What is the procedure in releasing CPU's?

daos

Senior member
Jan 2, 2003
940
0
0
does anyone know the proper procedure that these giant corporations(Intel, AMD) go through in deciding which processor is released and how many of them at a time are manufactured? for example, does Intel have a certain percentage of processors that HAVE to yeild at a certain speed before they are released. if not, then what happens, do they underclock them and sell them as less than what was expected? lets say the processor was supposed to yeild at 2.8ghz stable. they produce 1000 CPUs and only 85% of them actually run at 2.8ghz. the other 15% run stable at 2.6ghz. would they actually underclock the CPU and sell it at 2.53?

is there a percentage of processors NEEDED to run at a given speed and a given "yeild" to mass produce that CPU? and if so, then how many would be produced at that given speed? how exactly is the "yeild" found?

 

AbsolutDealage

Platinum Member
Dec 20, 2002
2,675
0
0
OK, there is a whole lot of info here, and I don't exactly know how in-depth you want this, but here goes.

When AMD/Intel release a "new" speed of processor, it is not really a new chip. The core of the processor, which is the layout of transistors on the chip, actually is exactly the same (unless, of course, they release a new core). What ends up happening is they fabricate, for instance, 1000 chips. All these chips are the same core, same exact internal design. They then begin to put these processors through a rigerous testing process. All of the parts of the chip are tested to ensure proper operation. So, in our little example, say 850 of the chips pass this test. The other 150 chips are thrown away. Now, they take the remaining chips and they start to ramp up both the clock multiplier and the ambient tempurature around the chip. As the multiplier/tempurature gradually increases, more and more chips begin to fail. The company, who has done extensive testing on this procedure, knows that when a chip fails at a certain multiplier/tempurature, then that chip can run reliably at a certain speed. That chip is then fixed at the stable multiplier and batched together with like chips.

The reason that specific chips run at different stable speeds than other chips is due to a very large range of reasons. These can range from the overall sterility of the manufacturing environment, tempurature, equipment, etc. Basically, you are creating such small structures on the silicon that there is an inherant margin of error. That margin of error produces minute differences between every single chip.

There is another whole ball of wax (which I will not get into) which involves the actual release of the new speeds of processors, and it is largely a marketing function.

Anyway, so the basic point is that the manufacturer niether "underclocks" nor "overclocks" thier CPUs, they run them at the highest tempurature at which that particular chip will run stable.

The second part of your question is more of a marketing question as well. If they have not produced enough chips at a certain speed to meet the demand, then they will either drop the price on a higher speed chip, or they will make more chips of the popular speed by locking the multipliers of better chips at that lower speed.

I hope all this gibberish helps you in some way ;)
 

daos

Senior member
Jan 2, 2003
940
0
0
boy i tell you what, that sure did shed some light on the situation. that was a great explanation. if you use MSN messanger or AIM, please contact me so we can talk a bit more in depth about this. i am very curious to know more about this subject.

AIM=daos1234
MSN=kenyon_lee@hotmail.com
Yahoo=daos1234

thanks again buddy and i hope to hear from you soon. :)
 

pm

Elite Member Mobile Devices
Jan 25, 2000
7,419
22
81
I posted some of my comments on this subject here.

FWIW, I mostly agree with AbsolutDealage's post.
 

ProviaFan

Lifer
Mar 17, 2001
14,993
1
0
Originally posted by: dejitaru
Meh, so you can't overclock different machines to the same speed with consistent results?
Exactly. That's why you should never buy a processor based on its supposed overclocking ability; you can not be sure that it will overclock beyond any set point (or even at all). Now, some processors have a history of overclocking pretty well (Pentium 4 1.6GHz processor comes to mind), but even then you can't be absolutely sure that your model will get the same fantastic results that everyone else's does. In summary, just get what speed of processor you know you need, then try to overclock if you wish; consider any success an added bonus. :)
 

AbsolutDealage

Platinum Member
Dec 20, 2002
2,675
0
0
Originally posted by: pm
I posted some of my comments on this subject here..

Very well put together post.

FWIW, I mostly agree with AbsolutDealage's post.

Why thank you ;)

I mostly wast going for more of a lay man's explanation of the entire process here, however, for a more detailed & technical explanation of the process take a look at pm's thread.

 

pm

Elite Member Mobile Devices
Jan 25, 2000
7,419
22
81
Originally posted by: AbsolutDealage
Very well put together post.
Why thank you. ;)
FWIW, I mostly agree with AbsolutDealage's post.
Why thank you ;) I mostly wast going for more of a lay man's explanation of the entire process here, however, for a more detailed & technical explanation of the process take a look at pm's thread.[/quote] If I disagreed with anything it was the terminology - which I why I added mostly. And I felt funny writing it so that's why I added the "For What It's Worth". You are right, my posts are probably a little too detailed and probably hard to read (they aren't to me, but I'm a poor judge I guess), yours is definitely more readable although some things that you wrote are the not way that I would term them.
 

Eskimo

Member
Jun 18, 2000
134
0
0
Originally posted by: daos
is there a percentage of processors NEEDED to run at a given speed and a given "yeild" to mass produce that CPU? and if so, then how many would be produced at that given speed? how exactly is the "yeild" found?

Yes. As pm described in his post your speed distribution usually fits a gaussian. Yield can be defined most simply as a ratio of the number of good die out of the number of total die. Throughout the semiconductor manufacturing process there are a number of yield metrics each of which are important. It's an immutable mathematical fact that your overall yield can never be higher than your lowest in-line yield. Inside the fab you have line yield, parametric yield, and sort yield. These three metrics multiply to give you your silicon yield. Then you have additional loss potentials such as package yield, burn-in yield, and assorted other back end tests that are beyond the scope of your question.

As far as a 'magic' number for yield that gates the shipment of a part it does exist but it is less of an engineering calculation and more of a business decision. A manufacturers sales, marketing and planning organization takes orders for upcoming parts and tries to guage market demand. Sales talks to major customers to try to get a sense of how many units they can comfortably sell, and then being who they are inflate those numbers a bit ;). Marketing determines when it is most strategic to the business to introduce a new product. They may decide to do this before the actual volume of manufacturing has reached the end of the line in the fab. It's up to planning to handle the logistics of ensuring sufficient resources within the fab such as raw wafer starts and lot priority is assigned so that they can get the die quantity out that is required by Sales and the customers. Sometimes along the way something unforseen can impact the production and a fab will not be able to meet it's die commits for that week/month. In cases such as this the company is usually forced to either put a customer on allocation or to try to sell them an alternate product.