What if OC'ing was merely disinformation?

[Zenoth]

That might have been discussed here before but I'll take the risk at mentioning it perhaps again.

I had a discussion today with a friend, just talking about random things until we discussed about gaming and it eventually led to over-clocking. Well at one point he said something without necessarily thinking too much about it himself, he wasn't building up that idea since years or anything like that, you know... he just threw that idea out of nowhere and at first we kind of laughed about it. Then later I messed around with my system's over-clock settings and what he told me came back in my mind: what if over-clocking was just an intended mislead about what a company's CPU(s) "can do", building up its quality value and ensuring good performance at the stock exchanges/market shares?

For instance, the company's engineers from the start plan and then conceive a chip that in the first place was already technically capable of and set to a certain speed (not limit) which was initially used as a stability reference and "good-to-go" status for the market. But then the company deliberately under-clocks the chip by say... (just randomly chosen number here) 1Ghz, and then they ship that chip series on the market at that, for us consumers, "stock" speed. We then naive (perhaps) consumers have fun over-clocking those chips and then we all come in discussion forums here and we some times even keep getting surprised by supposedly and indeed seemingly "how good" can that one chip in particular or the whole family of it can over-clock.

When I look at some chips since a few years mostly and see all those 1Ghz and some times 2Ghz+ over-clocks being achieved as it seems more easily done with each new generations I now wonder, after having that discussion with my friend... would it be possible? Could such a disgraceful potentially real disinformation be the cause of the reasons why some chips since at least a few years seem to over-clock oh-so-damn-well? That in fact we were not over-clocking but were merely just putting our chips back at their exact (or very near) unknown (to us) actual stock speed? I know it might be far-fetched, well... yes it certainly is. But just out of craziness-thinking, food for thought stuff... what do you guys think of the possibility? And certainly of the implications of such a thing.

 

[Billb2]

99.9% of computer buyers don't even know what over clocking is, much less letting the overclockability of a CPU effect their buying choices.

 

[Idontcare]

I think you are mixing up the words, its not disinformation in any sense that I know of the use of the word disinformation, but yes absolutely what you describe does happen.

Particularly with the entire budget line of processors. Celerons and Semprons.

CPU manufacturers are intentionally underclocking Celerons and Semprons and in some cases have their cache subsystem crippled so as to ensure the consumer only gets what they pay for performance-wise.

This typically does not happen across an entire product line, the top-end SKU is usually also the top-end parts that can be validated to perform at that speed while still maintaining a standard of product lifetime with engineering margin (to be used up by us overclockers in exchange for higher clocks for the duration of its shortened lifetime).

GPU parts are no different. Usually the low end discreet graphics cards are intentionally handicapped parts from the high-end.

This is less likely to be true when a new technology node is released (and yields are low) or when a new chip architecture is released (and yields are low)...and is more likely to be occurring the older and more mature the technology node becomes and/or architecture.

 

[Zenoth]

Originally posted by: Billb2
99.9% of computer buyers don't even know what over clocking is, much less letting the overclockability of a CPU effect their buying choices.

Ok and you'll have to tell me where you've got that number from? I know a lot of people who know what over-clocking is, they just don't know how to do it. If they knew, they'd do it (well at least that's what they say). So unless you've just made that number up to give yourself some placebo credibility you'll be able to provide me with the source of that information would you?

 

[Zenoth]

Originally posted by: Idontcare
I think you are mixing up the words, its not disinformation in any sense that I know of the use of the word disinformation, but yes absolutely what you describe does happen.

Particularly with the entire budget line of processors. Celerons and Semprons.

CPU manufacturers are intentionally underclocking Celerons and Semprons and in some cases have their cache subsystem crippled so as to ensure the consumer only gets what they pay for performance-wise.

This typically does not happen across an entire product line, the top-end SKU is usually also the top-end parts that can be validated to perform at that speed while still maintaining a standard of product lifetime with engineering margin (to be used up by us overclockers in exchange for higher clocks for the duration of its shortened lifetime).

GPU parts are no different. Usually the low end discreet graphics cards are intentionally handicapped parts from the high-end.

This is less likely to be true when a new technology node is released (and yields are low) or when a new chip architecture is released (and yields are low)...and is more likely to be occurring the older and more mature the technology node becomes and/or architecture.

