efikkanWell, Nvidia have about 10x the production volume of comparable chips, which means the various piles of "usable" chips becomes larger, and when they are produced by the millions, this will result in thousands of such chips.

It still doesn't explain why they have much more "waste" chips than AMD compared to fully functional ones.

AusWolfIt still doesn't explain why they have much more "waste" chips than AMD compared to fully functional ones.

-Its just a question of naming, I think.

AMD will release an "LE" chip, like the 7900GRE as a catchall for their faulty N31 dies.

NV would have released the same chip as a "shadow" 7800xt, making some deeper cuts to arrive at 60cu.

NV has more dies (AMD is trying to address the whole market with 2GCDs, 1MCD, and N33) and there are more chances for them to rob sales from themselves by naming a defective AD104 die xx80LE than just calling it a 4070.

AusWolfIt still doesn't explain why they have much more "waste" chips than AMD compared to fully functional ones.

It does, as running much more wafers means there will be more in all the various failed bins, and some of them will have enough to consider making a "product" from it.
It's not worth it if there is only a few hundred samples.

Hardly a refresh. Only refresh than makes sense is to rebadge the 4070 (non-super) as 4060 Ti at $499.

efikkanIt does, as running much more wafers means there will be more in all the various failed bins, and some of them will have enough to consider making a "product" from it.
It's not worth it if there is only a few hundred samples.

Are you saying that defective Navi 32 and 33 chips land in the bin? It sounds excessive to me when they could just make a 7700 non-XT, or a 7500 XT instead, and still make some money on them.

AusWolfAre you saying that defective Navi 32 and 33 chips land in the bin? It sounds excessive to me when they could just make a 7700 non-XT, or a 7500 XT instead, and still make some money on them.

Whenever AMD (or the others) develop a new chip they assess the first batch of production wafers to segment the chips into various product tiers, including desktop parts, mobile parts, enterprise parts, special OEM parts, and possibly special tiers for single customers. All of these will be various bins, including the ones which are not yet intended for a product. And the very last bin will of course be discarded chips. AMD have some flexibility in deciding how many tiers they want, but too few tiers will result in large variation inside a single bin, and the resulting product will have to be specced based on the minimum requirement of the bin. As yeids improve, the bins might also change, and we often get product refreshes. Still, there will usually be various bins of semi-functioning dies.

And here comes the point; the more wafers you run, the more chips will be in each bin, resulting in higher potential to turn them into products.
I.e. if 0.5% of dies are having large defects on otherwise good chips, then making 100,000 vs. 1,000,000 of these chips would result in 500 vs 5,000 in the bin. As Nvidia produces much more chips than AMD, they have lots of more options of using these bins. But AMD does it too, including OEM only products or chips for custom products you've never heard of.