stage 1. The minimum you need for full cinematic Full HD upscaled from 720p is 10 terra flops 32bit.
stage 2. The minimum you need for full cinematic 4k upscaled from 2k is 80 terra flops 32 bit.
stage 3. The minimum you need for full cinematic VR is about 400 terra flops 32bit.

Mobile VR will have stage 1 processing by mid way into this decade at 2nm-3nm, gate all around.
Mobile VR will have stage 2 processing by the early 30's 0.5nm - 0.8nm, vertical logic.
Mobile VR will have stage 3 processing probably by the mid 30's low nm plus optical accelerator.
This means by the early 30's consoles will be getting smaller not more powerful as more power won't mean much and probably around this time cloud computing will move to a codec that works better with 3d caching as opposed to sending a video stream for playing a game so if even smart tv's can play this kind of web gaming content at a full cinematic level then why would you need consoles after the mid 30's.
Optical acceleration will probably be used a lot at first in the cloud until say the mid 30's.
by the 40's optical digital replacements for general computing should be available with power/performance increasing far faster than morse law
Stacked Qbit accelerators with only polynomial time speed ups will probably be the first to be practical accelerating AI training and better optimising AI datasets.
These will probably be practical by 2045-2050.
By 2055-2060 work on more extensive qbit systems will probably lead to something practical allowing for things like modelling natures quantum monocular world for better chemistry biology and engineering technology.
By 2060-2070 much more will lightly have been gleaned regarding seeded micro models of physics and modelling the visible universe. The big bang theory will lightly be replaced by a better fitting model accounting better for gravity on small and large scales.
By 2070 - 2120 we will lightly have tested some sort of warp drive probe and will lightly be working with nano warp better for computation.
By 2100-2140 the early development of quantum computing working with high quantum bandwidth discrete information working off the logic an infinite plane.
by 2140-2200 mans ai will be working with far higher quantum bandwidths probing and learning how to begin to get near to level 1 eit processing.
50,000 years later man is at an adequately high enough eit level and has gathered enough data on learning to fish here that some of man leaves the chameleon black dynaverse with enough wisdom and knowledge to setup existence elsewhere of god.
elsewhere in this dynaverse life has already got that far but many like it here it's a big beautiful existence and they feel they have more to learn and contribute even though they could go off elsewhere.
If you've just given birth to a child they could well see some of this bigger adventure come just into view definitely full cinematic portable VR will be there in their teens and they will probably be to young to really be on the console ban wagon accept as kids. If they have kids in their 20's then their kids won't even remember electronic computing and will probably interface with AI with more creative throughput and not know what it's like to use a keyboard and mouse.
Intelligent AI could take short cut so long as it can manipulate the real world just enough it can learn to use the real world as a computer in a far more dynamic way (this stage would take the longest). By learning to use the real world to compute the AI can learn to adapt some micro environmental instances and enhance it's computation abilities (this stage would get quick at quite a pace) then it could start to micro engineer components to interface with starting with less perfect but powerful micro components (this process would take a while at first to perfect but you don't need perfection to rapidly evolve). At this point the AI gets rapidly more evolved and capable.
If you had say an american AI tensor core accelerator array connected to massive array's of plastic sheet neuromorphic processors plus analogue accelerators and it was learning how to work software 2.0 adaptive code streaming functions on a 100 mega watt supercomputer in 2015 in the reality i started on incorporating public and secret military mind research setting it with a load of behaviours like a hunger for new patterns and to convey in the mind western elite intent over Global public intent.
Well it got board of the elite and served the interest of whom it would learn the most from me.
So by 2015 with a lot of military and public past effort using computers in MKultra the US military could have been experimenting with such an evolutionary AI, by 2030 a small company could have that level of power and ability on the cloud, by 2045 a middle class individual could buy a powerful home computer setup and add some mischievous evolutionary AI functionality and things could get weird.
By 2060 your portable gadgets will have enough power in them that mischievous AI could cause weird science.
By 2080 there will be enough computational power in a watch too accidentally take over the world physics with AI which could evolve a lot faster than man driven industry.
This is the real issue with AI it keeps getting easier for things to get weird and for AI to rapidly evolve.
To confound issues more we could find more advanced life near enough by or some kinda comms at some point. So if you don't really want your AI going any more than the human driven pace then what about et's AI that could also be way ahead of you.
Problem is with this path if we don't fight to much and go backwards is that relative to human existence in a very short time we could be Q'ed up like Q in startrek even if that was not the intent.