Intel Designs Faster, 3D Transistor

lee1 writes "Intel has found a way to keep on the Moore's Law track by making smaller, faster and lower-power computer chips by building 3D transistors. They are already manufacturing microprocessors using this new design, called a FINFET (for fin field-effect transistor), which incorporates a small pillar, or fin, of silicon that sticks up above the surface of the chip. Intel said that it expected to be able to make chips that run as much as 37 percent faster in low-voltage applications and use as much as 50 percent less power. Products based on the new technology may appear some time later this year."

Comments

Rainbow Dash

[Anonymous Coward]

It needs to be about 20% cooler.

Re:Rainbow Dash

[Black Art]

If it uses 50% less power, it will be cooler. The Atom processors generate very little heat compared with the more power-hungry Xeon and Core Duo chipsets.

Intel has become very aware of power usage as well as heat. The days of the room heating CPU are hopefully behind us.

Re:Rainbow Dash

[Anonymous Coward]

If it uses 50% less power, it will be cooler. The Atom processors generate very little heat compared with the more power-hungry Xeon and Core Duo chipsets.

Intel has become very aware of power usage as well as heat. The days of the room heating CPU are hopefully behind us.

Nope. Because no matter how efficient these processors become, there will always be somebody wanting a faster processor, and hey! The power envelope just got more room, just cram more of 'em in there!

Re:Rainbow Dash

[Anonymous Coward]

"If it uses 50% less power, it will be cooler."

Um, no? If you also manage to cram in 4 times as many transistors as before, that's twice as hot.

Re:Rainbow Dash

[Sebastopol]

power and heat are the same thing

Re:Rainbow Dash

[the_humeister]

They may be correlated, but they're not the same thing.

Power=rate at which work is performed
Heat=energy transfer from one place to another

Re:Rainbow Dash

[Sebastopol]

when you refer to heat in a cpu system, you are referring to heat flux, which is joules per second, which is power. i'm not just playing with units. conceptually they are the same: the amount of heat transfer is identical to the amount of work done.

Re:Rainbow Dash

[Anonymous Coward]

the amount of heat transfer is identical to the amount of work done.

Only in terms of units. Actually dU = Q + W, where Q is the heat transfer, W is the work and U is the amount of energy that you put into/take from the system. Either you were being intentionally obtuse, or you don't see that power in this context is referring to the electrical (I^2.R, I assume) power that you're putting into the processor and heat is the waste energy that you get out of it.

I'm not au fait with reversible computing and its implications on the thermodynamics of processors, however the difference between energy in and 'work'* out is necessarily non-zero in contemporary designs.

*Work to me is either d(PV) or F.d or I^2.R but there are of course many others. I might even count heat transfer if we were talking about heat pumps but they aren't processors. Work is usually defined as the useful energy moved in or out of a system.

BTW I used d instead of (delta) because, well... actually after this I give up on moaning about the lack of greek letters here.

Re:Rainbow Dash

[Sebastopol]

i was being neither. i think both of you are focusing too specifically on one domain or the other and not realizing physics as a whole is the same.

even in your equation Q + W, you only gave half of the equation dU is part of the heat transfer equation, but you ignored the transferring part, you know, the part that takes time, the dt.

Try this on: are you telling me that you can put 100W into a processor and less than that comes out as heat? because it all comes out as heat. hence they are the same physical property. if you disagree, congratulations on your nobel prize!

i think you are hung up on car engines vs. transistors, when in reality the heat transfer and power are the same thing in both cases.

this is a huge problem in engineering schools. the concepts are taught too rigidly, and the relationships are never fully explored which leads to misconceptions like the one you are arguing.

Re:Rainbow Dash

[mr1911]

No, because the fundamentals are changing. In the power equation, voltage is squared. So if you reduce the voltage, the power goes down more. You can do more operations per watt. That equals less watts for the equivalent operations.

Re:Rainbow Dash

[Sebastopol]

its a complex equation, but you're safe assuming the cube of voltage. you can be a little more specific: only switching power is proportional to the square of voltage. leakage is proportional just a little over the cube, and rushthrough is a little more than squared with power due to the dependency of transition time in th linear region (slope) on voltage. plus there's also frequency and its dependency on voltage, and temperature affect on leakage and Idsat. but the OP has a valid point, power density does increase, so the cooling solution has to be beefed up regardless because even less watts, over a smaller area, requires much more cooling capacity. it's the hot spot that you need to defend against.

Re:Rainbow Dash

[Anonymous Coward]

Try this on: are you telling me that you can put 100W into a processor and less than that comes out as heat?

That's exactly what I'm saying, and I'm saying that because there is a change in entropy (the information kind). Studying more than one view of entropy is something you accused me of not having done. In a nutshell TANSTAAFL: all those useful computations come at a cost.

U, Q and W can either be Watts or Joules (so long as they match) just as thermodynamic entropy can be specific (per kg) or not. It doesn't really matter in this case since we're not going to assign values to any of them; the energy we put in is less than the heat that comes up no matter how long or short the time interval is. What we're interested in is the ratios between them: that tells you how efficient a processor is.

Try this on again*: dU = Q + W. Now they're all Watts instead of Joules; are you telling me W is zero for processors? I don't think it is.

*I'm sorry, but when I read the preview I just couldn't remove the delta: the internal energy of the system is the starting energy plus/minus (Q + W). This would have been easier if I hadn't resorted to using d, which I suppose might have lead you to think there was a dt needed too.

Re:Rainbow Dash

[Anonymous Coward]

I'm not going to argue this point because I don't know a great deal about electrical power but it's not vital to my argument anyway. Suffice it to say that a certain amount of a electrical power is put into the processor, some is used to perform computation and the rest comes out as waste heat. It doesn't particularly matter how you measure the input power so long as it's in the same units as the output; it could even be the number of calories input/output over (not per) a fortnight or eV if it makes you happy.

The issue I have with the parent is that he/she is telling us the power put in is identical to the power that comes out as heat.

I say it isn't; I say it's greater.

Your point however, is insightful (so you get a +1, Have a Nice Day, etc.): you appear to be saying that the efficiency (ops/Watt, perhaps?) increases if the voltage is lowered. If you'd care to explain this further and include a little on what limits there are on doing so and why (Feynman, forgive me ;), you'll probably get another +1 for the effort.

