4000 series is coming out soon. If I were you, I'd wait, since 3000 series is going to become an obsolete piece of crap within a few months, at most.

Cries in 1000 series... no, wait, my GTX 1080 is still a perfectly fine card for 1080p gaiming! Gosh darn it, maybe, just maybe older generation hardware doesn't expire the second something newer comes out...

It doesn't, older hardware makes its way to the ''developing'' countries where they're sold at higher-than-MRP because newer hardware is even more overpriced and we don't have other options. So do not despair, the third world envies you with their $100 GT 730 4GBs and $250 GTX 1050tis.

I know, I was being sarcastic... my current GTX 1080 is the only GPU I've ever splurged on, to be honest. The rest of my life I was stuck with entry level or at best lower-mid range GPUs, so I know that pain well ;).

P.S. The 1050Ti is still a great card for older games at 1080p (I finished the Witcher 3 on a 1050Ti). It all depends what games you want to play.

Everything you want to play now would probably run just fine on a 1650 Super, if you can find one for normal prices now. I don't know what the AMD equivalent of that is off the top of my head. That said if you're going to play anything more demanding like... off the top of my head... Elex 2, Mafia remake, etc... then I would look more at a 3060.

Once prices normalized I got a 3070ti and I do kind of feel like it was pointless, as my 2070 could run everything out now just fine and Starfield is a year away and 4000s will be out by then. The 3000 series was a huge performance jump though, I doubt the 4000 series will be. If you're gonna try and wait for the 5000 series then I do think a 3060 is a better option, again assuming it goes down to normal price.

4000/ Lovelace will be a big jump in performance at least at the top end- pretty much has to be. It may well be a rather oddly distributed increase though. And while we may get Lovelace and RDNA 3 cards this year they almost certainly won't be anything approaching affordable, they'll be the halo offerings with high prices.

But performance wise you're looking at both a node switch to a considerably better process and another big increase in transistor/ cuda core etc counts. Balance will be that TSMC is expensive relative to Samsung, and the extra transistors will outstrip efficiency gains. So while the 700W rumours are likely just that something like 450W is fairly realistic for top end. arch wise it's more or less Ampere+ so not much to gain, and they will be massive monolithic chips with a decent likelihood of being imperfect (ie a lot will be sold as a lower SKU) but for everything else there are gains.

The open question is certainly how much improvement there will be at the lower/ mid end though, and that is most relevant to most people. That depends a lot on which node they manufacture on. Since Lovelace was a bit of a stop gap in response to RDNA 2 being more competitive than expected and AMD being ahead in MCM they may well have difficulty getting consumer level volumes out of TSMC which has a lead in booking time of literally years in some cases. They did book capacity early, but we're only certain that it was for pro/ server level volumes.

The two most likely situations are that everything is TSMC and the affordable cards are slow in arriving and more expensive or that they use a different foundry or older TSMC process and the cards are slower/ have less improvement relative to the top end, perhaps significantly so, but are cheaper.

Yup, the feeling of finally having something that can run the games you want is great :D Sounds like you got yours before the current madness started so props for dodging that bullet. I also got my first good GPU just before the pandemic and I thank my RNG for it every time I see the hardware prices today.

The 1050ti is indeed great but the price for which it (and the 3050 / ti) sells in my country is absurd considering its position in the stack and its age.

If OP can find a 1050ti or 1060 6GB for a good price though, it might suit his needs very well.

3000 series was like an 80% performance bump over 2000. I don't see us getting those kinds of gains back-to-back. Would be happy to be wrong though.

Well, it's all rumours so far so practical mileage is likely to vary rather a lot, but the reasoning goes...

Process 8nm Samsung to TSMC 5nm should give something like 40% improvement by itself on an equal power basis, at least in theory.

The top chips are meant to be absolutely massive, bigger than Ampere's despite the node shrink, and have more power draw also despite the node shrink. That implies a lot more (+70%) transistors. That doesn't equate to 70% more performance though of course, but may well translate to ~35%.

OTOH, 'IPC' and memory gains are likely to be minimal. 'IPC' did not actually improve that much for Ampere though either, that too was mostly node improvement and more transistors (though quite a few less tensors). Still, the two above imply a big jump in absolute performance. Probably a decent efficiency gain too, but a 700mm^2 TSMC 5nm chip will not be cheap, so likely nowhere near as big an improvement in price/ performance.

The really interesting thing will be seeing how AMD's offering turn out, since Lovelace is only really a thing due to RDNA2 outperforming expectations massively and nVidia worrying about AMD's lead in MCM. In theory (again) 3x40CU gives +50% improvement even if RDNA3 is identical in every respect to RDNA2, and not taking their node change into account.

Would certainly be cool to see.