Just how much does a CPU cost?

[Shalmanese]

Hi, I just wanted to get a breakdown on the individual costings of a CPU.

- How much does the physical silicon cost to produce

- How much does Typical R&D cost

- How much does advertising, Promotion cost

- What are Intel/AMD's Margins on each CPU

- What are a typical retailers margin on each CPU

I understand if any of this information is top secret but any info at all would be good

 

[zsouthboy]

Intel/AMD 0wns j00!

As for your questions, the only one you'll get a definite answer on is how much the physical silicon costs to produce, which I do not recall at this moment but I will think of it.

The rest of your ???'s may not get very exact answers, but I am also wondering about those. hmmmmm....

zs

 

[Eskimo]

<< How much does the physical silicon cost to produce >>



This is not publically disclosed information, according to some rumor sites a P4 at .18um costs ~$100 to manufacture. The cost of manufacturing a CPU is a function of die size and yield.



<< - How much does Typical R&D cost >>



It takes a large amount of capital investment in R&D by both the CPU manufacturer as well as the manufacturers of the assorted tools that are used to make CPUs. From available filings with the SEC one can see that a company like AMD spends 13-20% of quarterly revenue on R&D.



<< How much does advertising, Promotion cost >>



I heard that Intel spends about $1 Billion a year on marketing, AMD much less of course.


<< What are Intel/AMD's Margins on each CPU >>



This is available in SEC filings as well, Intel has something like 50-60% gross margins usually. AMD is lower, 35-45% I believe.



<< What are a typical retailers margin on each CPU >>



Not really sure.

Hope that helps you.

 

[rimshaker]

Last i heard, the 0.18 coppermine P3 runs around $106 each for INTC to produce... and it still produces profits at this time.

The new 0.13 P4 costs roughly $55.... yes, $55! Quite a profit when they mark it up to almost $500.

As for the other specific costs, you'll never find that out. It's literally secret internal stuff to any company. If that kind of information leaked out, think of the what would happen to the stocks of the companies. They would either skyrocket, or you would find another Enron....

 

[Evadman]

It may only cost $55, but you still have to factor in all the machines, the cleanroom ( bulding? ) maintance, workforce yada yada. Product = inexpensve, labor = expensive.

The raw material that the wafer is made from probably costs less than $1 a die. The machines for the masks and resists ( along with the either, resist itself, yada yada ) are going t be much more. all these hidden costs are never thought of by us "Devil-Spawned end users"

2 points for who correctly identifies my quote

 

[rimshaker]

No, that's it... $55. That is THE floor price for intel. You can't take into account things like the machines or the clean room facilities. Those machines have produced millions of individual cpu cores over the years. How you gonna calculate a meaningful result? Divide the total cost of the equipment by a couple million?? The materials needed to make them, yes those count, but not the buildings, workers salary, etc.

 

[Evadman]

But Intel Still has to pay those workers, and for up keep of buildings and such. That money has to come from somewhere. WIthout the workers like PM, we wouldn't get squat out of Intel. Plus I am pretty sure that they needed new masking machines to go to the smaller process, so that should be built into the price right?

 

[Sid03]

i recently read somewhere that it costs like $2 billion dollars to build a fab. how can you say that shouldn't factor in when you say things like "quite a profit when they mark it up to..."? and i wonder how much it costs to run a fab that costs $2 bil to build.

that's like saying that cd's only cost $.25, so that's quite a profit for software companies.

there's a lot more to business then the cost of the raw materials. i'll give you $55 of silicon, copper and chemicals, and i'll bet you won't find any use for it.

 

[lesch2k]

in a lot of these costs you have to look at sunk and marginal costs, and the companies desire to average the total costs over the entire production.

based upon the published prices of chips my guess is that there is around a 40% markup although this varies between resllers and companies (Dell) which make complete systems.

 

[Daovonnaex]

<< "Devil-Spawned end users"

2 points for who correctly identifies my quote >>


It's from one of the Elbonian programmers from when Dilbert went to Elbonia.

 

[CTho9305]

I dont' think we'll be able to come up with real numbers without the help of an intel/amd employee, who would be fired if they provided that info maybe its so cheap they fear millions of startup competitors if the real numbers get out

 

[rimshaker]

Listen, I didn't come up with $55 all by myself. I just simply read it from somewhere... some EE site i think. I don't know the exact specifics of how they came out with that number, but it makes sense to me. With each die being greatly reduced to the 0.13u process, AND the fact they're now using 300mm (12in) wafers instead of 200mm (8in), they get so much more cores out of EACH wafer slice... provided the yields are very good.

 

[CTho9305]

I also saw the $55 number.

 

[Sid03]

OOOOHHHHHHHHHHHHH... you READ it. well now that's totally different.

by the way, i read that nostrodamus predicted 9/11/01.


i'm just saying that $55 might cover the raw materials. but theres no way that it covers all the other associated costs (r&d, fabs, labor, designing, etc...)

by that theory, if you had $5,500,000,000... you should be able to produce 100,000,000 .13micron p4's. the only way i'm believing that, is if you aren't including the costs of the fabs, labor, etc.