Re:Rainbow Dash

[Sebastopol]

I think you're getting hung up on the equations.

Let's keep it simple:

You supply 100W of power to black box. If less than 100W of heat transfers out (minus entropy), you've just observed a violation of physics or made a huge discovery.

Voltage regulator pumps in the energy in, heat sink sucks it out (well, so does the board); energy in energy out over time = watts.

Heat transfer = electrical energy transfer = power

There is no measure of "efficiency" because if a CPU demands 100W, it will expel 100W in heat, there's no way around that (minus entropy). (I'm also assuming a closed boundary around the cpu here, for the sake of the visual.)

I think what you're after when you say "Efficiency" is how much computation is done per Watt. But that's whole 'nother ball o' wax. If you are still thinking Carnot Efficiency, remember that is only a measure of the transfer of energy between two opposing mechanisms in a cycle: the boundary around the entire system remains constant (minus entropy).

Re:Rainbow Dash

[Sebastopol]

thought of another example:

if you're delivering energy to a CPU at some # of Watts, you have to remove it just as fast... and it doesn't just vanish. how is it removed? heat transfer. also watts.

Is this getting boring, should I go back to moving more than -2 meters in QWOP now?

Re:Rainbow Dash

[Anonymous Coward]

but the OP has a valid point, power density does increase, so the cooling solution has to be beefed up regardless because even less watts, over a smaller area, requires much more cooling capacity. it's the hot spot that you need to defend against.

Now the ball's in my court - so to speak. FYI, cooling capacity is Joules/second (i.e. Watts) whereas what needs to go up when one has a smaller core (a hot spot, as you say) is heat flux, which is Watts/m^2.

No mod points from me for your post because the first part was quite informative, if a little hard to read. Presentation counts and capitalisation is the first step. Have a Nice Day.

Re:Rainbow Dash

[kyle5t]

This might be true, technically speaking, if storing the information that results from each calculation required work to be done (for instance, you have to do work to change the magnetic moment of the grains in a hard drive platter). But in the CPU this information is just represented with a potential.

Re:Rainbow Dash

[Kjella]

Yep. And in the server world there's also consolidation, you always want fewer and more powerful processors if the cost isn't too high. The typical CPU in desktops and laptops is a different matter though. Maybe the question is more obvious the other way around "Would you like a quieter computer with longer battery life?" and with other marginal benefits like lighter heat sink, lower power bill, probably longer life because of lower temperature and so on. For example many office workers would do exactly the same even if their laptop used half power.

To take another example, if my phone uses half the power I wouldn't make twice as many calls - I pretty much stay connected 24x7 and make all the calls I'd like anyway. Same with say my TV, I wouldn't watch twice as much TV if it used half power. Of course I could start using my phone to play Angry Birds on my way to work instead of reading a book - which may or may not be a power savings depending on how "green" making and distributing that book was - potentially I could now use more power because I have a more efficient CPU. And I could get myself a new and bigger TV that is much bigger and still fills the same power envelope. Except I'll probably have to take into consideration other things like cost, size, weight so maybe I won't.

On the whole though, I think energy efficiency works out as a good thing. Look at fridges, freezers, washing machines and so on. Sure, there's a certain effect that now that we have refrigeration, we'll use it a lot more and have bigger fridges. But in the end it's easier to just make them more efficient and hope the use tops out than to say "Please eat food that doesn't need cold storage and use your clothes until they're really dirty." I think it's same with computers, make them efficient until people say "Actually, this 5w computer is enough for me. Great that there are 100w+ CPUs for those that need them, but I'll rather take the other benefits."

Re:Rainbow Dash

[sexconker]

You're an idiot, Starscream. When we slip by their early warning systems in their own shuttle and destroy Autobot City, the Autobots will be vanquished forever.

You're an idiot if you think the "entropy" (the bits being 0/1) represents anywhere near the magnitude of the work done to flip those bits. Even if you DO think that, or there is some system where it's true, you can't use high bits as a heatsink (worksink) because the entire storage of the system (every single bit you care to count) only counts when being flipped. You can set all bits to 1 in a few clock cycles, and then everything else is 100% heat.

A high voltage (a bit that's on) is potential. There is no work done in keeping a bit on, just as there is no work done holding something above your head. It's the lifting that does the work. In this lifting, the "entropy" as you like to refer to it (the word is "potential", you fuckwit, please get a dictionary and learn basic physics) comes from the work, meaning the rest of the work is heat. You have reduced the amount of work that contributes to heat. But if these voltages drop down, guess what, they work against you.

We store energy in molten salt. This works because it takes forever to heat up the salt and it has a large heat capacity. By the time the heat capacity is full, the sun has gone down.
Flipping bits doesn't work this way because flipping bits is easy (bits have nearly 0 heat capacity) and because the entire system is only useful when it's powered (and heat is being pumped into it).

Re:Rainbow Dash

[Anonymous Coward]

I only need to comment on one part of that:

I think you're getting hung up on the equations.

Let's keep it simple:

You supply 100W of power to black box. If less than 100W of heat transfers out (minus entropy), you've just observed a violation of physics or made a huge discovery.

What you said was the power in and the heat out are identical. You can't just hand-wave away entropy changes because the processor is meant to change the entropy in the system. What I'm trying to say to you is that because of the change - call it work if you want - the heat the processor expels is less than the power you put into it.

Don't talk to me about violating the second law while simultaneously saying something that amounts to 'computation doesn't expend energy.' You'll note that I mentioned reversible computing in my original reply, also take note that I'm not talking about that here but it, too, expends energy because otherwise it would be violating the laws of thermodynamics.

I never mentioned Carnot efficiency but you should know that it's the ratio of the heat taken from the hot side to the heat deposited in the cold side - the difference between them is work. The "mechanism" is what lies between. More generally, efficiency is the ratio of useful work to energy input and what I'm trying to get across to you is that the work done by a processor (addition, branching, ANDing, whatever) results in a change of entropy in the system, which comes with the unit of Joules. Hence, you can obtain the thermodynamic efficiency of such a system from the amount of electrical power you put in and how much you reduced the entropy in the system. In comparison with the Carnot efficiency we're counting the entropy change as the useful work done by the system.

thought of another example:

if you're delivering energy to a CPU at some # of Watts, you have to remove it just as fast... and it doesn't just vanish. how is it removed? heat transfer. also watts.

