Why Are CPU Prices Increasing So Much?

[Sakha]

The other day, watching to buy a new computer I saw CPU prices, and I asked myself why prices are so high.
I mean, its just a tiny piece of semiconductor.
Does the high price comes from the design of them, or it's that manufacturing them are expensive and machines to work at such small dimensions are expensive?

 

[russ_watters]

It's all in the R&D. And I wouldn't say they are all that expensive historically.

 

[Greg Bernhardt]

Like russ said, the materials are likely relatively cheap. But the money it takes to develop new technologies is quite large.

 

[waht]

A relatively good CPU is just under $200 bucks. The price is mostly set by the supply-demand law, but it's an excellent value considering what went into making it.

 

[Sakha]

So what would you guys estimate be the price of the actual CPU, not counting the R&D, nor the equipment, just the actual product? $25?

 

[mgb_phys]

So what would you guys estimate be the price of the actual CPU, not counting the R&D, nor the equipment, just the actual product? $25?

After R+D the biggest production cost is paying for the plant.
A 45nm fab with 300mm wafers producing around 25,000 wafers a month costs $3-5 Billion to build.

The area cost of a Pentium type CPU on 300mm is around $10-15.

 

[Sakha]

After R+D the biggest production cost is paying for the plant.
A 45nm fab with 300mm wafers producing around 25,000 wafers a month costs $3-5 Billion to build.

The area cost of a Pentium type CPU on 300mm is around $10-15.

Wow.. that's a lot.
So R&D is what, $20 Billion, for let's say, a new model?

 

[nuby]

Brand is expensive too.

 

[mgb_phys]

Wow.. that's a lot.
So R&D is what, $20 Billion, for let's say, a new model?

Intel spends around $5B a year on R+D.
There is also a lot of spend on getting the plant to operate at peak efficency.

 

[Dale]

Also, you have to consider the small window of time for recovering the R&D cost. Because the technology is continually improving the sales lifetime of a new CPU design is very short, so the R&D cost also has to be recovered quickly and cannot be spread over a decade of sales.

 

[Nick M]

I think prices are very reasonable for what goes into putting one into a consumer's hands.

Although admittedly they aren't bread & milk type purchases for the average person.
My new workstation has a pair of 2.8GHz Harpertown Quad-Cores... and I've done some sacrificing to get that machine under my desk.

 

[raj123mehta]

The price at manufacturing time is low then many tax and other expenses make price high, second thing if price are very low but other manufacturer selling it on high price and people are purchasing it without hesitation than who will deny this and will sell on low price?
________________________
Raj Mehta
Uninterruptable Power supplies

 

[michinobu]

The other day, watching to buy a new computer I saw CPU prices, and I asked myself why prices are so high.
I mean, its just a tiny piece of semiconductor.
Does the high price comes from the design of them, or it's that manufacturing them are expensive and machines to work at such small dimensions are expensive?

Dude, why don't you go to a flipping marketing forum or a computer engineering forum? Costs concerning CPU's isn't a computer science related issue unless you're talking about how computer science techniques can be used to predict optimal prices for CPUs provided that you have certain factors that go into it, etc.

But, this by in of itself isn't a csci issue.

 

[Enthalpy]

I feel processors are very cheap as compared to the huge effort to develop and produce them. So cheap in fact, that only two companies have survived.

They cost 200$ instead of 20$ because customers are willing to pay 200$ to get a better one. And they do get something for their 200$.

If you don't want to give that amount, buy older or used hardware as I do. Very interesting for computers, as they lose their merchant value too quickly.

About material costs: Cpu-grade silicon isn't sand. It's 10s times as expensive as gold per kg.

 

[nuby]

The funny this is people don't understand a standard 1.8 ghz CPU is all they need (normally).. But they all want the 2.5 ghz duos, etc, which cost $200 more... and many people don't realize the most important hardware on a computer is the motherboard.. And the huge bottle neck in overall performance is the slow speed of the hard disk.

 

[HallsofIvy]

So what would you guys estimate be the price of the actual CPU, not counting the R&D, nor the equipment, just the actual product? $25?

How much would a Picasso cost if you only count the cost of paints and canvas?

 

[CRGreathouse]

About material costs: Cpu-grade silicon isn't sand. It's 10s times as expensive as gold per kg.

Really? A quarter-million dollars for a kilogram?

Forgive my skepticism, but do you have a source for that?

 

[mgb_phys]

The raw silicon is very cheap - I think the poster is confused by some statistic of the cost of a finished chip. A CPU 1cm square and 0.1mm thick would be expensive / kg.
A raw 300mm Si wafer is a few $100, it's a few $1000 to put the CPUs on and process.

Just out of interest, the latest 45nm quad-core CPU die is 220mm^2 and costs $1K
The most expensive painting ever sold is 138*138cm = 1.9M mm^2 and cost $135M,
so paint is still 15x as expensive as sand.

edit - I had the Klimt portrait listed as the most expensive.

 

[CRGreathouse]

A CPU 1cm square and 0.1mm thick would be expensive / kg.

I agree with that. I remember reading (perhaps ten years ago, when chips were bigger!) that the price of processors was more than gold per pound.

Just out of interest, the latest 45nm quad-core CPU die is 220mm^2 and costs $1K
The most expensive painting ever sold is 138*138cm = 1.9M mm^2 and cost $135M,
so paint is still 15x as expensive as sand.

Okay, now you have me hooked. I found a list on Wikipedia of the most expensive paintings
http://en.wikipedia.org/wiki/List_of_most_expensive_paintings
and used their inflation-adjusted numbers for sale price along with the dimensions for paintings (as found by Google or the like). The best I've found so far is van Gogh's "Portrait of Dr. Gachet" with an inflation-adjusted value of $868 per square mm.

The Mona Lisa is 2-3 times larger, but might be worth five times as much...

 

[raj123mehta]

Dude, why don't you go to a flipping marketing forum or a computer engineering forum? Costs concerning CPU's isn't a computer science related issue unless you're talking about how computer science techniques can be used to predict optimal prices for CPUs provided that you have certain factors that go into it, etc.

But, this by in of itself isn't a csci issue.

Exactly ! I completely Agree with you. Well said..




_____________________
Raj Mehta
http://www.powercontinuity.co.uk"

 

[Werg22]

Aren't modern CPU's very difficult to engineer due to the push for more speed all while having hit a ceiling a few year back?

 

[mgb_phys]

All CPUs are difficult to engineer, always have been and always will be!
When they hit the 3GHz ceiling it got trickier having to get more performance out of smarter pipelines, branch predicts and other optimisations. If anything multi-core chips are easier, just copy and paste the core design 2 or 4 times!

Of course getting 2Billion transistors on a 1/2in square bit of silicon to work is harder - but that's the fabs problem!

 

[mmix]

http://opensparc.net opensparc is based on FPGA, and it's license is free.