Dump question

[joek71]

I ran a test for memory.
And it gave me these results.
NMI: Parity Check \ Memory Parity Error

What does this mean?

Specs:
Abit KG7-R
AMD 1.33 FSB266
Mushkin Ram- (2) Sticks 512DDRam each
NViDia GeForce3 Ti 20
Philips Sound - 760
2 Mactor 30 Gig HardDrives.

What does parity mean? Is there a problem with my ram?
Please advise.

Thanks in advance,
Joe

[KMS]

hiya parity errors normally indicate bad memory, mismatched memory, and a few other things .... rage ati fury cards used to cause them when the wrong drivers were installed. What are you doing the check with ?? software i assume as apposed to a system parity error ?

Heres what parity checkings all about :
A parity check is done using a single bit error correction code. To have such a code, one extra bit needs to be added to the data bits. It parity is used at the byte level, the for each 8 data bits a ninth bit is required. This extra bit can be set as the odd or even result of the addition of the other 8 bits. When the nine bits are read back, the same calculation is made and the parity bit which is read back should equal the calculation. If not, at least one bit has been altered.

Parity is an easy but not leak-tight way of checking data integrity. It has the advantage of requiring little electronics and requirinf only one extra bit. Parity was used early on in mainframe systems to enhance reliability.

There was a time that computer electronics were so unreliable that often the computer system broke down before the calculation had com to an end. In the era of vacuum tube computers, there was a bunch of scientist whose function was merely to calculate the odds a certain program would run to completion without the system to fail. If the system failed, this had to be detected and parity was a convenient way.

When IBM introduced the PC the parity technology was partly copied from the mainframe designs. However, there has never been an operationg system for the PC to support it. When a memory parity error occurs in a PC, the CPU receives an interrupt. Most OS's merely show a message on screen that the error occured an that is it.

In modern mainframe systems, the software is usually able to recover fully from a memory parity error because often true virtual memory management is used where there is a fat change there is still valid data on disk. Also, if memory protection is done in a professional way, only one user or task needs to suffer from the error. Moreover, a good OS is capable to re-launch the task which caused the fault.

Since the parity check has never been of much use in a PC computer manufacturers started to abandon the technology in the early nineties. This was also a time that memory was relatively expensive and competition was gearing up. Systems not requiring parity memory could be sold cheaper with little drawbacks for the end user.

This was a time, however, that memory size was not too big and memory speeds were slow, with often tens of nanoseconds margin on the timing. Today all this has changed. New memories call for a type of error protection better than parity because of other reasons those of which the original mainframes suffered. Today ECC technology has been adapted to mainframes and this technology is being introduced in personal computers.

[tripodal]

ok, now that ive found the resident memory expert, perhaps i could tag on another question.

What are the differences between buffered/unbuffered registered/unregistered memory?

im just curious so please dont go into such great detail as before, dont want to use too much of your time.

[Gibbon]

Hi tripodal,

Buffered modules contain a buffer (usually when there's a lot of memory) to help the chip set cope with the large electrical load required. The buffer electrically isolates the memory from the controller to minimize the load that the chipset sees.

Registered modules contain a register that delays all information transferred to the module by one clock cycle. Again, this is usually done on modules with a lot of memory to help ensure that the data is properly handled.

Most buffered and registered modules also have ECC and are used in servers. FPM and EDO modules are either unbuffered or buffered. SDRAM modules are either unbuffered or registered.

Hope that helps ;)

Gibbon

[tripodal]

wow, thank you very much. so im assuming that with registered memory the performance can be slightly less than non registered?

[Gibbon]

Yeah thats right m8, the act of using the register requires a clock cycle of the module.

However it does allow large amounts of memory to be used, so is suitable for servers, etc...

Gibbon

[tripodal]

thanks again for straightening me out