You only have to remove the heat if you want the processor to remain at a constant temperature. Is this why you said the heat out must equal the power in? If so then in a way you are correct: practically this is the case because the difference between the two is very small, but don't say the two values are identical when the difference between them is non-zero.

Re:Rainbow Dash

[Technician]

From the announcement, it looks like a product split. Long battery life for portables such as tablets and smart phones without using a really slow chip, and extreme computing for rendering 3 D movies, server farms, and supercomputing. This too will take advantage of the instructions per watt. Your data center can be half the size, use half the cooling, half the power and still run faster. Or as you pointed out, you could have double the data center in the same old room.

In short, for the same processing the power used is much less. How much power you need will scale. Now it will scale in a smaller space with less power.

Re:Rainbow Dash

[Technician]

In performance per watt, the efficiency is measured in how much computing you can do per watt of power consumed. A computer sitting idle all day is wasting power generating heat. A computer rendering the latest Pixar film may draw more power and generate more heat, but it is also providing the results of the computation.
http://en.wikipedia.org/wiki/Performance_per_watt

Re:Rainbow Dash

[Anonymous Coward]

You're an idiot, Starscream. When we slip by their early warning systems in their own shuttle and destroy Autobot City, the Autobots will be vanquished forever.

Sure, whatever.

You're an idiot if you think the "entropy" (the bits being 0/1) represents anywhere near the magnitude of the work done to flip those bits.

I never said it did.

Even if you DO think that, or there is some system where it's true, you can't use high bits as a heatsink (worksink) because the entire storage of the system (every single bit you care to count) only counts when being flipped.

I fail to see how the storage capacity has anything to do with how often the bits change and I never suggested using 1 bits as a source of heat or anything else. The machine doesn't even need to be powered on for you to know how many bits you have to play with.

You can set all bits to 1 in a few clock cycles, and then everything else is 100% heat.

A high voltage (a bit that's on) is potential. There is no work done in keeping a bit on, just as there is no work done holding something above your head. It's the lifting that does the work. In this lifting, the "entropy" as you like to refer to it (the word is "potential", you fuckwit, please get a dictionary and learn basic physics) comes from the work, meaning the rest of the work is heat.

I'm quite aware of the difference between potential and entropy thank-you. Whether the bit is 0 or 1 is immaterial to information entropy, unless someone is around to correct me that actually understands information entropy and how it relates to the other kind. You apparently do not. You should learn some basic civility, BTW.

Still, if all the bits are 1 and there's no processing going on the processor wouldn't be using any energy at all, would it? I'll ignore for argument's sake the fact that memory often needs refreshing quite often. Charge leakage, and all that.

You have reduced the amount of work that contributes to heat. But if these voltages drop down, guess what, they work against you.

As I've been saying all along useful work is not a component waste heat, by definition - unless you're talking about a heater of course.

How we store the bits is immaterial. What matters is that the processor works with information and ultimately information and uncertainty about the same are a matter of entropy. Information entropy and thermodynamic entropy are related and a processor is essentially trading an increase in the latter for a decrease in the former.

We store energy in molten salt. This works because it takes forever to heat up the salt and it has a large heat capacity. By the time the heat capacity is full, the sun has gone down. Flipping bits doesn't work this way because flipping bits is easy (bits have nearly 0 heat capacity) and because the entire system is only useful when it's powered (and heat is being pumped into it).

Again, irrelevant. How bits are stored and changed is merely overhead on top of the energy needed to alter the information entropy in the system. Certainly the overhead is probably many, many orders larger than the useful work but that doesn't mean the latter doesn't exist at all. This is what Sebastopol implied originally.

Perhaps I've not made my position clear or perhaps you were just flaming... frankly after that start I've little interest in finding out.

tl;dr: While I'm waiting to post again I think I'll give a - probably faulty - example:

* You want to find out what 1 AND 1 equals and store the result somewhere; TTL or rod logic, it doesn't matter.

* You put some energy into your computer and the uncertainty about the value of the output bit and hence its entropy is reduced.

* This process is not 100% efficient, though, so you end up needing to put in more energy than the value of the entropy change. The di

Re:Rainbow Dash

[Anonymous Coward]

You can't just hand-wave away entropy changes

Sure you can. Because in this case, they're infinitesimal.

Re:Rainbow Dash

[Sebastopol]

exactly!

"the processor is meant to change the entropy in the system"?

I'm not sure what the parent was trying to claim with this indecipherable statement.

"you only have to remove heat if you want the processor to remain at a constant temperature"

which is ... ALL THE TIME!

floss

[callmebill]

fins... so, frickin lasers nearby?

Re:floss

[Anonymous Coward]

When you die are you going to be able to look back at your life and say "job well done?" What little I know about you (this post) points to "no".

Re:floss

[GungaDan]

When you die you are no longer able to look back at anything - you're dead. Other than that, nice trolling.

Re:floss

[Anonymous Coward]

So confident are we of the realm beyond? Or to put it your terminology, what if it's just a transition into another dimension of space/time?

Re:floss

[SMTB1963]

When you die are you going to be able to look back at your life and say "job well done?" What little I know about you (this post) points to "no".

You, on the other hand, will be able to look upon a lifetime of helpful and insightful posts on /. as ANONYMOUS COWARD. Let me be the first to say "job well done"!

Cooling

[tsa]

Is it possible to use that fin for cooling as well?

Re:Cooling

[Anonymous Coward]

No. The heat will warm the processor just like it does today and the processor will need to be cooled--just not as much due to the lower power consumption.

Re:Cooling

[Anonymous Coward]

Is it possible to use that fin for cooling as well?

Not likely, given that the fins look to be less than 100nm high - or about 20x smaller than infrared radiation...

Re:Cooling

[Sebastopol]

only at speeds over 25mph

Re:Cooling

[Noughmad]

How much is that in Kessel Runs per parsec?

Re:Cooling

[Anonymous Coward]

Silly... everyone knows there are ALWAYS 18 Kessel Runs per parsec.

It is - however 2.01216 x 10^-17 Kessel Runs per SECOND

Re:Cooling

[Anonymous Coward]

Surely you mean 18 parsecs per Kessel Run?

Re:Cooling

[splashbot]

It is greater than 0.08333333 three recurring.

