Launch availability won't matter to me as I am not building for a few months, but Ampere is a hot mess as far as I am concerned as an architecture and being on Samsung "8"nm. Sure, like Fermi prior, it performs, but it's got its problems and Nvidia is sandbagging on VRAM.
If Big Navi meets, is close to, or beats Ampere and is priced well, I will be jumping back over to Radeon for my next build, no problem. I've been on two Nvidia cards since my last Radeon cards. I'm ready.
I expect Ampere to refresh on 7nm and that will help. It won't be a magic bullet, but I expect it will alleviate some issues they have being on "8"nm from a power and yields perspective, and maybe clocks too. I think it's Fermi 1 to Fermi 2 all over again.
AdoredTV did an interesting analysis on VRAM going back to Fermi and how we generally saw a doubling for each tier each new generation. And Ti cards were typically half of a Titan. Then it started to get weird with Pascal and going into Turing and Ampere hasn't really progressed and if their marketing is believed can be seen as a regression if the 3080 is supposed to be a replacement for the 2080 Ti (Nvidia does keep calling 3080 their new "flagship" yet continually compares it to the 2080 on charts more than 2080 Ti which was flagship of last gen). More credence that the 3090 is a replacement for the 2080 Ti, not the Titan RTX, while having a $300 price hike.
Fermi mid-range: 1GB (GTX 460/560)
Fermi high-end: 1.5GB (GTX 480/580)
Kepler mid-range: 2GB (GTX 680)
Kepler high-end: 3GB (GTX 780 Ti)
Kepler Titan: 6GB (OG Titan)
Maxwell mid-range: 4GB (GTX 980)
Maxwell high end: 6GB (GTX 980 Ti)
Maxwell Titan: 12 GB (Titan X)
Pascal mid-range: 8GB (GTX 1080)
Pascal high-end: 11GB (GTX 1080 Ti)
Pascal Titan: 12GB (Titan Xp)
Turing mid-range: 8GB (RTX 2080/2080 Super)
Turing high-end: 11GB (RTX 2080 Ti)
Turing Titan: 24GB (Titan RTX)
Ampere mid-range: 10GB (RTX 3080)
Ampere high-end: 24GB (RTX 3090)
Ampere Titan: there isn't one (48GB if one comes out?)