the electrons/holes to leave their parent atoms. I know that for Arsenic this number is -173C but I don't know any others. Liquid nitrogen has a boiling point of about -200C, but I doubt the transistors would actually get so cold that the carriers stop moving completely. Anyhow, if that happened, then we wouldn't be reading this news story on
Mobility does decrease with temperature; the explanation that I have been taught is that lattice vibrations prevent the charge carriers from moving freely as the temperature goes up. It is possible that the velocity saturation limit goes up when the chip is so cold, but that is more of a physics issue than an electronics one so I don't know. The intuition could be that the lattice is so stable that higher currents are enabled. Alternatively, it is possible that the circuits do not nominally operate anywhere near velocity saturation and the supercooled regime accomplishes the same effect as increasing the voltage (without fear of breakdown/shoot-through) or increasing the dopant concentration (without a decrease in carrier mobility).
By the way, I forgot to mention that increasing dopant concentration has the side effect of decreasing the carrier mobility. So that is another reason why increasing the doping level beyond a certain point would not give you a faster transistor.
In other words, the speed of transmission is not constant. Hope that answered your question!
They've gone to plaid!
What is really hard to believe is that only 2 days after the successful overclock, Microsoft sent him (for free) a Ready for Vista decal for his PC.
Aw, crud. I can't resist now.
Imagine how hot a Beowulf cluster of those could get!
And why are they running W95? Don't they know that it'll BSOD Much faster?
Or if it uses light perhaps?
The "speed of light," by definition, is the speed at which all electric fields propagate (not just optical ones). Even though the wire is treated as an object with constant voltage on it, physically, the electric field which creates that voltage is outside of the wire. In fact, you'll find that as long as the conductance of the wire is sufficiently high, it has little effect on the speed of signal propagation. This is because at the frequencies being discussed, the wires behave more like transmission lines than the ideal, lumped-element model used in circuit analysis.
What's actually more important to the propagation speed is the permittivity and permeability of the dielectric (insulator) surrounding the wire. As it turns out, the speed of signal propagation is identically equal to the speed of light in the dielectric medium (not by coincidence, of course). I may be wrong about this, but I believe that modern processors still use undoped silicon as the interconnect dielectric medium, which means that the signal propagation speed is c/3.4.
In other words, GP was right. In fact, this is just the speed of light in vacuum. In practice the waves propagate in a medium with a higher dielectric constant (epsilon_r>1), and therefore the wavelength will be even lower. Moreover, even when circuit dimensions are lamba/10, transmission line effects will come info play...
He probably threw in c as 300,000km/s instead of 300,000,000m/s. That happens a lot, and results in order-of-magnitude errors.
occultae nullus est respectus musicae - originally a Greek proverb
And as a harware engineer: As long as you dont boost the voltage too much (Which these guys prpbably did), you can not damage anything, so go for it.
Isn't that sort of like going to Lambeau Field and seeing a football and explaining to everyone that its safe to throw it?
Perhaps you should stick to free speech. The free beer is screwing up your physics...
Actually, I believe they removed the double-pumped ALU from Prescott and later to allow for even higher clock rates.
Opus: the Swiss army knife of audio codec
...the fan on the GPU in the photo with the Fluke thermometer. Why isn't the fan spinning?
Informatus Technologicus
Of course, in one clock cycle a signal does not need to travel from one side of the chip to the other, just to the next stage in the pipeline. But that still is a cool statistic!
Pentium 4 is based on a Netburst architecture.
Core 2 duo is derived from a Pentium D architecture (which was itself carried from Pentium3 / Pentium2 / Pentium Pro).
They're completly different animals and definitly not doing the same stuf during 1 cycle.
C2D 6.8Ghz can't be compared to the 8Ghz overclock.
That's also why Intel switched from advertising MHz/GHz to advertising number of cores. Otherwise, newer and faster would have been considered by joe 6-pck because he's been trained to look for the "GHz" and Core Duo happens to have lower clock for similar performance as pentium 4s.