Re:Cooling

[Victor Liu]

No. The fin is at one of the bottom-most layers of the chip. All the metalization layers for wires sit on top of the transistors.

37% faster!

[Ant P.]

You'll still have to cough up an extra $200 for the privilege of using all the transistors in your Intel hardware though. Or maybe this will bring an end to them segregating things like HW virtualisation based on how deep the users' pockets are.

Re:37% faster!

[the linux geek]

Oh, please. For consumer parts, the best price/performance by far is offered by Intel, specifically the i5-2500K. Intel doesn't price-gouge on consumer chips.

Re:37% faster!

[h4rr4r]

The i5-2500K is $230 for $15 less you can get a Phenom II X6 1100T Black Edition. Depending on what you are doing, say encoding video the Phenom is the way to go. Your best by far price/performance is just fanboy talk.

Re:37% faster!

[pablomme]

Are you aware of what is in a Core Solo processor?

Re:37% faster!

[smelch]

I bought one of those Phenom II X6 1100T Black Editions. Having six cores is pretty nice when you run a web server, file server and capture video and play starcraft II at the same time. I got blasted for not going intel, but I've had no problems at all with it and the chip was pretty cheap.

Re:37% faster!

[alen]

no, if you use a quicksync enabled app to encode video then Intel is faster than everyone

Re:37% faster!

[Jackazz]

What self-respecting slashdotter is buying a Core Solo processor?

Re:37% faster!

[Lunix Nutcase]

The i5-2500K is faster for the applications that most people are running which in many cases are single-threaded. Even when it comes to x264 encoding where the Phenom is faster, it's only by like a 10% margin at best. Not to mention that the Phenom is also more power hungry than the i5.

Re:37% faster!

[cforciea]

Sandy Bridge is the first time in a long time that Intel was remotely competitive with an AMD chip at the same price point. Many of us assume this bears some relation to the fact that early indicators put AMD's Bulldozer architecture on top of Intel's top end systems, so it is possible that we're going to see a role reversal because Intel needs to be more competitive in the mid-range market to keep in the game. This is, of course, all speculation, but it isn't hard to go back to pre-sandy bridge and see that the price/performance ratio on mainline consumer Intel chips prior to this year just plain sucked compared to AMD.

Re:37% faster!

[h4rr4r]

Which x264 does not support it so useless. For good reason as well, the encoder you are talking about is all in hardware and not very good.

Re:37% faster!

[h4rr4r]

Most people are not running only one application at a time.

The Phenom is more power hungry but the GPU anyone would buy for a setup like this is going to dwarf the small difference in TDP of the CPUs.

Re:37% faster!

[Anonymous Coward]

Living as a bitter malcontent is bad for you.

Re:37% faster!

[gman003]

Are you aware that AMD is even bigger on the whole "this processor has a defect in one core so we'll disable it and sell it cheaper" bit? They have dual-core processors that are just quad-core chips with half the thing disabled. Usually, those parts are disabled for valid reasons, but sometimes, if demand for the cheaper ones is too high (or they've got less defects than expected) they have to disable working parts. I know of several people who bought X3s and re-enabled the last core, giving them a quad-core processor at triple-core price. I also know that many of those people eventually reverted the chips back to triple-core, as the last core caused significant crashing.

Re:37% faster!

[PitaBred]

At stock speeds, sure. But my i5-2500K will clock at 4.4GHz at stock voltages, which puts it's encoding pretty well above the 1100T, overclocked or not. And any single-threaded tasks (of which most of what anyone does is) will be much better served by the 2500K.

I love AMD. I have two of their GPUs in my desktop here, my last build was a Phenom II 955. But right now, AMD doesn't even come close in the raw price, price/performance or performance/watt categories. I wish they did, and I really hope Bulldozer gets them competitive again. But as it stands now, AMD isn't a good choice for a completely new machine.

Re:37% faster!

[h4rr4r]

Most folks don't overclock, I only do that to my phone. Sure most applications are single process/thread but you normally run many at once.

I agree that at many price points Intel is untouchable right now, but at the i5-2500k that price is just barely in Intels favor.

I am waiting to build a new machine right now, if bulldozer flops it will be another Intel for me. My last machine was a prebuilt Q8300 that I got a steal on.

I wish I could use AMD GPUs but their linux driver sucks from what I have heard. Also from what I have seen in the past.

Re:37% faster!

[Anonymous Coward]

Oh come on, AMD doesn't hold a candle to Intel and hasn't for the past 5 years. My quad core i7 3.2ghz that I got a year ago for $299 is way faster than anything AMD had out at that time. Their GPUs are nice, however, they need better drivers!

Re:37% faster!

[Anonymous Coward]

The difference is in the details, The X6's and their associated chipsets are notorious for memory compatibility issues when using high speed ram, along with high power usage and high heat.

The Intel chip will overclock beyond your wildest dreams, and yet still puts out very little heat and uses far less wattage. That and you typically have to deal with far less compatibility issues when you go with intel chipsets.

An extra few bucks for less headaches sounds worth it to me.

Re:37% faster!

[RightSaidFred99]

Yeah, because the professionals use x264, right? The quick sync encoder is quite good, actually, but yeah you can choose between an imperceptible difference in quality or an additional 100X time to do the encoding.

Re:37% faster!

[spire3661]

Exactly. i5-2500k is a MONSTER of a chip for $200. The only reason you would go above this is for specialized media encoding. You dont NEED hyperthreading for anything other then specialized operations and time sensitive stuff. For that you should be expected to pay a premium. I would also like to add that the 1155 i7 is only $100 more.

Re:37% faster!

[lymond01]

I make sure to buy the Best Buy Monster cables to get the best video presentation possible.

Kidding. Purchasing a 50 foot HDMI Monster cable at 5% over cost was something like $16. I believe they charge over $100 retail.

Re:37% faster!

[PitaBred]

If you have a 2500K, you're gonna overclock. Otherwise you should have saved $25 and gotten the i5-2500, which is identical except for the unlocked multiplier. Same with the 1100T. Should have gotten a 1090T if you're not going to overclock. I'm still wondering how you say it's just barely in Intel's favor though, because the ONLY benchmark that the 1100T comes close to it on are POV Ray and some video encoding benches where it wins by a very small margin, and that's very seldom the only or even primary thing people do with their computers. All other benchmarks the 1100T gets trounced: http://www.anandtech.com/bench/Product/288?vs=203

Re:37% faster!

