Do Xilinx Fix Their Prices?

Steve,

As for older parts, they do not get any less expensive to make. So the price drops until the yields are stable, and then stops dropping. Happens to everyone. At some point, they get more expensive to make as their quantities go down, and the fab line equipment gets more expensive to run (obsolete processes). That is why we send out notices for the end of life of a product (eventually).

That is also why we then go to a new and less expensive technology as soon as we can! If we can make an FPGA for less, our business increases as the number of applications that can afford FPGAs increases.

If the ASIC business had 20,000 ASIC starts in a 20 billion dollar market three years ago, and only 2000 starts least year, you can easily see where increased business for FPGAs comes from.

The parts you mention here, in Peter's scale of FPGA 'life' of 15 years for every 1 year makes these parts very old (>90 years) and just plain old (75 years). Definitely parts that are in the "retired" phase (still playing golf actively, but not recommended for new designs).

As for why things cost less in quantity, that is Econ 101 (for non majors).

Austin

Let me answer some of your questions: The IC business is not so much different from other businesses. We ( i.e. any IC manufacturer, this is all not Xilinx-specific) sell largely through distributors, who are independent companies. We tell them a recommended retail price ( in the US auto stores this is called MSRP), but -just like the car salesman- they can sell it for any price they think is appropriate to optimize their profit. (Same as every grocery store does, too) Selling in large quantity is always cheaper than selling in small quantity. The world is full of examples. Somewhat unique to the IC business is the fast-paced innovation that allows us to make better and faster new chip designs at ever lower price per function. But that is due to smaller geometries, larger wafers, better defect density etc. Once a specific device has been in production a few years, there is little chance to lower its manufacturing cost, once the ramp-up problems are overcome, the yield has been stabilized and the testing effort has been optimized. The cost per quare inch of silicon is not going down. Cost reduction comes from smaller geometries and thus higher packing densities possible in the new designs. That's why you see no cost reductions for mature chips. Real old ones actually go up in price, as their manufacturing volume decreases. (You pay more for a 74161 today than we did 30 years ago). This is a very competitive business.

Peter Alfke

Steve wrote:

Well, not many other businesses see their products' performance/cost increase exponentially with time.

I don't doubt this is true, but it only covers the supply half of the equation. You also have to look at the demand side, because if a successful business model demanded cost reductions, that could probably be arranged on the supply side.

Once an FPGA has been designed into a circuit, engineering costs of changing to a different part are high (and has non-zero risk -- management hates that!). This is a natural form of "lock-in", or monopoly. So, without multi-vendor competition for an existing chip footprint, and prices _can_ stay unnaturally "high" (or in this case, not drop).

Where competition does kick in is for _new_ designs, which (duh) involve _new_ parts. Here the competition is, as Peter wrote, fierce. "Design wins" are a very big deal to both brand A and X. After a brief period of (they hope) high volume, high profile success -- part of which is leading-edge cost effectiveness from the buyer's perspective -- the business system settles down to the operating scenario of the previous paragraph.

So the fundamental difference between FPGAs and, say, SRAM (that leads to the effect that Steve noted and found unnatural), is the lack of cross-vendor competition for pin-compatible parts.

- Larry

Austin, there also is another effect: I do not recall that nuhorizons for example ever changed the prices listed on their website for any FPGA. I see a similar effect when asking for low quantity quotes with insight in germany.

But when the chips get older, it becomes easier to make the distri agree on a lower price for medium quantities.

For short: The distributors buy at lower prices over the lifetime of a an FPGA but this does not mean that they change the price lists.

Kolja Sulimma

So why are the prices *identical* to the cent, at different suppliers, in different countries, 4 years apart?

I must have missed Econ 101, so could you explain why there's such an enormous difference in price between the following?:

From:

XC2S400E & XC2S600E for $27 and $45, respectively, in 250,000-unit quantities, end 2002, and from:

the cheapest you can get them for is $55.45 and $170.00 respectively, for

Steve,

Quite frankly, I am amazed at how folks think about this. You have obviously never thought about that computer on your desk, and how it can be sold for $499! Or even your car, just go price the parts individually some time.

See below,

Aust>

They use the "manufacturer's suggested retail price." As mentioned, these were very old parts, and the pricing was now stable.

You learned the right stuff. Still applies. If a disti orders 100 parts (and they do) we have to process just that many parts for that one order. Disti's don't want to stock anything anymore, so that makes costs go.

Imagine Xilinx' dilemma: what do we build? and when do we build it? If we have an order for 100K parts spread out over a year, everything is trivial, and less costly. But if we have seemingly random orders popping in all of the time, we have to build ahead (risk) and sometimes scrap parts that are not moving.

