What prevents you from making a computer without RAM?

Question

Hard disks

Based on the news from Intel, together with Micron Technology, already this year... .
If we have such a fast large non-volatile and, in the long term, cheap memory, do we need to separate RAM and hard drives? Is it possible to combine architecturally these two computer devices?
This will mean that it will not be necessary to load the program or data from disk into RAM, but you can, for example, work directly in this memory.
This means that the devices will instantly turn on, because there will be no such stage as loading the system. After all, memory is non-volatile: where the system stopped, it continued from there.
What am I not considering? What are the restrictions? Who has any thoughts?
Is there enough speed? processor speed? what needs to be added in operating systems or do you need to redo everything in general?
UPD: PS The question is not specifically about this memory from the news. But in principle. Here we have a fast, capacious, non-volatile, cheap memory with a good service life. What will it lead to in technical terms?
UPD2: Colleagues, I propose to discuss the technical aspect. How should the OS develop if there is a large non-volatile RAM? Now I’m not particularly interested in whether this technology will take off / will not take off, or some of its shortcomings .... the future will show. I'm wondering how the OS and software will be arranged in such a computer.

Answer 1

Read about the advantages and disadvantages of eternal memory in the criticism of the Phantom OS.

Answer 2

DDR 3 throughput is about 20 gigabits, SSD throughput in random mode is 1.8 gigabits, 10 times less.
And yet, any flash memory will wear out, and without relocation and garbage collection, it does it pretty quickly.

Answer 3

The correct answer was given to you by Armenian Radio - this is the WEAR of the cells.
How long does Flash memory last?
SLC < 100k cycles
MLC < 10k cycles
eMLC < 40k cycles
TLC < 5k cycles.
Now let's calculate how long the "New" memory from Intel(r) lives.
Let's even take as a basis, those same 100k cycles. And we get about 100M cycles. Okay...
Now... put in this memory a stack that is updated on EVERY procedure call, and divided for each application, and so on. On average, a subroutine call in modern programs is from 10 to 100 instructions. Let the program hang in memory and work quite rarely, but make obligatory OS calls, respond to them, return data through buffers on the stack ... and so on 50-100 times per second.
To erase a page, again, you need to put its data somewhere (and we have NO RAM, there is only this new memory). Total - 2 rewrites per operation. And if some program starts to count the index of the array in the variable? It must be stored somewhere in RAM (but we don't have it). How long will the unfortunate sector live? Say, at a standard clock frequency of 100 MHz? About 1 second of continuous cell changes.
Pretty annoying, right?
And RAM is not subject to wear. At least not so fast.
I hope I drew the analogy accurately enough?

Answer 4

As soon as it becomes economically justified, then they will do it.