Whether this new "old GHz = new Num_of_Cores" marketing craze actually means something or is just hype (as opposed to trying to add task-specific coprocessors like IBM's Cell and AMD's Fusion etc.) is left to the speculation of the reader.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Obligatory Google Calculator link
:P
Yup.. I was wrong in the end, not him. Thus proving once more than moderators mod you up if you sound right, although you may be wrong. The original poster got modded down... Ain't that ironic
Of course maybe it'll be regulated out in time as people hit the clarification posts.
How about we have two signals moving away from each other at close to 2*c then?
My hypothetical computer is faster than yours!
Correct. Electric current moves through a wire slower than c (but can be on the same order of magnitude). So apparently width of chip = wavelength of clock isn't a problem or it would already be a problem.
Prov 9:8 Do not rebuke mockers or they will hate you; rebuke the wise and they will love you.
This brings back the old M&M's marketing phrase, with a twist:
"Pentium melts in your PC, not in your hand"
"It is a denial of justice not to stretch out a helping hand to the fallen; that is the common right of humanity."
The speed of light in a vacuum (c) is the absolute maximum speed at which information can travel. It doesn't matter how much you cool the chip or what materials you make it out of, given our current understanding of physics* you can't push anything through it faster than 3*10**8 m/s.
I wonder though. If 8 GHz (in two steps effectively 16 GHz) is bordering on impossible due to lightspeed... why was Intel claiming to have 10 GHz chips back then?
And in the end, they didn't produce them due to various issues, like heat, power consumption and so on. No one mentioned speed of information travel.. Weird huh...?
Did they plan to make the chips a lot smaller at that time?
and hope they have paid their fire insurance bill recently.
-- Tigger warning: This post may contain tiggers! --
Overclocking is cool and all, but this extends beyond what some would perhaps call useful.
For the proverbial Aunt Tilly, this may be true. However, "too much CPU capacity" is a non-existent concept for those of us who chronically peg the CPU at 95%+ doing raytracing or other kinds of graphics rendering including but not limited to video editing, large-scale database querying, molecular modeling (or pretty much any non-trivial simulation), automated generation of static reports for instant-access web pages, large-scale text analysis, machine learning of all kinds, and no doubt many other tasks.There is no such thing as "too much CPU capacity." There never will be. The more we have available, the more we things can do. The more things we can do, the more CPU capacity we need. Same goes for RAM, disk storage, net bandwidth, and pretty much any other finite resource.
If you don't believe it, or if this is a new concept for you, you're not trying hard enough.
...are we scared yet?
The interconnect dielectric is usually silicon dioxide, with a relative dielectric constant of 3.9. This puts the propagation speed at about c/2.
Contribute to civilization: ari.aynrand.org/donate
Speed of light limitation is a problem. I've read that in the P4, Intel had pipeline stages that did nothing but compensate for propagation delay across the chip.
Contribute to civilization: ari.aynrand.org/donate
Actually it travels around 18.7 meters in that 16GHz period. It'd travel just under 2 cm for a 16 THz chip.
Just off the top of my head 2 meters is about 146 MHz in free space so I believe you are off by a factor of 10^3. Dusting off my HP48 shows that 16 GHz has a wavelength of about 18.7 millimeters. I am used to doing these calculations based on wavelength for RF design but using period gives the same results. 16 GHz is 62.5 picoseconds which again yields 18.7 millimeters at the speed of light.
SiO2 has a dielectric constant of 3.9 for a 0.5 velocity of propagation and SiOC is about 2.7 for a 0.61 VoP so the actual distance is about 10 millimeters. IC designers have been dealing with propagation delays that are greater then the cycles times for a while now.
In fact, if they were to run a supercooled chip at the nominal clock frequency, they would have hold time violations and the chip would not work. In other words, the data would propagate so quickly that it would corrupt the previous piece of data.
That's not necessarily true - it's only the case if the logic paths speed up more than the clock paths. You get a hold time violation if one flip flop launches its data, the data gets through the logic, and arrives at the capturing flip flop before (or too soon after) the clock signal has arrived at that flip flop. As long as everything speeds up by about the same amount, the clock will arrive at the receiving flop quickly, and the "hold time" of the receiving flop (how long after the clock arrives you have to keep data stable) will go down too. Chip manufacturers use HUGE safety margins when it comes to hold violations (partly to handle process improvements by the fab, and partly because if there is a hold failure you can't fix it just by changing the clock speed--you just built a paper weight or keychain).
