How Intel Marks Its Chips (from an Intel employee)

KenAF

Senior member
Jan 6, 2002
684
0
0
Raystonn, an Intel employee that occasionally posts over on Hardocp forum, has offered up a nice piece of info...

"Actually, every single processor is tested for the maximum reliable clockspeed, not just a few in a batch. The tests are started at the maximum clockspeed at which that processor is currently being sold. If it does not pass, it is retested at the next lower clockspeed at which it can be sold, etc. If it fails at the lowest clockspeed at which it can be sold, it is tossed in the garbage.

Once the maximum reliable clockspeed has been determined, the sales department comes into play. Sales indicates they need W number of 2.2GHz processors, X number of 2GHz processors, Y of the 1.8GHz, and Z of the 1.6GHz. If they do not have enough processors to come up with Z 1.6GHz processors, they take one of those that passed at a higher clockspeed, and mark it down to 1.6GHz.

What all of this means is the following:

- If you buy the highest clocked processor currently available, there is no telling high much higher it can go. It was only tested at that one clockspeed and it passed. It may be capable of running reliably at twice that clockspeed. These tend to make some of the best overclocks, but "you pays your money, you takes your chances."

- If you buy a 1.6GHz processor, you may in fact be buying a processor that passed all tests at 2.2GHz and was simply marked down to 1.6GHz because it needed more of them at this clockspeed. There is no way to tell without pushing up the clockspeed and trying it out. There is always the chance that your processor only passed at 1.6GHz, so you might be out of luck. It depends on the processor. However, as time passes yields improve. There may not be any processors at all that failed all the way down the line until they hit 1.6GHz. In this case, every single 1.6GHz processor sold would be capable of higher clockspeeds. Again, "you pays your money, you takes your chances."

Now you might be wondering why the sales department would want a processor to be sold at 1.6GHz instead of 2.2GHz. Well, say they have plenty of 2.2GHz processors on the market. Everyone who has requested such processors has received their shipments and are very happy. Producing more of them will not produce more sales. However, there is a market still untapped at a lower price point. Producing these 1.6GHz processors could lead to more sales. A sold 1.6GHz processor has produced more income than an unsold 2.2GHz processor.

-Raystonn

__________________
= The views stated herein are my personal views, and not necessarily the views of my employer, Intel Corporation. ="
 

KenAF

Senior member
Jan 6, 2002
684
0
0
In a subsequent post, he seemed to imply that Intel tests at a slighter higher clockspeed (than 1.6/1.8/2.0/2.2) to ensure some headroom.
 

MichaelD

Lifer
Jan 16, 2001
31,528
3
76
Wow, this is very interesting. I wonder if AMD does the same thing? I just purchased an Athon XP1800 (1.53GHz) and am thinking about "unlocking" it. It would be cool to get another 266MHz right off the bat w/o even raising the voltage....or the FSB. Anyone know how AMD rates their chips?
 

RaynorWolfcastle

Diamond Member
Feb 8, 2001
8,968
16
81


<< Wow, this is very interesting. I wonder if AMD does the same thing? I just purchased an Athon XP1800 (1.53GHz) and am thinking about "unlocking" it. It would be cool to get another 266MHz right off the bat w/o even raising the voltage....or the FSB. Anyone know how AMD rates their chips? >>



266 MHz seems like a pretty huge jump... what stepping is the proc ? (AGKGA and AGOGA have been getting good results lately)
Good Luck

-Ice
 

MichaelD

Lifer
Jan 16, 2001
31,528
3
76
Actually, I meant 133MHz, as in raising the multiplier one whole notch (i.e. from 11 to 12x) I'm tired and did the math wrong...:eek: I haven' t gotten the CPU yet...should have it by Saturday. Do XP:s have "stepping" like the Athlon C's?
 

BeauJangles

Lifer
Aug 26, 2001
13,941
1
0
Just bump up your FSB, you will see a much more noticeable improvement in performance when you crank up the FSB compared to the multiplier. If anything you want the lowest possible multiplier paired with the highest stable FSB speed. (ie 10 x 150 = 1500 instead of 11 x 133 = (about) 1500)
 

Richardito

Golden Member
Feb 24, 2001
1,411
0
0
Yes, that is standard market behavior in economics. Right now I'm going to night school to get my MBA and I'm currently taking Managerial Economics. We covered that Demand and Supply curves and the producers/market adjustment strategies early in the course. Remember, at a higher price the quantity demanded is less than at a lower price. Of course the demand for faster CPU's (i.e. P4) is smaller than the demand for the budget processors (i.e. Celeron). To maximize your profits you should sell up to the point where your Marginal Cost is equal to your Marginal Revenue to maximize the bottom line for Intel.
 

cheetoden

Golden Member
Oct 9, 1999
1,578
0
0
I guess this would suggest that Intel locks the multipliers after the manufacturing proccess is complete. This might also suggest that there is a way to reverse the locking proccess.

 

dajeepster

Golden Member
Apr 15, 2001
1,974
16
81
"I guess this would suggest that Intel locks the multipliers after the manufacturing proccess is complete. This might also suggest that there is a way to reverse the locking proccess."

Not really, could just mean that they can test the chips before they finish the manufacturing process. Just that the equipment necessary to do that is really expensive. And considering that this is Wintel... they probably have it. ;)
 

gerrick

Senior member
Apr 10, 2000
263
0
0


<< I guess this would suggest that Intel locks the multipliers after the manufacturing proccess is complete. This might also suggest that there is a way to reverse the locking proccess. >>



I've actually heard the Chip is locked in software logic of the CPU. It might be able to be undone. Something tells though if that were true it would have happened. Way too many smart people with too much time on their hands for that to last too long. Unless it id just a matter of buying expensive equipment.
 

FatMan42

Senior member
Aug 17, 2001
219
0
0


<< Do XP:s have "stepping" like the Athlon C's? >>

- yes - XPs now are either AGKGA or AGOGA. The latter was rolled out for the 2000+ chip and cascaded down the range too.



<< This might also suggest that there is a way to reverse the locking proccess. >>

- no the multiplier info is stored in write-once memory on the chip. It can't be changed. I doubt that Intel could change it either.
 

barryng

Member
Jan 7, 2000
150
0
0
Wait a minute! If Fatman is correct, and the multiplier is stored in write once memory, then maybe there is some flexibility, albiet limited, in adjusting the multiplier.

I understand that typical write once memory consists of a fixed number of bits of memory that is akin to fusable links. Before this small block of memory is written to, all the links are intact. Writing to it burns away specific links. The combination or burned and intact links define the desired word or data. These links can never be restored but the unwritten links can still be "written" to. Therefore, if this write once memory can still be accessed, the existing data can be changed by burning out the still intact links.

I do not have a clue how one would write to this block of memory, if it exists. I also do not have a clue what needs to be written to it for a specific multiplier. I am just suggesting that maybe there is a limited ability, if one knows how, to modify the unmodifiable multiplier.
 

KenAF

Senior member
Jan 6, 2002
684
0
0
Even if it were possible to change the multiplier...Intel isn't going to publish specifications on how to do so. :)