Intel overclocking prevention, how secure?

[Belegost]

http://patft.uspto.gov/netacgi/nph-...,535,988.WKU.&OS=PN/6,535,988&RS=PN/6,535,988

Intel has applied for patent on a method for detecting an overclocked bus signal, and then preventing proper computer operation when detected.

The application for patent lists two types of circuits to be used. Both involve the use of reference clock signal, a counter to count the signal from the bus clock with respect to the reference clock, and a comparator to compare that count to predetermined values.

Two methods are listed for generating this reference clock; a crystal oscillator, or a ring oscillator. The diagrams included with the patent application show the circuit being contained on the chipset, which leads to my question for the EEs in the crowd:

Is it possible to full implement such a circuit on-die? Can the crystal oscillator, or the resistor-capacitor tank for the ring oscillator be put on the chipset die? Or will it be necessary for these components to be somewhat discrete, perhaps mounted on the chipset, but not on-die, or a discrete component on the motherboard?

As an overclocker, the difference is large. If the components are discrete, I could replace them and alter the reference frequency, especially easy in the case of a ring oscillator where I put in a pot for a resistor, and then o-scope the output til I get the freq I want. Otherwise, if intel can mount these completely on-die, then overclocking an intel processor becomes nearly impossible, since they certainly will force the other chipset manufacturers to include this technology to receive licensing rights.

 

[zsouthboy]

D'oh sorry

 

[SuperPickle]

Linkage

 

[AbsolutDealage]

There are ways of creating an oscillator on-die, however, my speculation is that it will be en external crystal used for the comparison. As it stands currently, an integrated oscillator is a very difficult thing to "tune" to a specific frequency. Also, this value would need to be set after the fabrication process, because of the inherant nature of the fabrication process (and the resultant speed ratings of individual dies).

To do this, they would need some kind of way to connect/disconnect special pins on the die. Also, they would be sacrificing a large amount of area on the chip and a highly complex design of fusing in order to have a wide range of possible frequencies. This would be not only a design nightmare, but it would be a waste of precious die area and a waste of design time.

This problem would be infinately easier with an external crystal for the oscillator. They could simply test the dies and then place the corresponding crystal on the packaging for the die. This would essentially eliminate a large amount of the "casual" overclocking that is happening today. In order to overclock the chip, a person would need experience with soldering surface-mount components, along with the equipment and the time to do so. Not many people have a signal generator and an oscilloscope lying around their house, nor would they know what to do if they had them. Also, at some point, you would need to then physically remount a crystal onto the package.. which is more soldering not to mention having the specific crystals that are needed just "laying around".

Personally, I don't think that Intel will attempt to completely get rid of overclocking... they know that if they did this, AMD would take a large market share from them, and they are too smart to do that.

 

[damonpip]

based on comparison of ratio of the system (processor) clock signal which is likely to be over-clocked and a fixed, stable reference clock signal which is highly unlikely to be over-clocked

A bit of hope, if it was on-die it would be IMPOSSIBLE to overclock the reference signal

 

[SuperTool]

Well, a reference clock would be temperature, process, and voltage dependent if it was generated by a ring oscillator.
So it would most likely be generated offchip either inside the package or on the motherboard with a crystal oscillator.
Synchronizing the reference clock with the cpu clock and the bus clock would require a synchronizer, which is another thing that could go wrong.
So whatever savings they gain from stopping overclocking will probably be offset by the reduced yields due to higher complexity.

 

[pm]

Nice answer, SuperTool.

That has been my observation of such things in the past. I know nothing of this scheme from working at Intel, but I did notice the diagram on the patent relies on an external reference crystal. They have the diagram that I saw over at the Inquirer.

 

[rommelrommel]

Originally posted by: AbsolutDealage


Personally, I don't think that Intel will attempt to completely get rid of overclocking... they know that if they did this, AMD would take a large market share from them, and they are too smart to do that.

Overclockers are hardly a large portion of the CPU market. We're actually a very tiny portion of the CPU market.

 

[Barnaby W. Füi]

Originally posted by: rommelrommel

Originally posted by: AbsolutDealage


Personally, I don't think that Intel will attempt to completely get rid of overclocking... they know that if they did this, AMD would take a large market share from them, and they are too smart to do that.

Overclockers are hardly a large portion of the CPU market. We're actually a very tiny portion of the CPU market.

But a portion nonetheless, and by going through with this, Intel would have to be confident that they were somehow helping revenues by doing this, and I don't see how that would be so.

 

[Shalmanese]

I would have thought that it would be relatively trivial for them to implement but also a relatively trivial problem which they really dont care about.

 

[AbsolutDealage]

Overclockers are hardly a large portion of the CPU market. We're actually a very tiny portion of the CPU market.

We're not that tiny... AMD has acknowledged that a significant portion of their customer base consists of "enthusiasts".