Ah, well that's nice to learn about

And, well I thought I'd have to use the word disinformation, but I'm not English, I'm not always certain about the words I need to use in certain contexts. Would it have to be "misinformation"? Because I thought that disinforming someone or a mass was to intentionally mislead people into believing a false truth. In this context a company would mislead the consumers into believing that over-clocking does indeed go beyond what the chip was designed for, when in real fact all OC'ing does is to go from an under-clocked state labeled as "stock speed" to the real stock speed that the engineers secretly tested before releasing the chip.

 

[Idontcare]

Originally posted by: Zenoth
Would it have to be "misinformation"? Because I thought that disinforming someone or a mass was to intentionally mislead people into believing a false truth. In this context a company would mislead the consumers into believing that over-clocking does indeed go beyond what the chip was designed for, when in real fact all OC'ing does is to go from an under-clocked state labeled as "stock speed" to the real stock speed that the engineers secretly tested before releasing the chip.

Misinformation would be more appropriate. In the US we use the term "sandbagging" which means to intentionally under-sell or under-report your capabilities.

For instance lets say you know you can run 7km in 45 minutes, but there is an organized race coming up which is 7km long so you tell everyone you can run 7km in 60 minutes. (which is true, 45min is less than 60min)

Then all your competitors get comfortable with the expectation of you only running the race in 60min, so maybe they train less aggressively because they just want to beat you and run 7km in 55min.

But then you show up and you run 7km in 50min (which you knew you could do all along) beating the competition by 5mins (they ran in 55min) but also still not exhausting yourself because you didn't run your absolute fastest (45min) as it wasn't needed.

That is called sandbagging your capabilities. Chip makers do it all the time.

 

[Lonyo]

Originally posted by: Idontcare

Originally posted by: Zenoth
Would it have to be "misinformation"? Because I thought that disinforming someone or a mass was to intentionally mislead people into believing a false truth. In this context a company would mislead the consumers into believing that over-clocking does indeed go beyond what the chip was designed for, when in real fact all OC'ing does is to go from an under-clocked state labeled as "stock speed" to the real stock speed that the engineers secretly tested before releasing the chip.

Misinformation would be more appropriate. In the US we use the term "sandbagging" which means to intentionally under-sell or under-report your capabilities.

For instance lets say you know you can run 7km in 45 minutes, but there is an organized race coming up which is 7km long so you tell everyone you can run 7km in 60 minutes. (which is true, 45min is less than 60min)

Then all your competitors get comfortable with the expectation of you only running the race in 60min, so maybe they train less aggressively because they just want to beat you and run 7km in 55min.

But then you show up and you run 7km in 50min (which you knew you could do all along) beating the competition by 5mins (they ran in 55min) but also still not exhausting yourself because you didn't run your absolute fastest (45min) as it wasn't needed.

That is called sandbagging your capabilities. Chip makers do it all the time.

Apart from those times being dire, it's also easier to have a lower target because it ensures you can keep yields very high, and then as your process matures, you can increase speeds without really risking yield.
It gives you headroom to release new processors in the future without too much work (partly sandbagging, but also you're still training, so your performance is getting better anyway, and you don't want to max yourself out really).

Also chipmakers want to make money. If they released a full range from 2GHz to 3.8GHz, either they would lose money at the lower end, crowd the market with too many different offerings, or end up having the top end processors priced at really really high levels so that they could differentiate all their products.

 

[rchiu]

You have to understand CPU business to see how the whole thing works. First of all, engineers cannot design a chip with a particular speed, let's say 3.0ghz. A lot of it is based on the manufacturing and how good the yield is. Even for chips with same design, one batch manufactured at a certain date at a certain plant can be better then the others. And this better is not even absolute, it is a probability curve, meaning you have a better chance to get a chip that runs at a higher speed.

And in the end, marketing has the power to decide how many CPU's in each speed category they wanna sell to maximize their profit. Of course marketing cannot pull out the number off the air, they have to look at the manufacturing yield and decide. But if they decide that there is 1 million demand for 2ghz cpu at xyz price, 200k demand for 3ghz cpu at abc price and that maximize their profit. They will mark 1 million cpu as 2ghz even tho 500k out of those 1 million is capable of running at 3ghz. I don't think this is disinformtion/misinformation. It's just the way they do business.

So what does that mean to you? The CPU you get may or may not run at a much better speed because of all the different factors involved. And it is for the adventurous to find out where the limit is.

 

[angry hampster]

Originally posted by: Idontcare


Particularly with the entire budget line of processors. Celerons and Semprons.