If you have any optimism about your business at all, it would be best to enter into a agreement and let the disti (and us) know where you think you are going, and how many you will need.

Distribution costs can't be

We can't seem to convince disti's to work for free, however, so they charge what they feel they need to in order to make a profit. Disti's also have 200+ FAEs of their own on their payrolls to support their products, as well as order entry systems, stocking(?), unsold inventory, stocking losses, uncollectable accounts (deadbeats), etc.

As for the book business, I was an author, and if an author gets 1 cent on their book, they are lucky. Ruthless business, with all of the money going to the publisher and retailers. Like perfume, or music CDs, cost of book: $3, price of book $75...... the $9 shipping is a complete rip-off, they already made their profit, now they are icing their cake.

Because they are a fair representation of the costs associated with small numbers of parts ordered through distribution to allow for a profitable business by the distis and reps. They also represent the unwillingness of a customer to enter into a contract which would allow a scheduled delivery of parts over the long term, which is where the real savings start to kick in.

You commit, we commit. You do not commit, we supply product at a fair price based on the fact that you will go away after the one order.

I've often priced the parts for building a computer, and they add up to something within 15% of the price of buying the computer from Dell. Moreover, the price for Intel CPUs in the shop is the same to within about 15% as the price stated for thousand-unit quantities in their press releases.

I believe FPGAs are comparably complicated to Intel CPUs, and I don't think there's as much as an order of magnitude difference in production quantity.

Is the market volatility for FPGAs that much greater?

I can understand that attitude for people buying ten thousand chips; but where do you expect people to get the experience with FPGAs that they have with microprocessors, when state-of-the-art FPGAs are two orders of magnitude more expensive and an order of magnitude less convenient to acquire?

But, again, why doesn't the same argument apply to CPUs, for which there are half a dozen distributors in most towns, fairly happily distributing the things for a couple of percent profit margin.

Tom

An Intel cpu doesn't cost 4.70$ on whatever quantity. And the margin on intel cpus is sufficient on all levels.

The cost is at the FPGA representative, distributing the stuff. They get the questions asked.

You say it. There are half a dozend shop selling cpus per town. You go there, get a cpu, no questions asked, no questions answered. They wouldn't be able to answer anyway.

There may be one FPGA representaive per state. And you ask a lot of questions. Not because you're more stupid than a cpu buyer, but because placing a cpu and applying an FPGA are completely different.

Rene

Buying XC2V6000 at nuhorizons I get a 600$ discount going from 24 pieces to 100 pieces. You really think that there are than 14.000$ distribution cost for distributing 24 FPGAs? (24 times 600$)

Austin: Some of these are self made problems. In germany for example Avnet and Insight are the only Xilinx distributors for your FPGAs. These are huge distributors that have an organization tailored for large customers with large orders. (My last insight order involved about a dozen emails from me and three letters from Insight. I can see that that is costly, but it is unecessary)

Both are reluctant to even talk to me about 2.000? orders and I have prove that they outright lied to me about stock at least twice.

Why don't you get yourself a small distributor in addition to these? One without hundreds of FAEs. Without trade fair appearances that cost hundreds of thousands of dollars. One that is happy to stock a number of standard parts and make a 30% profit with them?

Kolja Sulimma

You may have priced the subassemblies such as the motherboard, CD-ROM drive, etc. Try pricing the actual components (chips, passives, etc.) in small quantity. You'll be lucky if you can get a total BOM cost less than five times Dell's price.

The car example is even better. If you buy all the parts from Toyota (or any of the manufacturers) to assemble a new car, you'll pay ten times the price of the car.

This is because the distribution chain is willing to accept very low margins on the CPUs, because they can move such high volume. Try buying a north bridge chip and see how that experience compares.

So why don't you sell off your web pages for people who don't want FAE support and the rest?

Kolja,

Now you bring up another interesting factor: are we interested in small business, or big business, or both? Are our distributors interested in small business, big business, or both?

I think everyone would say they are interested in ALL business, but...

Not so much different than when you go to a store, buy one can of Coca-Cola, and then ask for a discount, and help with your other bags as you leave the store.....not to mention you would like it a little colder, please.

Do you go to the store often? Is your business worth it? Are they losing money for such a small service? Serious questions.

Small business is not afraid to innovate, and sometimes wildly suceeds(or at least they do here in Silicon Valley on occasion).

I remember a paper by a certain tiny company on networking in San Francisco in 1988 that was laughed at by the attendees of the conference (it was Cisco -- who is laughing now?).

Austin

I just got some XCS30-3TQ144C for $24, although there was another broker who assured me that $67 was a really great price for them. (Yeah, a really great price for HIM!)

