little study about memory

1. Sort of. The microprocessor cache architecture is coherent with respect to the contents of the system's address space. This means that the contents of the cache will always be a subset of the main memory. Whenever the value at some memory address is changed it is first changed in the microprocessor's local copy (in the cache), and this change is then written back to the memory some time later when it is appropriate to do so. It also means that whenever the value at some address is loaded into the microprocessor from memory, the value is first copied into the cache.

Problems can occur when the value at a memory location is changed outside the scope of a running task. A good example of this is when an interrupt occurs and changes a value. The running task which was interrupted is not made aware that the value has changed and while any cached copies of the address will be updated, any copies of that data that's stored in CPU registers or on the program stack will not be. This problem occurs quite often in parallel programming.

2. The RAM does not access the HDD directly, but a program (such as GPU and storage drivers) can take advantage of the platform's DMA controller to offload data movement operations between primary memory and peripherals, or between primary memory and secondary storage. The DMA controller is a non-turing-complete co-processor which is purpose built to offload data transfer operations from the CPU. The DMA controller is given control of the necessary system buses when the CPU is busy doing other things, which allows for bulk data transfers to occur between peripherals and system memory without taking up valuable CPU time.

3. It depends. There is little about most microprocessors that explicitly require SDRAM to operate. In fact, one of the reasons why system firmware (usually conforming to the BIOS or UEFI specifications) is so minimalist and compact is that it has to fit on a small EEPROM chip and has to be able to fit the necessary elements entirely in the CPU cache until the SDRAM has been configured and initialized. Early PCs had the firmware on a floppy disk instead, which was eventually moved to the EEPROM and Flash chips that we have now. So it's only a small stretch to believe that a PC could run all program code from a slow secondary storage device such as a hard disk without any sort of primary memory except the CPU cache. While it may be possible, it would be extremely ill advised as hard disks are cripplingly slow compared to SDRAM. Furthermore, it may not be possible to put x86 CPUs from Intel or AMD into higher modes of operation without SDRAM ; this is not the case universally however.