So, $150 per dead part, $150-200 million total. That means 1-1.33 million dead parts that they are admitting to. Small batch, our flabby posterior. Moving on, there are about 250 million computers sold worldwide per year, and about half of those are laptops. The bad chips, G84 and G86 among others, were sold for about 16 months, so 250 million / 2 * 1.33 = 166 million laptops or so in the time period.
Nvidia had about a 70 per cent mobile market share, and laptops have about a 30 per cent GPU attach rate. 166 million * .7 * .3 ~= 35million parts sold into laptops. Of that, not all are the officially 'affected' parts, several lines and older products don't have the same tendency for early death. Another educated estimate says about half the parts are potentially affected, lets call it 18 million bad parts. That would put the admitted failure rate at between five and eight per cent so far.
Using the high number of eight per cent and the low number of $150 million, we can figure out that the the total cost of a recall, again with NV paying only half, is around (100/8)*$150 million = $1.875 billion. Nvidia only has about $1.6 billion in the bank, so this could put a crimp on the decoration at the company non-denominational winter festivities party that does not endorse or disclaim any particular faith, religion, or point of view.
Basically, what it looks like is a recall could bankrupt Nvidia, even if they only pay half. They are quite desperate to minimise this and shut things up because they know what happens if it gets out. It is out, and why they keep continuing to pretend is beyond us, but that is corporate culture for you.
. . .
In any case, the numbers in the end don't lie. Nvidia is admitting responsibility by paying half of HPs cost, and getting twitchy about the cost of having to fix the real problem. Won't it be interesting to see how this ends up? And if Nvidia survives.