RAM speed

The final part of the specification of a memory upgrade is the speed of the devices needed. As already  describe in the previous chapter, access speed is measured  in nano-seconds or ns. The slowest devices have  access times of 150ns to 120ns. Medium speed devices are 100ns and fast access chips are 80ns, 65ns and 60ns. In most cases it isn’t possible to look at the design of a machine and easily deduce the speed of RAM chips needed. Indeed some systems will allow  you to use one  of a number of speed by  imposing additional  wait states for slower chips. If machine uses  page mode memory  access then it may  even be more  stringent in the type of chips  it uses than  simply specifying an overall access speed.

The simplest solution to finding out what type  of chips  your machine needs it to look in the manual! Failing this you cloud open the case and look for the area  where the existing  memory is installed  and read the device  code on one of the  chips – see Reading the chips. You can usually recognize where the memory  device are either  because they will be the only SIP/SIMM device or because they will be  arranged in a regular rectangular array of identical  DIL chips.

It doesn’t really matter if you use  chips that are faster than your machine  needs. They won’t  male it work any faster and they will cost more but at least they will work.  This is one solution if you simply  cannot find out what speed  of memory device to use. On the order hand, Using chips that are slower than your  machine needs will cause memory errors to be reported during the power on  Self Test (POST) routine . It is even possible that if the chips  are only a little  slower than your machine needs then they will pass the POST routine but fail intermittently later on when your machine has  warmed up a  little. (RAM chips are  more tolerant of being worked faster when they are cooler.)

Most machines have  to be informed , either via jumpers, dip switches or software setup, if what speed  chips you are using  so that they can the  appropriate number of wait states. Once again introducing more wait states than necessary will

Reading and chips

One f the most  Intermediating   aspect of trying to buy extra RAM in the way that chips,  SIPs and SIMMs are described  in catalogues an manuals. Part of the problem is due to each manufacturer assigning their one product codes to each type of chip they produce. The assignment is fairly  arbitrary and so  you shouldn’t expect too much sense in the sort of numbers marked on a chip but they can provide a general guidelines to the chip’s type.
The first part of a chip’s part number usually  indicated the amount and organization of the memory  that the chip provides. The same amount of memory can be organized in many different ways. For example a chip that stores 1Mbit  can be arranged to provide  128Kx8bits (i.e 128Kbytes), 256Kx4bits or 1Mx1bit of storage. As most computers can only work with  memory locations that can store a complete byte, The organization of the chip indicates the smallest number that can be used to increase memory  capacity. For example , You could use one 128Kx8bit chip to increase memory by 128Kbytes,  but you would need at least two 256Kx4bit  chips (i.e 256KBytes) and eight Mx1bit chips (i.e 1 Mbyte). (The ignores the complication of needing a parity bit and memory banks, see later.)
The final digits of chip’s part number generally give its speed in nano-seconds  but often leaving out or adding in extra zeros . For example, The product code for an 80ns chip might end in -80, -08 or just -8 as the manufactures choose!
Often you will find chips listed simply by their organization and speed. For example , 256Kx4 120ns DRAM, 1Mx9 80ns SIMM 256Kx1 100ns SIP ect.. On other occasions you will find their full part numbers quoted. For example , a 41256-80 is a 256Kx1bit 80ns DRAM, 41464-12 is a 64Kx4bit 120ns  DRAM and a p21010-08 is a 1Mx1bit 80ns DRAM. Form these examples you can see that there  is some connection between part numbers and chip types -  but not enough to be certain without looking them up! If you are at all in doubt about the type of memory that you machine needs then check its  manual or contact one of the specialist memory supplies listed at the end of this all posts.

parity

The first personal computers were mainly used for recreation and the consequences of any undetected error was slight. However, if a machine is used for business or any serious  purpose it has to incorporate some method of error detection. The simplest and most commonly used method is party checking. This involves adding and extra  bit to every byte of data stored so as to make the number of 1 bits even. This a called even parity checking as opposed to making the total number of 1 bits odd i.e odd parity For example, if the data is 011101100 then the parity bit is 1 because there are five 1s in the data and making the parity bit 1 makes the total six which is even. If the data is 10100000 then the parity bit is 0 because the total number of 1s is already even. Each time data is read from memory is parity is checked. If a single bit has changed, either a 0 turning in to 1 or a 1 into a 0, since the data was stored  then the total number of 1  bits will not be even  and a parity error will be detected

The problem with parity checking is that while it is certain to detect  a change in a single bit ,if two bits change then this  leaves the total  number of 1 bits even and a parity error will nit occur. However, in modern computer  systems the probability of a single bit error is so small that the probability of two such  errors in the same  memory location a vanishingly small. Indeed, such errors are so rate that it is arguable that parity checking itself is redundant.

As is takes eight bits to store a single byte, adding an extra bit for parity checking makes the total number of bits need equal to nine. This is the reason that 386/486  systems need nine  1MByte chips or 9xMByte SIPs or SIMMs for every Mbytes of memory you also have an additional 1MByte dedicated to parity checking! In same machines you can disable parity checking and make use of the unused chips to increase the amount of memory available. As transient memory errors are very rare, and the memory is  tested for permanent faults every time you switch the machine on,  you might consider this a reasonable trade-off.