Why is that? The major cost is the one-time cost of designing the chip, not producing it. Software is even more extreme: the distribution costs are essentially zero. That doesn't mean the most economically efficient solution is for the software to sell for nothing.
Say you have 2 audiences: Gamers and professionals.
For gamers, a low performance value yields a value of $10, and a high performance value yields a value of $12.
For professionals, a low performance value yields a value of $5, and a high performance value yields a value of $20.
If nVidia sold a single design to all of them, with a high performance value, then gamers would be willing to pay less than $12, and professionals would pay less than $20.
If they want to sell to the entire market, they have to price it less than $12, and lose around $8 from each professional buyer.
However, if they sell 2 designs (e.g: one a cripple of the other), they can derive around $10 of value from each gamer, and $20 from each professional.
The reason this is inefficient, is because for each gamer buying the crippled design, there is a loss of $2 of real value. This loss is a real value loss that could be avoided if there was some other way to extract the actual value for gamers and professionals without crippling the product.
In short, the inefficiency is not that the price is not near the production cost, but that the intentional crippling of the product (necessary to extract maximal value via market segmentation) is causing an actual value loss in the economy.
However in your latter scenario, the net loss is only 2$ versus your original scenario where the net loss is 8$. (lose 8$ per card to the perfessional, versus gamer losing 2$ of performance)
So its socially optimum to continue to do what Nvidia is doing ;).
In the original scenario, there is no economic value loss, there is only monetary loss for nVidia. i.e: the surplass 8$ value are captured by the professionals rather than nVidia.
If they sell at a >4:1 ratio in favor of the gamer version, it is still suboptimal.
If I had to guess, I would say their gamer card prices are by and large good approximations of the value they provide. No idea about professional card prices.
That's not really true, for small runs design time is a significant cost, but at scale it's mostly a question of manufacturing and capital costs.
Consider, Intel produces a wide range of CPU designs specifically because production capacity is what is most expensive. Video cards are something of a special case because they are highly redundant so companies can sell highly damaged chips as slightly different models. However, demand rarely matches the rate of defects so often chips are sold below capacity.
What confuses the issue is 'pro' cards that may preform worse than 'gamer' cards, but cost a lot more. However, that's more a case of different needs. Consider, stability is a lot more important to the 'pro' market which down clocking provides. Toss in some driver tweaks and a small market and you end up with a high priced product that does not cost more to manufacture.