My server
obviously hasn't seen the producers
More than just my physics, I can tell you that...
It's still amazing how much you learn by being corrected. The beer will have washed away all that by the morning, but still...
Free beer is never free as in speech. Free speech is always free as in beer.
"Did they blow anything up?", my wife asks. No? Well then, move along, nothing to see here!
Actually there is a way to get around the whole 'speed of light' issue - don't use light.
... now that is 'hauling ass fast', also known as immeasurably fast. When you turn on a light by flipping a switch - the light takes a measurable amount of time to get to you, but when does the light actually turn on? The instant you flip that switch - ahhh, the magic of electricity running at immeasurable speeds over wire.
Yea, the 'photonic computer' guys didn't think that one all the way through, did they?
Use electricity instead, have it run on little traces cut in silicon like the old days, but then seal the silicon in a dark ceramic casing so no light gets in, and put the whole thing in a computer case WITHOUT the clear panels - have to keep out the light.
Light is fast, no doubt, but it is measurably fast (186,000 miles per second, as I recall) - but regular electricity running in the dark across wires (or traces on silicon)
Think about it - every scientist in the past century has measured the speed of light - but how many have been able to measure the speed of electricity in a wire?
None?
Bingo!
And what kind of tools do they use to measure the speed of light?
Electronic tools made with electricity running on wires?
Bingo!
Glonoinha the MebiByte Slayer
in fact... speed of light in vacuum is more or less 3E8, in coper or semiconductors will be far less... n=c/vp, were n is the refraction index of the material. for copper n~1,1, so vp~2,73E8 (1.7cm at 16GHz); for silicon n~4.24, so vp=0.7E8 (0.4cm at 16GHz). As we can see, this setup is even more questionable.
Math is beautiful... e^(pi*i)+1=0
6400 MHz ought to be enough for anybody.
"Destroy science and religion. Science would re-emerge exactly the same; but not religion." - Penn Jillette, paraphrased
Your explanation sounds like it could be reasonable for an asynchronous circuit but not for standard CMOS. Assuming that the processor is using a balanced clock tree, the clock signal should arrive at the two flops at the same instant, no matter what the propagation delay through the clock tree is. As the chip gets colder, the propagation delay through the datapath decreases until finally a hold time violation occurs. Throughout this entire process, the clocks can be assumed to hit both of those flops simultaneously which is valid for a balanced clock tree.
I did make a mistake in saying that hold time is dependent on clock frequency. This is not true for the reason that I explained above. However, the hold time will eventually be violated as the propagation delay between registers shrinks with decreasing temperature no matter what the clock frequency is.
FWIW, the speed of light in a vacuum isn't "more or less 3E8". It's precisely 299792458 m/s. The meter is actually defined as the distance light travels in 1/299792458 seconds.
Cow Cube
except that copper being a conductor would have very strong absorption (which translates into a large imaginary part of the refractive index), therefore the light would never propagate more than a few hundred nanometers. in aluminium for example the penetration depth is about 100 nm and i guess in most metals is not higher than a fraction of a micron (10^-6 m)
Check out these Australian cigarette packets.
... and then they built the supercollider.
they exceed the theoretical maximum of warp 10 ?
I sat down to write a new sig tonight and all I did was make the chair warm.
Sure you can! If the temperature changes too fast (creating differential thermal expansion/contraction) you can physically crack the chip (or a PCB, or whatever).
"[Regarding the 'cloud,'] ownership was what made America different than Russia." -- Woz
No, it's central to that microarchitecture. Intel can't remove it without completely redesigning the chip.
How fast electrical signals travel through the wires is depending on the material the wires are made of.
Actually, the velocity of propagation equals the reciprocal of the square root of the dielectric constant of the material through which that signal passes.
--fatboy
10 GHz was probably not a real number. It could have been like AMD's Athlon 3100+, where it'd be that much of an increase in power, rather than actual cycles per second. Or, we're looking at calculations on both sides of a cycle, same way the poster further up got 16 GHz for the ALU on this chip. If you only do 5 billion cycles per second, but you can do two things on each cycle, you get 10 billion things. Then they went with an extra core instead.
