Navi 33 (their midrange chip, for like 7600X) will be very cheap indeed - it is both 6nm GCD and 6nm MCD io-dies, with a 128b bus. While the cache hasn't grown this gen - it is much faster apparently (in bandwidth) so it's a throughput vs hitrate tradeoff.
Navi 31, the 7900X or whatever, it kinda depends where the performance lands I think. If they match or beat NVIDIA they can pick their price... I think they would probably pick $1499 or similar to at least come in under some perceived high prices. If the cutdown comes in at $999 with a little more equitable configuration that'd do fine, but I'd expect a pretty decent step between the AMD offerings as well. For $1199 or so, eh maybe I guess, they certainly didn't hold back from pricing RDNA2 high against ampere so it's possible, but it wouldn't satisfy any of the critics lol. People want cheap cards and that's not going to happen until at least miner sellthrough starts to slow down - I figure six months before even AMD starts to really crank navi33 out probably.
(Navi 33 may launch at CES and have availability in like, late feb or something, with production ramping, but I bet it's probably March before we see even AMD willing to go... the miner dump is still just getting started, people seem to have been caught completely unawares... somehow. lmao)
Navi 33 was originally planned to go first, pushing it back tells me they're skittish about miner inventory too imo. So I don't think they're going to launch some super-cheap 6800 non-XT like SKU probably... with one chip I dunno if AMD will go much below $1k either maaaaybe $900. Remember, they have to do that with the big Navi 31 chip, and Jensen ain't wrong about node costs either, people are gonna have to get used to it, 5nm R&D+wafer costs are skyrocketing even compared to 7nm, and they're gonna increase again hugely at 3nm. 6nm is gonna be real real popular for a long long time, because it's the only decently-priced node TSMC makes right now.
You're not wrong, Walter, you're just an asshole. Or at least not wrong in general - the Ada price structure is much more about keeping price competition away from Ampere inventory. Can't launch anything too cheap or it'll push Ampere and miner cards down even further. Even the "cheap" stuff like 4080 12GB is really a 4070 or 4070 ti and probably would be an eye-watering $700 or $800 or something in a little more normal market. He's not wrong that it would have been more expensive than people wanted no matter what, node and R&D costs are getting more and more bullshit and the gains are smaller and have more and more caveats and the old "just make it 2x faster per gen!" is no longer reasonable in a post-Moore's Law and post-Dennard Scaling world.
(calling it a "4060" is bullshit though, because that ignores the contribution of the cache... it's like calling a 6900XT a "RX 480 successor" because it only has a 256b bus. The cache changes things and anyone who's doing the "4060" stuff can be pretty safely discounted as being hysterical.)
This power density on 5nm shown so far with Zen4 and Ada probably will not spare AMD either - I bet they're at 400W or higher as rumored a while back, if they're around the same performance as a 4090. 5nm is just super super dense and clocking it high takes a lot of juice.