Mark Roulo (nothanks@xxx.com) on 8/4/10 wrote:
>
>I was hoping (still am, actually!) for some insight into
>the relative profits between the lines, though. For
>example, if the bottom 80% by volume GPU sales only
>contribute 10% of the profits, then I can see surviving
>w/o the lower 80%.

No. No. No. No.

You cannot count/think that way. It sounds "obviously
correct", but it is, in fact, total and utter garbage.

Why? Because a lot of the costs are fixed. The cost of
things like driver development is largely fixed. The
cost of much of the engineering is similarly largely
fixed.

And those costs aren't "per SKU". They are generally
across the whole (or at least a large part) of the line.

So let's go through what that actually means. Say that
nVidia sells a hundred million low-cost chips, and only
5 million high-cost ones. And lets say that the profit
on the low-cost chips is zero just to make the
point really obvious, and all of nVidia's actual profit
comes from the high-cost ones.

According to your logic, nVidia would make as much money
if it just dumped the totally profit-less low end.

But think about it. "profit" isn't actually really money.
It's the difference in cost and sales - and those are the
"real money" numbers that you need to look at. And if many
of the costs are relatively fixed, and not very directly
related to how many units are shipped, what happens?

Think about it. Really.

Those "low end" cards that made zero profit were
still paying for costs. Now, they didn't pay any more for
the costs that they brought in, but they very much supported
the whole operation.

And if they go away, the sales numbers go down, and they
go down a lot more than the cost numbers did. The cost stays
relatively constant - remember how driver development was
not cheap? So suddenly the "costs" of the high end actually
go up, even though you got rid of something that didn't
help your profit at all.

Now, admittedly, there are few totally fixed costs. The
driver development would certainly be a bit cheaper if you
didn't need to worry about cards that had just half the
raster pipelines or whatever. And there are development
costs to doing those half-unit designs too, as well as with
just shrinking and maintaining old designs and selling them
in the low end.

Even "pure silicon" cost isn't linear with volume. Not only
do you want to bin those chips (by speed grades and number
of working units), your silicon supplier is likely to give
you better prices per wafer if you do lots of them. Or just
preferred treatment and care more when you have troubles
with your vias.

So no. It's a total mistake to think that you can skip the
low-grade chips because they don't bring in much profit.
They may not give you those sexy profit margins, but they
are (or should be) what pays for your fixed costs.

The high end is pure candy. It may be what you enjoy, but
without the bread and butter you're going to be very very
unhealthy.

Linus