I see your informative link, and raise you a pithy comment.
what! Not precisely 299792458.7?
A Good Troll is better than a Bad Human.
Indeed. Light travels just under 2 centimetres in the 16 GHz period.
So? Who says that the clock domain stretches from one side of the die to the other? I'm no guru on CPU architecture but I would be surprised if anything switching at the maximal rate were not confined to a much smaller area.
Yeah, that makes more sense: I don't know what I was thinking. It's quite difficult to grow crystals on already fabricated devices. :-)
P4? Please. I might have actually cared if it was a Core2.
Now I could be completely wrong here, I've taken a whopping one physics courses in my career, but the speed of electricity is not immeasurable, especially using AC current. AC current, as I recall, is just a bunch of electrons bumping into each other moving in one direction and If i remember correctly the net speed in the direction of current is quite slow. This is not true of DC, don't even ask me about that euro trash. And as a side note, do we not know the speed an electron travels in "orbit" about the nucleus? Is that not the "speed of electricity? Also, is any of this true or does UC Santa Cruz just have some crazy-ass drugs?
Measuring the speed of light to 1% accuracy with junk-drawer parts and Ebay bargain istruments is not trivial, but it can be done.
The determined Real Programmer can write Fortran programs in any language.
Many. Electricity in a wire can be likened to those pendulum ball desk thingies. But without the strings holding them in place. However, you are correct in that the electrons turn on the light faster over the same distance than it would take light to travel. The individual flow of electrons is quite slow, in fact you can walk faster than any individual electron moves through the wire.
It's possible because they used the flux capacitor
in b4 9000
FWIW, they're not running Windows 95; it's XP with the "Windows Classic" theme applied.
This, of course, is assuming that any signal inside the P4 needs to travel across the whole chip in one clock period. I believe the entire philosophy behind the Netburst design was to avoid signals traveling too far. Signals that need to get from one side of the chip to another are given multiple clock cycles (and subsequent signals after them are pipelined) to do so.
Analog components like capacitors and inductors can catastrophically fail at high frequencies. Electrolytic capacitors can explode.
Some digital electronics are more sensitive to high frequencies than others.
But it still won't run Oblivion smoothly...
You moved your mouse. Please restart Windows for changes to take effect.
You are confusing penetration depth in a material with attenuation of a propagating electromagnetic wave over the surface of a conductor.
Math is beautiful... e^(pi*i)+1=0
p4 with 8Ghz.... must be as fast as core solo with 1.8 GHz.
At least!!!!
Actually it is the perceived speed of light that is the upper limit. In other words, the maximum speed light can ever be measured at is C.
e rThanLight
The key point in the above is the word 'measured'. There is nothing that prevents the universe from providing an alternate way of propagating information. For example:
http://www.archive.org/details/IanWoolfMozartFast
There are phenomena like the Tunneling effect or the Quantum Entanglement effect that have not yet totally understood and explained...these phenomena might one day be used for breaking the speed of light barrier.
I am actually more impressed with the 3rd result, obtained using a PII originally clocked at 333 Mhz.
It managed to reach 7640.28 Mhz. Reference here : http://valid.x86-secret.com/records.php .
Direct link to results here: http://valid.x86-secret.com/show_oc.php?id=159352
Light travels only in vacuum at 3 * 10^8 m/s. In copper, light speed is around 2/3 of that value. I forgot what the value for silicon was though, but somewhat in that area.
So the wavelength would be:
vacuum: x = c / f = 3 * 10^8 * 1/16 * 10^-9 m = 3/16 * 10^-1 = 0,1875 m
copper/silicon: x = c(copper) / f = 2 * 10^8 * 1/16 * 10^-9 m = 2/16 * 10^-1 = 0,125 m
Even shorter! I begin to doubt whether this 8 GHz are real...
It's also how fast your circuits can switch, and how fast the signal can travel on the wires. The execution core of a Pentium 4 also happens to be double-pumped (i.e., it performs operations on both edges of the clock signal). Essentially, those ALUs would be switching at 16GHz ... I, personally, take this with a grain of salt.