[Anonymous Coward]

No, the K processors are not identical except unlocked multiplier. K-models are also lacking some new instructions that Intel decided that only "business" users can have

Re:37% faster!

[Chuck Chunder]

Otherwise you should have saved $25 and gotten the i5-2500, which is identical except for the unlocked multiplier.

There are a number of differences between the K and non-K varients. The K variants have a better GPU which is why I bought one. I have no idea how much better because no-one seems to want to benchmark K vs non-K varients but as I want to drive high res monitors (not for games) I figured I may as well go for it.

Re:37% faster!

[Anonymous Coward]

Why purchase the i5-2500k if you are not overclocking?

The i5-2500 is exactly the same @ $11 cheaper; in fact it is better for the non-overclocker as it supports hardware IO virtualization:
http://ark.intel.com/Compare.aspx?ids=52210,52209,

Re:37% faster!

[petermgreen]

Most people are not running only one application at a time.

That rather depends on your definition of running.

Yes they may have a number of apps loaded into memory but the chances are most of those apps are sitting idle waiting for user input events with only one or two doing actual computing work at any given time. Further in my experiance most desktop apps only have one thread doing significant work. The result is for general desktop use four fast sandy bridge cores are going to be a far better choice than six slower phenom II cores.

Re:37% faster!

[aiht]

Are you aware that AMD is even bigger on the whole "this processor has a defect in one core so we'll disable it and sell it cheaper" bit? They have dual-core processors that are just quad-core chips with half the thing disabled. Usually, those parts are disabled for valid reasons, but sometimes, if demand for the cheaper ones is too high (or they've got less defects than expected) they have to disable working parts. I know of several people who bought X3s and re-enabled the last core, giving them a quad-core processor at triple-core price. I also know that many of those people eventually reverted the chips back to triple-core, as the last core caused significant crashing.

Then obviously those cores were disabled for a good reason. That hardly backs up your point about choosing to disable working parts.
However, your point is still valid. I am currently using an x2 550. The third core works fine, the fourth core is bad. I have been happily running with three cores enabled for over a year now.

Re:37% faster!

[jon3k]

if you want to cherry pick really low utilization use-cases how about we compare AES performance? the reality is for 99% of consumers the intel chip will absolutely blow AMDs offerings out of the water. dollar for dollar, watt for watt or any other metric you'd like to use.

Re:37% faster!

[jon3k]

You realize the 2500K is a quad core CPU right? And you understand the concept of preemptive multitasking right? So explain to me how you came to the conclusion that for most consumer use-cases that 6 slower cores are better than 4 faster cores?

I can see the shape of ads to come

[cygnwolf]

lots of blather about the 'next dimmension' of computer chips... 3d Inside or some such. Still, nice to see that moore's law still holds... at least for now...

Re:I can see the shape of ads to come

[SomePgmr]

Still, nice to see that moore's law still holds... at least for now...

Well I can't see into the future, but I have to say... it's encouraging that we've been saying that for a long, long time. :)

Re:I can see the shape of ads to come

[cartman]

The NYTimes article on this topic had a quotation from an intel researcher, who claimed they're almost certain they can reach 10nm, probably by 2015.

It appears that Moore's law has at least a few years left to it.

Re:I can see the shape of ads to come

[Thud457]

The generalized form of Moore's law describes society, not so much physics.

Re:I can see the shape of ads to come

[tsotha]

Assuming progress is incremental There are a lot of things in the pipeline that might mean orders of magnitude smaller design rules. One of them will work out eventually.

Re:I can see the shape of ads to come

[petermgreen]

Afaict the lattice constant of most semiconductors is about half a nanometer, other semiconductors are similar. That is 40 times smaller than 20 nm.

In other words if 20nm stuff comes out in 2012 and features sizes halve every 3-4 years (equivalent to doubling transistor counts every 1.5-3 years assuming all else remains equal) so then arround 2028-2033 we will be down to feature sizes that are comparable with the size of the repeating pattern of that crystal structure. My understanding is that as we approach that point we will get to a stage where we can no longer treat the crystal as being a smooth homogeneous material which will either require a radical rethink of how we design transistors, a move to stacking transistors in 3D (which will work for a while but unless we find a totally new way of making chips will mean an explosion of mask counts) or stopping progress

Re:I can see the shape of ads to come

[CtownNighrider]

Does anyone know off hand what the physical limit is where the uncertainty principle (if thats what it is) that prevents electrons from flowing throw a transistor?

And what will this do...

[Anonymous Coward]

to heat dissipation?

Re:And what will this do...

[teslafreak]

Not much. You're talking about stuff on the nano-scale. So small, it won't even look different to the naked eye. The same thermal compounds should work with about the same amount of efficiency.

Re:And what will this do...

[Coren22]

Especially since this will be inside of the package, and not exposed to the thermal compound.

Re:And what will this do...

[mr1911]

Yeah, considering you need several thousand of them grouped together to be perceptible to the naked eye.

Re:And what will this do...

[ChrisMaple]

By having the gate on both big sides of the active region (conventional MOS has the gate on only one big side) the gate control is about twice as effective. The active region can be turned off better, and turned on harder. This means lower power dissipation, other things being equal.

Other things aren't completely equal. The larger gate area means more capacitance, which lowers speed and raises dissipation. That is why they're talking about 37% faster and 50% lower dissipation, instead of (magically) 100% faster and 75% lower dissipation.

its 2D

[aepurniet]

they are not 3D, they are just thinner and deeper than the standard, we still dont see transistors on top of each other. the latice is still pretty much 2D. i ussually dont complain too much, but slash dot summaries are batting way below the mendoza line.

Re:its 2D

[Anonymous Coward]

You do realize this is the same site that tries to get you to purchase books from publishers like Packt, right? What's funny is that the last one was rated 3/10 yet the editor still pushed the Amazon Affiliate link trying to get people to buy it despite the fact that it's a garbage book. Slashdot hasn't had credibility in ages.

Re:its 2D

[bryan1945]

Even the actual article wasn't very good- "streetscapes" and "phone poles"? Yeesh. I find myself coming here less often in the past year.