 

[sao123]

Just another example of corporate greed...

Intel would have to be confident that they were somehow helping revenues by doing this, and I don't see how that would be so.

Fact: All chips of a generation come from the same die. And therefore cost the same to make.
Fact: CPU's are priced by the clock signal generator, which generates a different rate of [insert # nere] Mhz, changed by flipping a switch or something.

By preventing overclocking...Intel makes you pay for the extra Mhz.
IE... you cant buy overclock and buy the cheap CPU and overclock it, you have to pay the extra $[insert amount here].


For a chip it costs just pennies to make, look at the proce differences...
Both moade from same die, same chip, just enable more features hyperthreading, or enable a clock rate increase.
compare prices...
P4 1.7 Ghz W/512K CACHE 400FSB 478-PIN RETAILED - $132
P4 2.6 Ghz W/512K CACHE 400FSB 478-PIN RETAILED - $254

P4 2.4B Ghz W/512K CACHE 533MHZ 478-PIN RETAILED - $170
P4 3.06 GHZ W/512K CACHE 533MHZ 478-PIN RETAILED - $581

 

[pm]

Fact: All chips of a generation come from the same die. And therefore cost the same to make.
Fact: CPU's are priced by the clock signal generator, which generates a different rate of [insert # nere] Mhz, changed by flipping a switch or something.

These facts are not exactly factual.

Microprocessors from a given design are all made from the same mask set, which is a form of photographic negative that is used to produce the electrical circuit patterns on the silicon wafer. Unfortunately at the dimensions that chips are manufactured at, there is some variance in the output across a chip, a wafer, and across multiple wafers. Although this variance is small, it affects the operations of the transistors in several key ways (speed, noise, power leakage, etc.). So across a given wafer which contains, let's say, 200 microprocessors there will be a variation of transistor properties in different chips. Because the clock frequency that a given chip will clock at without errors is determined by how fast the transistors are in key circuits on the chip, what this means is that some chips will clock faster than others.

If you had a lot of microprocessors at your disposal and you graphed them with maximum clockable speed without errors in the X-axis and the number of parts at a given speed in the Y-axis, what you would see is what is known mathematically as a "normal" (or Gaussian) statistical distribution. You'd find that you have a very few ultra-fast parts, and some really slow parts and most are clustered in the center.

At this point a manufacturer could do one of two things, they could draw a cut off line at a given frequency, toss out the slow ones, and sell the rest at a specific frequency - say the middle to lower end of the speed range. Or they can do what every high-volume semiconductor manufacturer does, and has done since the 70's, and bin them into frequency catagories and sell them at a sliding pricing schedule with the highest performance at the top and the lower performing at a lower price.

What you, sao123, seem to be suggesting in your post is that regardless of speed, all parts should be sold at the same price. In this alternate world, whenever Newegg gets a new shipment of parts in, they'd receive a few fast ones, a lot of medium range ones and a few slow ones. Would it be first come first serve for a 3.06GHz part? If you aren't fast enough ordering, you'd end up with a slow one. Or, more realistically, the distributors would start doing what manufacturers are doing now and would start a sliding pricing scale. In the past they have been known to do this with some models that were packaged at a given location. (for example, Costa Rica Pentium 4 $200. Malaysia Pentium 4 $210). So either the manufacturers will charge more, or the capitalistic market will do it for them.

I don't have a lot of references that explain what I posted above in readable detail that aren't expensive college textbooks - although I can post a few if you have access to a decent college library. Here is one that talks about the subject.

For a chip it costs just pennies to make, look at the proce differences...

Pennies? A fab costs about US$3 billion to build currently. Then there are the salaries of the workers, the cost of materials (a wafer alone is not exactly cheap), packaging costs, testing costs, and yield fall-out from the parts that don't work. Even leaving aside the costs of the building, I can't see how any large modern microprocessor could be manufactured for pennies or even for under a couple of dollars. And we haven't even discussed the bit that's important to me: my salary. A modern design takes several hundred engineers several years to design. That cost needs to be factored in as well.


Patrick Mahoney
Microprocessor Design Engineer
Intel Corp.

 

[Agent004]

Originally posted by: AbsolutDealage

Overclockers are hardly a large portion of the CPU market. We're actually a very tiny portion of the CPU market.

We're not that tiny... AMD has acknowledged that a significant portion of their customer base consists of "enthusiasts".

In fact, AMD has make this rather public in early life of athlon, ' we will not lock multipliers '. What's the implications of this?

Average joe knows nothing multipliers, motherboard manufacturer knows and so are "enthusiasts" knows what it can do.