You do know that 8GHz was the target speed that intel always intended to get netburst up to, don't you? An intel representative even stated once "We have positive indications to be able to take Netburst to the 10 GHz space." Only power management and thermal dissipation issues prevented them from releasing a commercial product that was this fast. As I understand it, this particular feat is merely a replication of something Intel did in their own labs over a year ago (although I forget where I read the description of that setup, and a quick google hasn't turned it up).
Gravitation has been measured to travel up to 1.2 times faster than the speed of light, although these measurments are not agreed upon by the entire scientific community.
See http://arxiv.org/abs/astro-ph/0302294 for statement.
Indeed. Light travels just under 2 centimetres in the 16 GHz period. The Pentium 4 core is not much smaller than this... it seems like they're pushing their luck on order-of-magnitude estimates alone.
1. Only one pipeline, which only performs ADD instructions, is clocked at twice the core speed. This section of the chip is *very* small, probably containing only in the order of 1,000 transistors. Some additional support circuitry (e.g. the register write-back stage and instruction queue) will also need to operate at this speed, probably for a total of around 5-10,000 transistors, or an insignificant fraction of the total die area of the chip.
2. One of the design principles of the P4 was to break down the processing into 20 stages so that critical signals didn't have to travel much distance across the chip within a single cycle.
The P4 was originally expected to reach speeds of approximately 10GHz (i.e. with its fast ALU pipeline running at 20GHz); I suspect some fundamental (but easily calculated) limitation like signal propogation delays would kick in at that point -- Intel would have known about these limits back when they made that statement.
speed scales inversely with the square root of permittivity*permeability.
Also FatPhil on SoylentNews, id 863
I for one welcome our overclocking overlords.
'Course you'll want to run BeOS so you can keep an eye on is_computer_on_fire()
How much is that in nano-fortnights per rod?
The AACS key is NOT 0xF606EEFD628B1CA427BEA93A9CA9773F
Um no. If there is a way for *information* to propagate faster than the speed of light, this violates causality. For example, from one reference frame, I could repeat what you just said, and in another reference frame, you could repeat what I just said. One can cause the other, but they both can't cause each other, which would be possible with sending information faster than c.
OK that was a bad explanation. Just know that information travelling faster than the speed of light violates causality, according to special relativity.
Quantum entanglement and "spooky action at a distance" can cause a change in a system to propagate faster than the speed of light (instantly, in fact), but it cannot be used to send information. And yes, IIAQP (quantum physicist).
8Ghz? that's easy. All you need is 4.77Ghz processor, and then you press the turbo button
You are interpreting .98+/-.19 c, which is the result given in the paper. It does not mean that some of the signal went at .98-.19=.81c and some of it went at .98+.19=1.17c. It means the researchers are 95% confident that what they measured traveled between .81c and 1.17c.
Ah, yes, you are correct. It does leave room for the possibility of >c speed of information, but I don't really believe in it myself.
Scientists? Take a good oscilloscope and a fast rise pulse generator (a crystal can from a 486 board does for your purposes), some twenty feet of cable, and measure it yourself!
This is essentially, BTW, what you can buy in packaged form as a TDR.
Alternate experiment: multiple length cable loops around a 74S04, and a frequency counter.
"Electronic tools made with electricity running on wires?" - No need, you can do a lot with interference patterns, phase shifts etc... without leaving the optical domain. And if any wire is involved... you can very easily ascertain the speed in it by varying its length and observing the difference.
This is a question I've had for a while:
/dev/urandom'). Has this ever been scientifically tested with heavily overclocked processors? Or is it just plain theoretically impossible?
An overclocked processor generates a massive amount of heat, so much that specialized cooling systems need to be installed. Here's my question: is the amount of energy in heat generated greater than the amount of energy required to power the processor?
I believe this may be the case so long as the machine is doing "work" (i.e. hauling bits around at insane speeds, see 'od
-dave
6th Street Radio @ddombrowsky
Only 2GHz more and the P4 will reach the 10GHz once predicted for it.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
I don't see much of a problem with hold-time issues since the hold-time requirements for gates receiving the data also get better with cooling.
I would worry instead about long, pipelined wires, where several bits are in flight simultaneously. While gates get much faster with cooling, signal propagation down a wire doesn't improve very much, so a gate at the other end of a wire would expect a particular bit much sooner than it was available.