Re:its 2D

[Anonymous Coward]

You are right that 3D is a bullshit term here; but this time slashdot is not to blame. Almost all the IT press is describing it in those terms. This is probably due to this nice headline from Intel's press release: "Intel Reinvents Transistors Using New 3-D Structure". Allow me to quote from ChipsZilla's website:
"The three-dimensional Tri-Gate transistors represent a fundamental departure from the two-dimensional planar transistor structure..."
It is just like Nostradamus predicted.

Re:its 2D

[denobug]

they are not 3D, they are just thinner and deeper than the standard, we still dont see transistors on top of each other. the latice is still pretty much 2D. i ussually dont complain too much, but slash dot summaries are batting way below the mendoza line.

No the structure is totally different. Look at how the source, drain, and the gate are arranged. Different geometry here.

The article from AnanTech has the best explanation of the technique I have come across so far today:

http://www.anandtech.com/show/4313/intel-announces-first-22nm-3d-trigate-transistors-shipping-in-2h-2011

Re:its 2D

[Coren22]

No one reads Slashdot for the article itself, people come to Slashdot for the conversation. If you are trying to say the article is bad, you are doing it wrong.

Re:its 2D

[emarsman]

Haven't transistors always been 3D? You may draw them 2D in layout, but it's still a gate sitting on top of a channel with stacks of metal... it has L x W x H. I think that's called 3D. Silly Intel marketing.

Re:its 2D

[SuperSlacker64]

And yet you're still here....

Re:its 2D

[Belial6]

No kidding, if they could make them in only 2 dimensions, now THAT would be impressive!

Re:its 2D

[smellsofbikes]

they are not 3D, they are just thinner and deeper than the standard, we still dont see transistors on top of each other. the latice is still pretty much 2D. i ussually dont complain too much, but slash dot summaries are batting way below the mendoza line.

No the structure is totally different. Look at how the source, drain, and the gate are arranged. Different geometry here.

The article from AnanTech has the best explanation of the technique I have come across so far today:
http://www.anandtech.com/show/4313/intel-announces-first-22nm-3d-trigate-transistors-shipping-in-2h-2011

Well... the geometry is still pretty similar. What they've done is raised the source/drain up and wrapped the gate around it, which means they have increased gate area, and can pretty much further increase it as much as they want, without changing the surface area required by the transistor.

However, it seems to me that this is going to increase the gate oxide leakage, which is a significant part of the power burned non-usefully by FETs, and also increase the gate capacitance, which means the gate drive current is going to increase. Gate drive current is already a huge part of the power required by (and burned by, and heat having to be removed from) a fet, and rises as the clock switching speed rises.

Am I wrong about this? Does this design evade these two problems somehow? Making a transistor smaller while increasing the power it uses and the heat it dissipates seems like going in almost exactly the wrong direction. Obviously they're making it work, so I'm wondering what I'm missing.

Re:its 2D

[Anonymous Coward]

WOW You are right! Better call Intel's 1-800 and set them straight! That pack of idiots will probably elect you king of the lab for saving their bacons!

Re:its 2D

[lee1]

Allow me to rise in defense of my summary. The press article that I summarized described the transistors as "3D", as did the people from Intel that they used as sources. Your quarrel, which is valid, is with them, not with the summary. Also, Intel's use of the term "3D" can be justified; even though the array of transistors might be planar, each transistor employs a 3D geometry in a nontrivial way. And, who or what is mendoza?

Re:its 2D

[Anonymous Coward]

they are not 3D, they are just thinner and deeper than the standard, we still dont see transistors on top of each other. the latice is still pretty much 2D. i ussually dont complain too much, but slash dot summaries are batting way below the mendoza line.

No the structure is totally different. Look at how the source, drain, and the gate are arranged. Different geometry here.

The article from AnanTech has the best explanation of the technique I have come across so far today:
http://www.anandtech.com/show/4313/intel-announces-first-22nm-3d-trigate-transistors-shipping-in-2h-2011

Well... the geometry is still pretty similar. What they've done is raised the source/drain up and wrapped the gate around it, which means they have increased gate area, and can pretty much further increase it as much as they want, without changing the surface area required by the transistor.

However, it seems to me that this is going to increase the gate oxide leakage, which is a significant part of the power burned non-usefully by FETs, and also increase the gate capacitance, which means the gate drive current is going to increase. Gate drive current is already a huge part of the power required by (and burned by, and heat having to be removed from) a fet, and rises as the clock switching speed rises.

Am I wrong about this? Does this design evade these two problems somehow? Making a transistor smaller while increasing the power it uses and the heat it dissipates seems like going in almost exactly the wrong direction. Obviously they're making it work, so I'm wondering what I'm missing.

You are correct in your assumption that the gate drive current will go up, but the real issue is that as the gate width decreases, the on current increases, but the standby current increases much faster. And at 22nm the gate is too small to accurately control the inversion layer, meaning there is not enough difference between the ON and OFF voltages to differentiate between a logic 1 and 0. So the solution is to increase the gate width while still being able to scale down the channel width. One way to achieve this is by raising the channel up so that there are essentially 3 contact points for the gate(s). This roughly doubles the gate width which, in turn, allows for more control over the transistor at lower voltages and smaller architectures. Another side-effect of the wider gate is that it allows for a greater flow of current which will cater to the performance crowd. Now back to your question; yes, it might raise the drive current some, but the power savings will really come from the lower voltage standby state.

Re:its 2D

[Anonymous Coward]

Mendoza was a baseball player who sucked at batting but was good at fielding. He was just bad enough (but good at fielding) that it didn't make sense to replace him with some kid who might suck in the field. Instead, he just muddled along doing poorly but not poorly enough to lose his job. His batting average (I think it was around .200) became known as the "Mendoza line"; the worst you can do without losing your job.

Re:its 2D

[Anonymous Coward]

You make it sound like actual working 2D transistors are not a big deal. I never even knew they existed ! I mean, even if they could somehow cut a transistor thinner and thinner until it had no width at all, how do they put it into a computer without them becoming just part of a 3D structure ?

Re:its 2D

[aepurniet]

your right i have seen this everywhere called 3D, my quarrel is with this label. rereading it, your summary does accurately summarize the referenced news item, but i would hope that the summary can get a little deeper into the issue and accurately summarize the event, rather than just the news item. or that may be our job...