CPU manufacturers are intentionally underclocking Celerons and Semprons and in some cases have their cache subsystem crippled so as to ensure the consumer only gets what they pay for performance-wise.

What about the low-end Conroe Core 2 Duos? The E6400 is just a 6420 with half the L2 cache disabled. ...Luckily mine's an engineering sample with all 4MB

 

[batmang]

Originally posted by: Idontcare

Originally posted by: Zenoth
Would it have to be "misinformation"? Because I thought that disinforming someone or a mass was to intentionally mislead people into believing a false truth. In this context a company would mislead the consumers into believing that over-clocking does indeed go beyond what the chip was designed for, when in real fact all OC'ing does is to go from an under-clocked state labeled as "stock speed" to the real stock speed that the engineers secretly tested before releasing the chip.

Misinformation would be more appropriate. In the US we use the term "sandbagging" which means to intentionally under-sell or under-report your capabilities.

For instance lets say you know you can run 7km in 45 minutes, but there is an organized race coming up which is 7km long so you tell everyone you can run 7km in 60 minutes. (which is true, 45min is less than 60min)

Then all your competitors get comfortable with the expectation of you only running the race in 60min, so maybe they train less aggressively because they just want to beat you and run 7km in 55min.

But then you show up and you run 7km in 50min (which you knew you could do all along) beating the competition by 5mins (they ran in 55min) but also still not exhausting yourself because you didn't run your absolute fastest (45min) as it wasn't needed.

That is called sandbagging your capabilities. Chip makers do it all the time.

Sorry, but your wrong. Why? Because my Phenom 9600 doesn't overclock for shit. JK. (no, seriously.)

 

[aigomorla]

Originally posted by: batmang

Sorry, but your wrong. Why? Because my Phenom 9600 doesn't overclock for shit. JK. (no, seriously.)

R O F L

Fell off the chair on this one.

then here comes my yorkie running the 7km in 30 min making the people around go

 

[Idontcare]

Originally posted by: aigomorla

Originally posted by: batmang

Sorry, but your wrong. Why? Because my Phenom 9600 doesn't overclock for shit. JK. (no, seriously.)

R O F L

Fell off the chair on this one.

then here comes my yorkie running the 7km in 30 min making the people around go

I thought your Yorkie was a Malay chip...but 30min for 7km sounds like it must be a Kenyan chip

 

[Rubycon]

The technical term for overclocking is speed margining. Chips need a decent margin to be reliable. The overclocker sacrifices this margin or safety net for some speed knowing that better cooling or non nominal (higher voltage) may be required for reliable operation. Of course reliability in an outside of nominal condition can never be guaranteed and such data provided by a computer (under ideal conditions) cannot be fully trusted. Stress testing is not proven at all particularly using prime or distributed computing because it does not exercise every feature set available! Only the manufacturer can do this thus if you want your system to be as reliable as possible and a producer of faithful data (output) it should be run at default speed.

Does this mean that (non o/c) systems never crash or produce errors? Of course not and that just affirms the issue - adding another layer of potential failure or trouble spot in normal operations. This is why a production system (or server) should NEVER be run out of specification. The consequences of anomalies far outweigh the cost advantages of the increased speed.

For the enthusiast OTOH, o/c is pure joy as there's nothing more satisfying than slapping a $200 Q6600 in a system and getting it running at 9x400 easily outperforming a $1500 QX9770. Think of Celeron 300A (which ironically back in the summer of 1998 could be had for the same price - $200; and the P2 450 which cost about half of the QX9770 but was top dog at the time). If those numbers don't put a smile on your face than nothing will.

 

[PlasmaBomb]

nm...

 

[aigomorla]

Originally posted by: Rubycon
For the enthusiast OTOH, o/c is pure joy as there's nothing more satisfying than slapping a $200 Q6600 in a system and getting it running at 9x400 easily outperforming a $1500 QX9770. Think of Celeron 300A (which ironically back in the summer of 1998 could be had for the same price - $200; and the P2 450 which cost about half of the QX9770 but was top dog at the time). If those numbers don't put a smile on your face than nothing will.

yeah i felt kinda dumb when my friends 300A was keeping almost an = pace with my 450.

Because of this i needed 2 machines a 300A oc'd on a DFI to 450. And a regular 450.

The faster cache on the 300A gave my 450 a good run for the money.

 

[Rubycon]

Originally posted by: aigomorla

yeah i felt kinda dumb when my friends 300A was keeping almost an = pace with my 450.