Of course it depends exactly on how the circuit is constructed. If the clock driving the gates on the receive end suffers the same propagation delay as the pipelined wire, then everything might work out.
What's really neat in all this how at some point you need to think like Einstein and ask what does it mean for two events in a circuit be simultaneous.
Surly power to frequency ratio is what is now important.
Those pendulum ball desk thingies would only transmit signals instantaneously if they were perfectly rigid, which they are not. In reality, they only transmit signals at the speed of sound through steel (or whatever they're made of). And in fact, relativity sets an upper limit on rigidity, because the speed of sound can't be faster than c. (To take an 1800's view, c is the speed of sound in aether.)
To those of you who think signals in a wire travel faster than c, you're simply wrong. The signal velocity will always be less than c. This does not preclude measuring the speed of light with electrical equipment, you just have to be a bit more careful to take other delays into account. There are cases where the phase velocity could be higher than c, but that doesn't allow you to transmit information or energy that fast.
If you were able to transmit information faster than light, then to an observer moving past you sufficiently quickly, you would be transmitting information back in time. That's why we're pretty sure it's impossible.
Light doesn't travel in copper - at least not very far.
It's not wasting time, I'm educating myself.
And there's nothing funny about imaginary numbers! Children, stop laughing!!
the thing is, rarely are clock trees built as completely balanced structures... usually some sort of clock-tree-synthesis tool is used, and while the tree may be balanced for a particular PVT, care has to be taken to balance transistor delay with metal delay, across the entire structure. hold fixing is a major part of timing closure on an asic.
but in general, i agree with your point, the faster the chip runs (not clock frequency but signal delay time), eventually you'll get hold violations.
"onward!" cried the copper man, little knowing brass corrupts...
errr... by "signal delay time" i meant combinational (eg transistor) delay. that's what i get for trying to think on my day off...
btw, when did this successive-submit delay show up? it's kind of annoying...
"onward!" cried the copper man, little knowing brass corrupts...
Your explanation sounds like it could be reasonable for an asynchronous circuit but not for standard CMOS. Assuming that the processor is using a balanced clock tree, the clock signal should arrive at the two flops at the same instant, no matter what the propagation delay through the clock tree is.
I was talking about CMOS. As the other reply to your message points out, in the real world, the clock tree is never perfectly balanced.
Throughout this entire process, the clocks can be assumed to hit both of those flops simultaneously which is valid for a balanced clock tree.
Not necessarily. Even if we assume that the clock was perfect under normal operating conditions, you'll have temperature variations across the die and other effects (e.g. device variations) that add clock skew.
As the chip gets colder, the propagation delay through the datapath decreases until finally a hold time violation occurs. [snip] However, the hold time will eventually be violated as the propagation delay between registers shrinks with decreasing temperature no matter what the clock frequency is.
No. The hold time drops as the transistors that make up flops gets faster. If you assume a sense-amp style flip-flop, your hold time is probably determined by how long it takes after the clock pulse that the sensing nodes are isolated from the data input pins. The circuitry that does this isolation speeds up when the chip cools, reducing the hold time. You won't introduce a hold violation unless the datapath logic speeds up significantly relative to the devices making up the flip-flops.
Keep in mind that a hold time violation occurs when after a clock pulse, flip-flop A has launched its data while flip-flop B is still trying to record the data that was present on its input at the end of the previous cycle, but the new data from A arrives at B and corrupts the data. It's the same clock edge that is triggering A's launch and B's capture. The problem is worst when the clock arrives later at B than it arrives at A due to non-idealities. In order to exacerbate hold issues, the cooling would have to speed up the clock to A but not also speed up the clock to B. If both A and B get their clocks early, or they both get the clock late, everything is still OK.
My server
I never claimed information can be sent in faster-than-light speeds in the classic sense, i.e. waves travelling through spacetime. But information can travel faster than FTL using the tunnelling effect, as it is obvious from the transmission of Bethoven's 40th symphony posted above.
FTL information transmission does not violate causality, because, for example, at the moment you repeat my words I would already have spoken. The instance you hear my words is not the instance I said those words.
Do you have a link?
Prov 9:8 Do not rebuke mockers or they will hate you; rebuke the wise and they will love you.