|
Non-parity versus ECC modules
Standard memory, also called non-parity memory, uses 8 bits to store an 8-bit byte. ECC memory (Error Correcting Code memory), sometimes called parity memory, uses 9 bits to store an 8-bit byte. The extra bit provided by ECC memory is used to store error detection and correction information.
A non-parity memory module can neither detect nor correct errors. An ECC memory module can detect all multi-bit errors, correct all single-bit errors, and correct some multi-bit errors. Memory errors are so rare that most desktop systems use non-parity memory, which is less expensive and faster than ECC memory. In fact, most desktop chipsets do not support ECC memory. If you install ECC memory in such a system, it may not recognize the memory at all, but more likely it will simply treat the ECC memory as non-parity memory, ignoring the extra bit.
ECC memory is occasionally used in desktop systems, but is much more common in servers and other large, critical systems. Because ECC modules contain additional memory chips, in the ratio of 9:8, they typically cost 10% to 15% more than similar non-parity modules. Also, because the circuitry on ECC modules that calculates and stores ECC values imposes some overhead, ECC modules are marginally slower than similar non-parity modules.
Unbuffered versus registered modules
Unbuffered memory modules allow the memory controller to interact directly with the memory chips on the module. Registered memory (also called buffered memory) modules place an additional layer of circuitry between the memory controller and the memory chips.
Registered memory is necessary in some environments, because all memory controllers have limitations on how many devices (individual memory chips) they can control, which in turn limits the maximum capacity of the memory modules they can use. When a memory controller interacts with an unbuffered memory module, it controls every memory chip directly. When a memory controller interacts with a registered memory module, it works only with the buffer circuitry; the actual memory chips are invisible to it.
The sole advantage of registered memory is that it permits a system to use higher-capacity memory modules. (The highest capacity modules at any given time are often available only in registered form.) The disadvantages of registered memory are that it is considerably more expensive than unbuffered memory and noticeably slower because of the additional overhead incurred from using a buffer.
Most desktop systems support only unbuffered memory modules. A few can use either registered or unbuffered memory modules. A very few desktop systems notably, Socket 940 AMD models require registered memory. If you are upgrading the memory in your system, we recommend that you use registered memory modules only in the following situations:
The system accepts only registered memory modules.
The system accepts unbuffered or registered memory modules, but registered modules are already installed.
The amount of memory you want to install requires using modules that are available only in registered form.
EITHER/OR
Most systems that support unbuffered and registered modules can use only one or the other at a time.
|
