Is it Better to Have Temporary Memory and Permanent Storage as Separate Components?

[DemonDragonJ]

Most computers have permanent storage and temporary memory as two separate components, and, in the past, the fact that these components did not operate at the same speed meant that the overall system was slower than it should be, but, now that solid-state storage is gaining popularity, that discrepancy is disappearing, making computers overall much faster and more efficient.

I wonder if, at some point in the future, temporary memory and permanent storage shall eventually merge into a single component, eliminating whatever delays remain in the computing process. That may seem to be beneficial, but I believe that it is better to keep them separate, and my reason for believing so is that temporary memory is cleared out and refreshed every time that a device is shut down or restarted, which means that any errors or corrupt data that is not in permanent storage shall be erased, but, without temporary memory, that data shall remain until it is manually expunged by the user, or perhaps by automated software.

What does everyone else say about this? Is it better to have temporary memory and permanent storage as separate components, or is it better to merge them?

[Airbozo]

I agree with having separate memory and storage systems and the technology will dictate the same setup for some time to come. There is storage technology on the market today that allows you to plug in SSD type storage into the memory bus. Right now they include their own battery subsystem to keep the contents stable when there is no power. The bus between the memory sockets and the CPU is blazing fast when compared to even the most recent storage technology, nvme (which uses the PCIe lanes). As processors advance, expect to see less SATA and USB (or other slow bandwidth interfaces) and more pcie lanes. Also expect to see more memory bandwidth. The interface between the on-cpu memory controller and memory bus will also continue to speed up.

I have been to some Intel briefings where they have even talked about co-processors again. Part of the idea is to move the interface or signals closer to the physical ports to reduce delay. most dual socket server motherboards perform better because the PCIe lanes and memory buses are split between the 2 cpu's. Imagine adding in a co-processor that deal only with memory or pcie and communicate directly with the host cpu via the qpi link (which is blazingly fast). This would allow the CPU to use more clock cycles for computing and less for communications much like dedicated network adapters off load any processing from the cpu.

[Konrad]

Sorry, I'm a little late on this one, lol.

The ideal solution would be a CPU die packed full of nonvolatile cache/RAM/storage memories. In reality things need to be spread out across discrete components because technology (and costs) prohibits this sort of computing storage density.

But @Airbozo - doesn't Intel already build these "co-processors"? Specifically, doesn't the PCH (high speed "Northbridge") component on the motherboard chipset already do what you describe? Many of Intel's processors and sockets and chipsets are modular and somewhat interchangeable, but in reality the PCH is just as critical and complex as the CPU, they're designed together to produce a working platform. The more stuff you pack onto the processor die the more complex, more expensive, and more hot it'll get, plus the PCH component enjoys physical proximity (lower latency) to the busses it serves. I'm imagining a your "Intel co-processor" basically just turning into a multi-processor multi-chipset motherboard ... with advantages and costs which would limit it to niche applications (since it would still be more affordable to scale up by simply doubling the number of motherboards, no?).

[richardlawson1489]

I agree. Separate is the best option.