I mentioned this in the Ars comments on the 3050, but I think this 3000/Ampere generation has two alarming trends that I'm not terribly fond of: power and price. In regard to power, I don't know if Nvidia just wasn't happy with the performance at prior power levels (e.g. 250W at the high end), but having to consume an extra 100W just to hit what should be a generational performance gap isn't terribly impressive. What's worse is that I've always heard that you can sacrifice a relatively small amount of performance if you drop your voltage (and the resulting wattage) a bit. Doing that puts it close to the previous power targets, and it's still a higher performer. It also doesn't help that Ampere also had the benefit of a smaller node in addition to its upgraded architecture and still failed to hit good power targets.
Pricing is a hard thing to really talk about in some respects. The market is such a mess right now that it's hard to say how the 3000-series cards would've been priced if things were comparable to what we had 2-3 years ago. However, we can pick up on pricing trends that are arguably still getting worse. The 2000-series cards were the first to introduce the hefty $1200 pricing with the 2080 Ti, and we've seen the same with the 3080 Ti. The three cards that I've been mostly fine with this generation are the 3060 Ti, 3070, and 3080 (10GB) as they've had decent pricing overall. Albeit, I think it'd be far better if the 3060 didn't exist and the 3060 Ti was at the 3060's price point.
I think it's also worth mentioning that Nvidia branching out from simply graphics is also adding extra complexity and cost to cards, but I wonder how worthwhile it is to gamers in the end? For example, do we think that ray tracing has been worth the extra cost? Removing the RT cores would produce smaller dies and help ditch the RT tax related to all the R&D that it takes to handle it. On the other hand, it seems like the Tensor cores have been fairly useful for the AI processing with things like DLSS. It even gives us things like RTX Voice, which, to be fair, has been shown to work fine outside of RTX-based GPUs too.