Wow. This is so, so wrong. Wowzers.
Not the HW’s fault really, it is more like the storage API’s that were not able to let devs exploit the bandwidth. Once you can like with the new consoles and their SDK’s, it becomes a matter of reducing the cost of access and you add HW to transfer data, decompress it, etc... at zero CPU cost.
VFXVeteran
is correct about being able to build a desktop PC that can do essentially everything these consoles can do: if you can offer 32 GB of RAM or 64 GB (not that far fetched and has not been for a while) you can use a lot of RAM to prefetch tons more data in advance and store temporary data or data which gets reused a lot without having to stream it back and forth.
For now you’d only pay the cost with a higher initial loading time until the SSD’s catch up and devs start getting used to more direct API’s to manage storage (MS will bring DirectStorage over). Those kind of high species desktop PC’s have frequency and cores headroom to be able to go by without as much custom HW. Not a cheap endeavour and still a PC to assemble and manage over time.
Now, consoles wise and up to $399-499 in terms of price, double digits TFLOPS count, etc... there are other discussions. We have two relatively close systems overall with different design goals and their unique advantages. Xbox has fast access to storage and enough HW to accelerate it, but they pushed for more GPU grunt (in terms of shader ALU’s, RT resources, ROPS, and TMU’s which are all in the CU cluster), and have a fast CPU setup (with caveats, virtualisation tax like for the GPU, and highest frequency only if you disable SMT) and RAM (another caveat, fast pool matched to feed the higher performance shaders and a slow one... only one can serve requests at any given time)... not a bad HW and 12 TFLOPS should allow it to run the highest quality options and highest resolutions barring any bottlenecks we are not seeing albeit unlikely (the issue people have is how much can this be flexed in practical terms with a delta that is percentage wise quite small...).
PS5 tried to address and overpower most areas where developers had constraints and complaints and the TFLOPS crown was not the top one on their list: ease of programming, surprises/gotchas free design, and developers flexibility were.
Cerny’s main design goals were old consoles’ ROM-like access times (think GBA or SNES and addressing storage vs one of the various RAM pools they have... hence why they went with such a fast storage solution and all the HW they added on the SoC to make it transparent to developers and remove work from the CPU) and ensuring RAM was not wasted keeping data way ahead of time (just in case, to cover for low HDD speeds like in the past... Cerny stressed how they do not just want streaming to be easier, they want developers to rely on it a lot more also to free RAM up, to store in there only what they really need to use in the current frame/scene and no more [ideal, ymmv]), faster and less bottlenecks at the async compute shaders part (same HW resources there as XSX but much higher frequency for all the units that setup the scene and distribute workload inside the GPU) as well as an interesting quirk to remove even more work from the CPU related to geometry processing and optimisation (the aptly named Geometry Engine which seems something slightly more custom than Mesh Shaders and that they really wanted to stress as do-designer with AMD and something we may see on future AMD designs as they adopt it for their next generation desktop parts).
Given both systems are relatively small RAM wise (XSX only has 4 GB more RAM than Xbox One X) compared to current generation machines, how fast you get the data in and out of them is key. As much as RAM bandwidth and latency is key when you think about fetching data into L2 cache or serving misses.
I see both as well balanced pragmatic designs which will mean PC games minimum specs are due to rise quite a bit to catch up.