Re:its 2D

[Anonymous Coward]

I guess at one time it was cute how inept the eds are. But then sites like digg and later reddit rose up where the users get to pick the articles and the headlines.

Less and less do i want to be "corralled" into stories by editors who are clearly morons. On top of that this comment system is clumsy and cumbersome. I love the flexibility to modify how posts are scored but that one bright spot is really hardly enough.

Yay!

[ItsJustAPseudonym]

Millions of free accelerometers! (For high-G's anyway.)

Get Perpendicular?

[Anonymous Coward]

For some reason when I read the summary (no, not TFA) I envisioned the "Get Perpendicular" video and now I have that damned song stuck in my head. Gee, thanks a lot, Slashdot.

Get perpendicular!

[sleschdott]

D'oh - they should have thought of that much earlier! Get perpendicular - soon also available for SSDs. http://www.youtube.com/watch?v=-xPvD0Z9kz8

this 3D thing is catchy

[Anonymous Coward]

so.... the glasses go between he socket and the processor?
if no glasses it performs like a regular 2D processor? .....sorry

Re:this 3D thing is catchy

[Anonymous Coward]

"3D is a fad; they've tried it several times before. It will never catch on permanently!"

NPR

[Anonymous Coward]

kudos

Mesa Transistors

[Anonymous Coward]

"a small pillar, or fin, of silicon that sticks up above the surface of the chip"

Like a Mesa transistor? http://en.wikipedia.org/wiki/Mesa_transistor#Mesa_transistor
Transistors USED to be fabricated this way, before the founders of Intel developed the Planar process.

Re:Mesa Transistors

[ChrisMaple]

The planar process was developed at Fairchild by Jean Hoerni. (wikipedia).

3D Transistors. Seriously?

[doomsday_device]

What is indeed news is that intel is fielding them first.

Well, while it is nice a slashdot article has finally been written about FinFET's - there may already have been one, I just can't remember - these devices have been widely guessed to be a part of the 22 nm technology node for quite some time. (see: http://www.itrs.net/ and http://en.wikipedia.org/wiki/22_nanometer ).

They offer more effectivity for your gates as the field is not coming from one, but from 3 sides to the channel. That means a bit more scalability, but not much more. There is only a bit of improvement possible for the future in putting the gate below the channel as well (as hard as that may be, i, personally, don't think it would be worthwhile), so this won't save moore's law in the end.

It may not surprise you that they actually haven't been invented by intel, and are not new.

The term has been coined more than 10 years ago ( http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=823848 ) (find one of the free pdf's of this classic paper for yourself)

What is more interesting is how far down these transistors will scale in the extreme ultraviolet processes that are emerging right now.

Re:3D Transistors. Seriously?

[RightSaidFred99]

These aren't (common) FinFETs. They're trigate, and a little different than conventionally defined FinFET. See here.

Re:3D Transistors. Seriously?

[doomsday_device]

Well, the original FinFETs were twingate (as were some similar transistors before the term FinFET was coined for them), intel invented trigate. In the end, this still is the first kind of FinFET coming to market.

I just think inventing them in the first place is more inventive then refining them.

My point of them having been quite a long time in the making still stands. Just note the age of that press release (2002).

Oh dear. They called them 3D-Transistors even back then.

Re:3D Transistors. Seriously?

[RightSaidFred99]

I agree with your tautology that inventing them is more inventive ;)

But what's actually meaningful is deriving benefits from something. You can go off and invent all kinds of crazy shit in a lab, but if nobody ever does it it's meaningless.

Re:3D Transistors. Seriously?

[doomsday_device]

Yes. You may note, that was my original point. Intel fielding them before AMD and the rest does is what the real news is about here.

Nevermind the tautology. I'm tired and drunk. Sorry. ;-)

But a link to the inventors would have been ... well ... informative.

And the kind of FinFET technology used is more a matter of what is more effective in your semiconductor process. Whether it's twingsate, trigate, quadgate, whatever. There's a good chance what they published years ago is not all that similar to what they did for this.

Next up - HD Transistors

[sagaciousb]

What other trending marketing buzzwords can we apply to unrelated technology to get attention? Social Transistors? Transistors 2.0? Cloud Transistors? Tablet Transistors? iTransistor. What would truly be remarkable would be to develop a 2-dimensional transistor. Infinite layers of them with no height added to the die - and it'd still be slim enough to fit in a tablet!

Re:Next up - HD Transistors

[Belial6]

Even better, get it down to 1 dimension.

Re:Next up - HD Transistors

[sagaciousb]

That's what I'm talking about. Singularity Transistor.

Re:Next up - HD Transistors

[bill_mcgonigle]

Even better, get it down to 1 dimension.

You store your whole transistor in this universe? Amateur.

Twisted Transistor

[jeffeb3]

What's next? A Twisted Transistor?

In a "finFET", the gate surrounds the channel

[slew]

Haven't transistors always been 3D? You may draw them 2D in layout, but it's still a gate sitting on top of a channel with stacks of metal... it has L x W x H. I think that's called 3D. Silly Intel marketing.

Actually integrated transistors have been "planar" for the most part (although there have been "vfets" and other types of 3d channels in the past)...

In planar transistors, the field that chokes off the source/drain path has been mediated by a gate which is just on top of the channel on one plane. Imagine an iron on of a ironing board heating it up the board when you turn it on. Although the ironing board and the iron are both 3-dimensional objects, the interface in a "plane" and the heat diffuses across this plane. In this analogy with a planar transistor, the channel is dug into in the ironing board and the iron is the gate.

In finFET, the gate surrounds the channel on 3 side. Imagine now a tube on an ironing board and the iron has a notch cut in it so the iron surrounds the tube on three side. When you turn on the iron, heat diffuses across all around the notch instead of interfacing on 1 plane. This is "3D" or finFET instead of planar. In this analogy with a finFET transistor, the channel is the tube on the ironing board and the iron with a notch is the gate.

As you might imagine, the finFET architecture should have a better capability to turn on and off the channel since there is field is wrapping around the channel instead of just being applied to one side (okay that's simplification, but you get the idea).

Intel took, this finFET idea and added another twist with a "3", called tri-gate (or tri-channel depending on your point of view). This congolomeration of two independent ideas that both revolve around the number 3 is the kind of thing that drives marketing people to be silly ;^)

And for those that don't understand w/o a car analogy, imagine the difference in traction you get with bald tires on ice (planar where tire is the gate and ice is the channel), vs snow tires on dirt (finFET where the tire is the gate and dirt is the channel)...