 

[rimshaker]

<< OOOOHHHHHHHHHHHHH... you READ it. well now that's totally different. >>



What's wrong with reading information?? Like you have the necessary resources or contacts to make your own statement? It's not like I read it in The National Enquirer mag or something.

 

[Agent004]

The $55 figure is covering only the raw materials on the cpu. However, the price that intel sells to oem ($$ per 1000 and etc), is the one which covers margin, overheads and other expenses.

 

[Sid03]

maybe not from the national enquirer, but maybe from theinquirer? (http://www.theinquirer.net/06010207.htm)

an article from theinquirer about a report quoting a market analyst (not someone actually in the semiconductor industry) is not what i'd call "reliable" information.

and the way it says how the willamette costs about twice to manufacture as the northwood, and is also about twice the size... well obviously they are only talking about raw materials. if they were talking about all the intangibles, then it wouldn't be a comparable ratio reduction (size/cost)

 

[rimshaker]

Good point... but even an INTC market analyst has access to way more financial information than we could EVER get our eyes on.

 

[Evadman]

<<

<< "Devil-Spawned end users"

2 points for who correctly identifies my quote >>


It's from one of the Elbonian programmers from when Dilbert went to Elbonia. >>



Correct! you get 2 points and a cookie! *Hands out a cookie*

 

[microAmp]

<< Hi, I just wanted to get a breakdown on the individual costings of a CPU.
- How much does the physical silicon cost to produce
- How much does Typical R&D cost
- How much does advertising, Promotion cost
- What are Intel/AMD's Margins on each CPU
- What are a typical retailers margin on each CPU
I understand if any of this information is top secret but any info at all would be good >>



When I worked in a 150mm (6 inch) fab one wafer for productin was about $36 a test wafer for quals was about $27, the difference is the quality of the wafer.....

Also been in a 125mm (5 inch) fab and a test wafer was about $22 (Can't remember exactly) and production wafer was about $27.

For a 200mm wafer (8 inch) my guess would be around the $40 range.

Intel said the P4 at .13um would cost $55 to make but like someone said you have all the other factors.

I know AMD's Fab 25 had a 1 million dollar electric bill per month.

 

[microAmp]

<< I dont' think we'll be able to come up with real numbers without the help of an intel/amd employee, who would be fired if they provided that info maybe its so cheap they fear millions of startup competitors if the real numbers get out >>



Oh well,

I work at Motorola in Austin at MOS 11, former fab was MOS 8. I'll see if I get in contact with a friend over at MOS 13 which makes the G4 CPU's and going for 1 GHz now..... and see what the price of a 200mm wafer is.

 

[Zuidera]

Umn no actually USD55 could actually be the break even per chip (I can buy them for about that from my distributor).

What you have to remember that this is a Duron we are talking about. The average selling price per CPU is quite a bit higher.

Raw materials are negligble, Capital expenses are the name of the game, and they are overcome by high volumes (millions of cpus coming off highly automated production lines).

This keeps new entrants away (big fab costs, so you go fabless - but you still need a good (sellable) design that will achieve high volumes, big capital needed for R&D, wages, distribution system, etc)

You are highly unlikely to get many market entrants, the barriers to entry do exist, the crusoe has come out but the RISC chips are dying, so in net terms we have fewer companies.

btw 5% margin for retailers of PC HW.

 

[Superdoopercooper]

Ok... when I worked at company X, we used 8" wafers from a number of different Fabs. A processed 8" wafer could cost anywhere between $800 and $1500 depending on volumes, complexity, etc (if I'm remembering my #s correctly). We fit 3000 die on a wafer... so divide the total $$ by the # of die, and that gives you cost per die. Then... you have to add package costs, test costs... and any other cost associated with getting that one die to customer A.

You really can't count the fab costs in becuase as someone said... those are SUNK costs... and for fabless companies, they have nothing to factor in. Now... if you have your OWN fabs... then you really just pay for the raw silicon... all the processing is part of the $$ spent to build and run the fab. So, if over the lifetime of that Fab, the company produced $100M P4's.... and the Fab cost $2B, then you kind of have to add $20 of cost per devices. Problem is... you never know the lifetime of the fab and stuff.

So... if you take on of Burr Brown's audio DACs for DVD players or Sound cards... those sell for a couple of buck at most in high volumes (sometimes $1.50 or less). So, you know their cost per chip is likely less than $1.00.... if they want to hit the 50% margin that most Consumer Audio IC makers shoot for (rarely happens... but that is a whole nother store).

 

[Elledan]

<< OOOOHHHHHHHHHHHHH... you READ it. well now that's totally different.

by the way, i read that nostrodamus predicted 9/11/01. >>

FYI, he did not. It was a student from some University somewhere around 1986 who created a 'prediction' to show how easy it is to create a prediction which would come true. Apparently this student did a good job.




<< i'm just saying that $55 might cover the raw materials. but theres no way that it covers all the other associated costs (r&d, fabs, labor, designing, etc...) >>

A company invests in a building and machines to create chips with, but in the end the only things which determine the price of a chip are: R&D, materials, labor. Then the manufacturer and resellers add some more to the price so that they can make a profit.

 

[microAmp]

200mm wafer is $50