You are kind of derailing the thread with this… start by dividing the speed by the framerate and then think about latency of getting data, compression speed, and think about making it transparent to developers (how to map large files and chunks of large files so that it can be streamed and uncompressed small region by small region transparently to devs) and then you see that speed is one of the Many factors they spent transistors and R&D time on: custom SSD controller with DRAM cache and additional priority levels compared to NVME specs, custom I/O unit with custom do processors, dedicated DMA, dedicated low latency local memory pool, dedicated HW decompressor…
You keep focusing on a single element as if you had s gotcha, as if it was about how many GB can you transfer per second in a scene and how fast you can make smaller data transfer happen (how many ms to transfer X blocks of Y MB is a good question, how many parallel transfers, how can you suddenly get some small data chunk you have an immediate need of).
Then again people discussed it to death and you know it.
Cerny’s designs seem to have dual purposes: 1.) make it a lot easier to make games that take good use of the HW and 2.) if you invest your good time there are custom puzzles for you to solve to go even further. I think it is where the balance is with both the I/O stack and the custom Geometry Engine.
So, for some games, especially cross generation ones, the point is not how impossibly majestic they look, but how easy are they to make on PS5. Without wondering if you lost the tools or not.