It's part if AMD's strategic (marketing) plan to appeal to "enthusiasts" and market athlon through words of mouth to reach for "non-enthusiasts".

sao123

It's not all about corporate greed. If you don't need so much computing power, why pay for the high end (in you case, high end price = low end price) when low end will be suffice for your usage? Why pay for the unnecessary features which are enabled for no (noticeable improvement for the individual's usage) apparant(sp?) reasons?

Creating a price differential signals the different levels of needs and usage. Since the ancient times( don't belive me? Cave men actually have a very similar monetary system), premiums are always attached the latest and the greatest. Today's new will not worth the same in tomorrow and a price drop on the (old) range reflects the drop in value. The fact profit maximisation is just so happens to be one of the positive outcome of this doesn't change its purpose.

 

[wviperw]

It seems as if a few people think Intel is doing this because of the enthusiasts we commonly call "overclockers." However, this is not the case. Its the resellers who practice bad business by selling oc'ed processors which Intel will potentially be cracking down on. Though, I'm sure wiping out the enthusiasts' ability to overclock would be a small side benefit for Intel.

 

[sao123]

Hey, my opinion (however unrelated this might be) might be biased on the fact that I think there is an ideal society with the advantages of less capitalism and more communism. Otherwise known as a moneyless society. One that is not driven by the amassing of possessions and power.

Nothing personal against you but you still said...

we haven't even discussed the bit that's important to me: my salary.

So I can make the assumption, that corporate greed is probably based on or the result of collective individual greed.
And I am not accusing you of being greedy at all. Just the current state of our americal civilization.

Intel isnt alone in this matter... so is (microsoft) and the record & movie industries, and anyone who uses all this new antipiracy stuff to squeeze every last penny out of every last product.

 

[wviperw]

Originally posted by: sao123
Hey, my opinion (however unrelated this might be) might be biased on the fact that I think there is an ideal society with the advantages of less capitalism and more communism. Otherwise known as a moneyless society. One that is not driven by the amassing of possessions and power.

Nothing personal against you but you still said...

we haven't even discussed the bit that's important to me: my salary.

So I can make the assumption, that corporate greed is probably based on or the result of collective individual greed.
And I am not accusing you of being greedy at all. Just the current state of our americal civilization.

Intel isnt alone in this matter... so is (microsoft) and the record & movie industries, and anyone who uses all this new antipiracy stuff to squeeze every last penny out of every last product.

There's a difference between greed and wanting the salary that you know you deserve. And talking about Intel I wouldn't say this move is pure greed. Oh sure, there is probably a little of it in there but I would think there is other reasons as well.

 

[pm]

soa123, interesting that you completely dodged the entire point of my post which was to correct two points that you stated as fact - by noting that bin split is a statistical function and that pricing for top bins is based on typical supply and demand, and to comment that your estimate of how much a chip costs to make is off by several orders of magnitude - and instead focused entirely on something that I added as a bit of humor.

 

[CTho9305]

Originally posted by: pm
soa123, interesting that you completely dodged the entire point of my post which was to correct two points that you stated as fact - by noting that bin split is a statistical function and that pricing for top bins is based on typical supply and demand, and to comment that your estimate of how much a chip costs to make is off by several orders of magnitude - and instead focused entirely on something that I added as a bit of humor.

Just something to note, about WHY Intel might lock multipliers. I highly doubt any of the chip companies care about 0.5% of the population overclocking their processors. For that matter, when we fry chips, we post on Anandtech about how far it went before melting (free advertising) and then go buy another one.

However, what they WOULD care about is less-than-reputable vendors buying slower chips, bumping the multiplier up a few notches, then selling the processors as if they were really the faster speed. The chip is unstable, the customer says, "Intel sucks, their chips aren't stable", and switches to AMD. With a locked multiplier, Joe User is basically guaranteed that the chip he paid $600 for which claims to be 3ghz is in fact rated 3ghz and will be stable, since it was properly tested by Intel, whose tests are much more thorough than a little prime95.

If you look around you might be able to find some info about early P2 remarking - IIRC there was something in the cartridge that remarkers would change, then close it back up and sell it at a higher price to an unsuspecting customer.

 

[AbsolutDealage]

Originally posted by: sao123
Hey, my opinion (however unrelated this might be) might be biased on the fact that I think there is an ideal society with the advantages of less capitalism and more communism. Otherwise known as a moneyless society. One that is not driven by the amassing of possessions and power.

Ugh, it would be altogether too OT and too irritating to explain the impossibility of your "ideal" society. Another post, another forum maybe.

 

[Thraxen]

However, what they WOULD care about is less-than-reputable vendors buying slower chips, bumping the multiplier up a few notches, then selling the processors as if they were really the faster speed. The chip is unstable, the customer says, "Intel sucks, their chips aren't stable", and switches to AMD.

I guess the question at that point become how much of a problem is this? Is the number of people getting falsely labeled CPUs larger than the potential loss of the enthusiast market?