Thread: ROM as memory, pipeline questions

Could ROM be used to store the instructions of an OS?
Would it noticably increase speed?
Would it be cheaper?
Would it be much smaller?

Here are some questions I have about pipelines.
Are pipelines simply hierarchical memory?
Are pipelines only useful to reduce cost?
Why not just load a whole bunch of data onto a single bus? You could use a register as wide as the RAM, say a megabyte. Then, select one 16th, and output that to another circuit which selects one 16th, and continue until you are left with 16 bits. (It would take 5 of those steps.) You could use
2-d addressing/muxing for the selection circuits.

Originally Posted by NNet

Could ROM be used to store the instructions of an OS?

Yes it could. And this was commonly done before the development of cheap disk drives and in embedded systems (which, nowadays, tend to use flash as ROM).

Would it noticably increase speed?
Would it be cheaper?
Would it be much smaller?

Compared to what?
Probably faster and smaller than disk. Not sure about the cost per MB.
Similar density to DRAM. Faster than DRAM. Again, not sure about the cost; probably similar.

Are pipelines simply hierarchical memory?
Are pipelines only useful to reduce cost?

Not quite. Pipelining allows a number of different operations to occur in parallel, with each one typically at a different stage of the operation (row address decode, data selection, transfer data to bus, transfer data across bus to processor, etc). It is used for a number of different purposes such as increasing clock speed and hiding latency; the latter is more significant in memory design.

Why not just load a whole bunch of data onto a single bus? You could use a register as wide as the RAM, say a megabyte. Then, select one 16th, and output that to another circuit which selects one 16th, and continue until you are left with 16 bits. (It would take 5 of those steps.) You could use
2-d addressing/muxing for the selection circuits.

This is typically what happens inside large memory devices (especially DRAMs). The problem doing it outside the chip is that pins cost money and it takes more power to drive more signal lines. It is also harder to design a circuit board with very large numbers of signals. However, this approach is being adopted for the interface between memory and processor in things like 3D memory (see the Hybrid Memory Cube, for example) which allow tight integration of memory and logic, almost as if they were on one chip.

Question. Sorry my paragraph return doesn't work in here for some reason..... Is it worth to put Windows 7 on a Vista Machine?