I also got a quote a while ago for these parts from a Chinese supplier that was so low, they had to be either bootlegged or Chinese copies of the Xilinx part. I decided not to buy from that supplier, as they wanted a bank wire transfer in advance, too.

No, they get more expensive! I have found that the prices of 5 V Spartan chips is going UP, and maybe going up wildly! (Like the guy who though I'd pop for the $67 chips.)

Jon

Avnet and Insight are the two biggies here in the states, too, and you're right -- they don't wanna talk to you unless you're buying multiple trays. Nu Horizons does have some stock of CoolRunners and I was able to buy a handful, but maybe I just got lucky.

Arrow is usually much better about smaller quantities, but they don't handle Xilinx.

DigiKey has a very limited stock of Xilinx stuff: smaller 95xx CPLDs (all CoolRunner stuff is non-stock, mutliples of a tray), old XC3K and XC4K parts (still very expensive, but if you need 'em they got 'em). They do have some Spartan 2 stuff, but it's really hit or miss. DigiKey really oughta be selling the CoolRunners.

Or just let us buy small (prototype) quantities direct from Xilinx! Yeah, X offers only three CoolRunner parts, 32 (real cheap), 64 (cheap) and 384 (expensive)! I actually didn't care too much that the

Yeah, yeah, yeah, complain complain complain. It's great if you have a relationship with a distributor, but once they find out that you only want to build a handful of prototypes or whatever, they're not interested. There has to be a way for the small garage shop guys to get parts.

I don't know about the US PC prices, but in the UK you can make your own PC for not that much more than you buy a new ready-built PC from a shop. Compare that to Xilinx where small quantities are a few hundred percent more expensive than in large quantities.

It's called mass production. I thought someone who works at Xilinx would have heard of mass production?

You use the example of car prices above, and it's interesting to note that in the UK some of the big car companies were found guilty of price-fixing because they were applying pressure to their dealerships to sell at the price the car manufacturer wanted them to sell at, and if the dealership disobeyed the manufacturer then they faced losing the ability to sell their cars.

This sounds awfully similar to the Xilinx pricing policy...

Take the other end of the scale; what about your highest volume chips? Unless your production managers are extremely bad at planning then you're never going to do make 100 parts when that part sells X million per annum. But all your, say, Spartan IIE chips seem to follow a nice monotonically increasing price vs size graph for a single package type. Why? Because there seems to be very few distributors, so there's little competition, and no true competition if Xilinx order the distributors to sell the products at set prices.

Yes, but the fastest selling chips are still expensive in small quantities.

What about companies that are starting off? It just seems that if you're a big company that can place vast orders then you're alright; if you're small then you're not interested.

Do they though? Or do Xilinx have a say in the prices?

Books are a good example of mass production though because bestsellers sell at £5 or less, whereas slow selling books like niche engineering books sell at very high prices. You don't see this kind of thing with Xilinx's prices though because despite some of your fastest selling chips selling hundreds or thousands of times faster than your slower selling chips the price vs size curve of the chips is monotonic.

Compare that with bestselling books, which are cheap wherever you go, from the biggest to the smallest booksellers. This seems to be the biggest problem, because there seems to be so few places that you can buy Xilinx parts from in small quantities that they can effectively charge whatever they want. And from some of the posts to this and the other thread it seems that the distributors aren't interested at all in small quantities, so they price them at levels that discourage people to buy in small quantities.

You have to wonder whether Xilinx limits the number of suppliers in order to keep prices artificially high?

When do the "real savings start to kick in"? When you order 50,000 per annum? How can start-up companies do this?

-- Steve

Xilinx have a revenue of $1.2bn according to this:

Are you seriously trying to say that the cost of an FPGA representative being asked questions has anything other than a negligible effect on the prices of FPGAs?

You don't expect such shops to be able to answer your questions, but the problem is that there just aren't any shops that you can buy FPGAs from.

See above.

-- Steve

The total BOM cost in small quantities is irrelevant because you can buy motherboards from as little as £40 (~$60) in the shops. The issue is that if you buy Xilinx parts in large quantities then you're alright, if you don't buy in large quantities then you have to put up with high unit prices so those that want to buy high volume chips in small quantities don't benefit from the economies of scale for these parts.

-- Steve

I buy my FPGAs through a distributor. Not Altera direct. Somehow it took me longer to grasp certain ideas. I was talking on the phone with a FAE several times for half an hour or so. This FAE is likely getting a bit more than $5.95/hour. And this service somehow has to be paid. The fewer chips a customer purchases the more expensive he is, assuming the setup-cost per customer to be constant. So obviously the lower the chip count the higher the prices.

Rene