If that was true then they could already markup the cost for George Hotz (Tiny Corp) wannabees buying consumer gpus for "AI servers" that are sold for tremendous profit.
I think its more that wafers will be devoted to the AI chips while the gaming chips get less wafer starts and a more constrained inventory thus leading to higher prices. We might see the 5070 at the 4080 prices (1200) and the 5060 at the 4070 price (800) and so on and so forth. Who really knows.
GB202. GB102 is supposedly a half of GB100 and it won't come to gaming markets (likely missing RT h/w too).
It won't, it's not. The whole hysteria about how Nvidia suddenly decides that graphics don't matter anymore isn't based on anything but wishful thinking. Both Intel and AMD have infinitely more reasons to decide that and yet somehow they don't either.
DC products aren't constrained by wafers, they are constrained by the advanced packaging lines. These lines aren't being used for gaming chips.
We might see 5070 at 4080 prices and if it will perform like 4090 then it will be a sizeable perf/price gain and wouldn't matter at all.
In consumer space questionable matrix acceleration and RT aside, NVIDIA has one GPU faster than AMD. So asking them to "wake the f up" is a bit out of touch IMO.
If someone thinks that way about NVIDIA, then he/she probably doesn't understand NVIDIA well enough, they are an opportunistic and conquering entity, meaning they love to conquer new markets, for years they kept their Automotive segment on life support even though it brings them very little money, just because it allows them access to new markets. The Gaming segment brings them huge money, they are the market leader there. Gaming chips are also shared with the Professional Visualization segment (in which they are the market leader there too) and Automotive segment. They have no incentive to abandon any of these segments.
Their RTX 40 Super series is a direct contrast to that idea, they adjusted prices down and released more perf/$ parts, even though they don't really need to, but they did.
Why didn't they need to?
Well, they have that AI money, why would they care if a 4080 costs 1200$ instead of 1000$? It's the argument people are making to incorrectly expect a 1500$ 5080.
Do you have anything to back up the notion that they were "sitting on all that inventory" prior to 40 Super launch?
First of all, AI GPUs are NOT higher yields, they are chips sized at the reticle limit, which means they have much lower yeilds.
They can sell both at the same time, and make even more money. That's how businesses work.
That's not entirely true, only when you compare against Crypto bubbles, but exclude those bubbles, and the number is relatively stable each year.
That's not true, even in the hypothetical situation of gaming sales going way down, NVIDIA can subsidize whatever they need from the huge pile of AI money they have. That's what Intel and AMD do with their not successful Arc and Radeon segments anyway.
Sorry, I can't take this seriously, companies don't operate like that let alone NVIDIA. Besides, in your logic Jensen is keeping autmotive as a pet project while getting rid of the gaming which is the company's bread and butter since it started in the first place, that doesn't make any sense whatsoever. Jensen doesn't hate gaming, half his company is doing gaming related work, and part of the AI side is also doing gaming related stuff too, you expect the company to get rid of all those people?!
All of these chips are gaming GPUs with various sizes.