Because of this i needed 2 machines a 300A oc'd on a DFI to 450. And a regular 450.

The faster cache on the 300A gave my 450 a good run for the money.

Yes even with "only" 128kb mendocino could surpass deschutes due to its faster cache. Adding insult to injury I bought 4 and 3 did 466 flawlessly and one did 504. :Q Half a gig barrier broken!

EDIT: Oh and at 300 the 300A had so much of a margin we tried torturing it with no fan and installed Windows 98 on it from start to finish with no problems. The board (Abit BX6) had no way to monitor the onboard diode but I know it was hot - my thumb said so! :laugh:

 

[Billb2]

Originally posted by: Zenoth

Originally posted by: Billb2
99.9% of computer buyers don't even know what over clocking is, much less letting the overclockability of a CPU effect their buying choices.

Ok and you'll have to tell me where you've got that number from? I know a lot of people who know what over-clocking is, they just don't know how to do it. If they knew, they'd do it (well at least that's what they say). So unless you've just made that number up to give yourself some placebo credibility you'll be able to provide me with the source of that information would you?

I sell computers in a store that handles high end machines/parts/custom builds. And yes, my guess is just that, but it is based on years of selling to thousands of customers.

I would also guess that if you compare the number of people that you know that are aware of overclocking their computers to the number of people that you know that have computers, but don't know about overclocking, you would come to the same conclusion.

 

[Rubycon]

Hey just be glad RIAA wasn't into computers telling Intel what to do. If they were we'd have to deal with stuff like this! :shocked:

 

[Idontcare]

Originally posted by: Rubycon
If they were we'd have to deal with stuff like this! :shocked:

I am 100% confident such a day is coming.

 

[Markfw]

Originally posted by: Rubycon
Hey just be glad RIAA wasn't into computers telling Intel what to do. If they were we'd have to deal with stuff like this! :shocked:

Well, I certainly won't be buying any software that does that crap....

 

[soccerballtux]

Originally posted by: Markfw900

Originally posted by: Rubycon
Hey just be glad RIAA wasn't into computers telling Intel what to do. If they were we'd have to deal with stuff like this! :shocked:

Well, I certainly won't be buying any software that does that crap....

By the time this happens we can all just run linux.

But then we'd be painted by politicians as being anti-american, that we should support the american economy.

 

[myocardia]

Originally posted by: soccerballtux

Originally posted by: Markfw900

Originally posted by: Rubycon
Hey just be glad RIAA wasn't into computers telling Intel what to do. If they were we'd have to deal with stuff like this! :shocked:

Well, I certainly won't be buying any software that does that crap....

By the time this happens we can all just run linux.

But then we'd be painted by politicians as being anti-american, that we should support the Indian and Asian economies.

There, I fixed that for you. BTW, does anyone happen to know exactly what percentage of India's GNP comes directly from M$'s and Dell's coffers?

 

[Idontcare]

Originally posted by: myocardia

Originally posted by: soccerballtux

Originally posted by: Markfw900

Originally posted by: Rubycon
Hey just be glad RIAA wasn't into computers telling Intel what to do. If they were we'd have to deal with stuff like this! :shocked:

Well, I certainly won't be buying any software that does that crap....

By the time this happens we can all just run linux.

But then we'd be painted by politicians as being anti-american, that we should support the Indian and Asian economies.

There, I fixed that for you. BTW, does anyone happen to know exactly what percentage of India's GNP comes directly from M$'s and Dell's coffers?

Whatever it is I suspect it is dwarfed by the dollar transactions that come from Walmart's coffers to the same economic zones of discussion.

Did you know Walmart's global sales (>$400B/year) is nearly 2X the size as the entire Semiconductor industry ($250B/year)...including titans such as Intel and Samsung all combined?

We are so used to thinking Microsoft and Intel are these gigantic global monopoly titans...but their annual sales are 1/10 of Walmart.

 

[Diogenes2]

Intel was definitely into over/under clocking with the P3 .. Often, when they sold a higher clocked part, they just set the vcore higher ..

I never have liked the term OVER-clocked ... As long as it works, it's not ' OVER '-clocked ...

 

[aigomorla]

Originally posted by: Diogenes2
Intel was definitely into over/under clocking with the P3 .. Often, when they sold a higher clocked part, they just set the vcore higher ..

I never have liked the term OVER-clocked ... As long as it works, it's not ' OVER '-clocked ...

uhh... 3.5ghz is very far from the word "stock" tho. :X