Re:In a "finFET", the gate surrounds the channel

[blahplusplus]

Car analogy FTW! :D

3D?

[slapout]

3D? Aren't all transistors 3d? Along with every other physical item?

Re:3D?

[mark-t]

Transistors are only 3d in the same sense that a drawing on a piece of paper could be called 3D owing to the thickness of the paper it is drawn upon.

The silicon upon which the transistor exists is, of course, 3d... but the transistor itself is simply etched into its surface and treated with chemicals to produce the desired effect.

Re:3D?

[Darinbob]

But multiple layers requires 3D. It's more than just a drawing because some bits are on top of other bits.

Re:3D?

[tyrione]

3D? Aren't all transistors 3d? Along with every other physical item?

The point of 3D is to have 3 distinct dimension and in this case, 3 distinct dimensions where electrical current is controlled.

Re:3D?

[mark-t]

In which case you are talking about something with more than one transistor.

Re:3D?

[Darinbob]

No, you can't have a wire that has width cross another wire that has width and still be 2D.

Re:3D?

[strack]

wow dude. you managed to use the strict definition of something to point out a trivial fallacy in the article, then use that to make a patronising and trivial correction to make yourself seem smart. ooo. you used the word 'distinct' twice. in the same sentence. it must be the only 'fancy' word you know. i got a idea, unless you have something real to add to the conversation, shut the fuck up.

Re:3D?

[Anonymous Coward]

The components are designed in 2D. It doesn't matter if the physical device takes up 3 dimensions, the actual operation of the device is essentially 2 dimensional.

Re:3D?

[TopSpin]

It's 3D in the sense that it has more depth than the previous version. The original Castle Wolfenstien (1981) had very little depth relative to later incarnations, such as Wolfenstein 3D. The 3D aspect of the new Intel transistors improve power consumption by reducing current leakage. The new transistor may therefore have much smaller feature size. Likewise, Wolfenstien 3D improves game play by reducing the players perspective of the map and increasing anticipation.

You will notice that both the original Castle Wolfenstien and Wolfenstien 3D are actually rendered on a 2D planar surface, despite the use of the '3D' moniker in one of the titles. Intel is guilty of the same thing, but in reverse; so called 'planar' transistors are etched onto a flat surface, but they actually have depth, just like Wolfenstien doesn't, whereas part of the new transistors stick up from the surface and must, therefore, be characterized in a third dimension.

BadAnalogyGuy use to handle this sort of thing...

another car analogy

[Anonymous Coward]

Perhaps a better car analogy would be a planar transistor is a two-cylinder single-port injector engine and the intel 3d-tri-gate is a six-cylinder multiport injector engine...

Re:another car analogy

[drcheap]

Perhaps a better car analogy would be a planar transistor is a two-cylinder single-port injector engine and the intel 3d-tri-gate is a six-cylinder multiport injector engine...

No, the snow tires + dirt was better. You can return to the basement now.

Disadvantage

[ChrisMaple]

Although this provides a superior transistor, there is a drawback from the IC designer's standpoint, if the illustrations are accurate. The designer of an IC with planar FETs can control both the width and the length of the active region. With these FinFETs, only the length is under the designer's control. Width is fixed (it's the vertical dimension in the illustrations). The only way to get stronger drive is to parallel transistors, whereas in planar design the transistor is simply made wider.

Re:Disadvantage

[drhank1980]

This is a real disadvantage from an IC designer standpoint, I would expect that this is not as much of restriction as they are already contained by in the extreme limits imposed on the design by double (possibly triple in this case) patterning to get down to 22nm.

I did see a TSMC presentation earlier this spring at SPIE showing scaling down to the 1xnm node where they actually had little "pipes" of channel where the gate wrapped all the way around (Think of the gate line in a FINFET going through the "fins", in this the FINs became the gates). In this kind of process, designer would be allowed to modify W again on the transistor by making the pipes larger or having two or three of them as part of your transistor with minimal interconnect.

Makes sense that Apple is moving to Intel

[MojoRilla]

It now makes much more sense that Apple is having Intel manufacture the next generation IOS chips. Based on information from Ars, the finFET shines at lower voltages. So Apple is going to be using Intel not just because they can manufacture smaller than everyone else, but also because the resulting chips will be faster or take less power than anyone else.

Hunh?

[vegiVamp]

Faster *and* lower-powered, from *Intel* ?

Truly, the end is nigh.

Re:Hunh?

[The End Of Days]

Where have you been this decade?

And I'll decide when the end is nigh, thankyouverymuch.

isn't this just a mulitgate finFet

[viking80]

isn't this just a mulitgate finFet. That structure has been around for a long time.

Re:no its not 2D

[thehodapp]

Newsflash: AMD designs even faster, 4-D Transistor!

This is baby steps

[Anonymous Coward]

The technology for wafer-scale, multi-layer stacking already exists. The interesting thing is that he's also discovered that most of the heat and power requirements for chips are for external drivers, so you have all these border drivers on the CPU and Northbridge generating heat, and all the border drivers on the RAM generating heat. Stack them together, and you get rid of both, not to mention reducing the circuit path to microns. How would you like to see memory densities go up eightfold, while power consumption is cut by 90%? You'd think companies would be jumping all over that.

I have done work for the patent holder, and he's been unable to get any American companies interested until he commercializes it and makes it a big success, so they can grab his coat-tails. That includes Intel. http://www.tezzaron.com

Wake me up when this is less that five years out..

[fotoguzzi]

Are there articles from nine, four, three, two, and one years ago predicting this breakthrough? To someone not in the business, this seems to have come out of the blue--already in production by the time the news breaks. How long have they been working on this?

Re:Wake me up when this is less that five years ou

[fotoguzzi]

Answering myself: The idea started in 2002.

Re:Wake me up when this is less that five years ou

[fotoguzzi]

Intel's 3D Transistors One Step Closer to Reality Posted by ScuttleMonkey on 15:15 Monday 12 June 2006 from the closing-your-leaks dept. http://slashdot.org/story/06/06/12/224207/Intels-3D-Transistors-One-Step-Closer-to-Reality

3D

[Anonymous Coward]

Do you need special glasses to see them?