ROM

This is read-only memory, memory that can only be read, but cannot be written to. ROM is used in situations where the data must be held permanently. This is due to the fact that it is non-volatile memory. This means the data is “hard-wired” into the ROM chip. You can store the chip forever and the data will always be there. Besides, the data is very secure. The BIOS is stored on ROM because the user cannot disrupt the information.

There are different types of ROM, too:
Programmable ROM(PROM). This is basically a blank ROM chip that can be written to, but only once. It is much like a CD-R drive that burns the data into the CD. Some companies use special machinery to write PROMs for special purposes.
Erasable Programmable ROM (EPROM). This is just like PROM, except that you can erase the ROM by shining a special ultra-violet light into a sensor atop the ROM chip for a certain amount of time. Doing this wipes the data out, allowing it to be rewritten.
Electrically Erasable Programmable ROM (EEPROM). Also called flash BIOS. This ROM can be rewritten through the use of a special software program. Flash BIOS operates this way, allowing users to upgrade their BIOS.

ROM is slower than RAM, which is why some try to shadow it to increase speed.

RAM

Random Access Memory (RAM) is what most of us think of when we hear the word memory associated with computers. It is volatile memory, meaning all data is lost when power is turned off. The RAM is used for temporary storage of program data, allowing performance to be optimum.

Like ROM, there are different types of RAM:

Static RAM (SRAM)

. This RAM will maintain it’s data as long as power is provided to the memory chips. It does not need to be re-written periodically. In fact, the only time the data on the memory is refreshed or changed is when an actual write command is executed. SRAM is very fast, but is much more expensive than DRAM. SRAM is often used as cache memory due to its speed.

There are a few types of SRAM:

Async SRAM

An older type of SRAM used in many PC’s for L2 cache. It is asynchronous, meaning that it works independently of the system clock. This means that the CPU found itself waiting for info from the L2 cache.
Sync SRAM. This type of SRAM is synchronous, meaning it is synchronized with the system clock. While this speeds it up, it makes it rather expensive at the same time.
Pipeline Burst SRAM. Commonly used. SRAM requests are pipelined, meaning larger packets of data re sent to the memory at once, and acted on very quickly. This breed of SRAM can operate at bus speeds higher than 66MHz, so is often used.

Dynamic RAM (DRAM)

DRAM, unlike SRAM, must be continually re-written in order for it to maintain its data. This is done by placing the memory on a refresh circuit that re-writes the data several hundred time per second. DRAM is used for most system memory because it is cheap and small.

There are several types of DRAM, complicating the memory scene even more:

Fast Page Mode DRAM (FPM DRAM)

FPM DRAM is only slightly faster than regular DRAM. Before there was EDO RAM, FPM RAM was the main type used in PC’s. It is pretty slow stuff, with an access time of 120 ns. It was eventually tweaked to 60 ns, but FPM was still too slow to work on the 66MHz system bus. For this reason, FPM RAM was replaced by EDO RAM. FPM RAM is not much used today due to its slow speed, but is almost universally supported.

Extended Data Out DRAM (EDO DRAM)

EDO memory incorporates yet another tweak in the method of access. It allows one access to begin while another is being completed. While this might sound ingenious, the performance increase over FPM DRAM is only around 30%. EDO DRAM must be properly supported by the chipset. EDO RAM comes on a SIMM. EDO RAM cannot operate on a bus speed faster than 66MHz, so, with the increasing use of higher bus speeds, EDO RAM has taken the path of FPM RAM.

Burst EDO DRAM (BEDO DRAM)

Original EDO RAM was too slow for the newer systems coming out at the time. Therefore, a new method of memory access had to be developed to speed up the memory. Bursting was the method devised. This means that larger blocks of data were sent to the memory at a time, and each “block” of data not only carried the memory address of the immediate page, but info on the next several pages. Therefore, the next few accesses would not experience any delays due to the preceding memory requests. This technology increases EDO RAM speed up to around 10 ns, but it did not give it the ability to operate stably at bus speeds over 66MHz. BEDO RAM was an effort to make EDO RAM compete with SDRAM.

Synchronous DRAM (SDRAM)

SDRAM became the new standard after EDO bit the dust. Its speed is synchronous, meaning that it is directly dependent on the clock speed of the entire system. Standard SDRAM can handle higher bus speeds. In theory, it could operate at up to 100MHz, although it was found that many other variable factors went into whether or not it could stabily do so. The actual speed capacity of the module depended on the actual memory chips as well as design factors in the memory PCB itself.

To get around the variability, Intel created the PC100 standard. The PC100 standard ensures compatibility of SDRAM subsystems with Intel’s 100MHz FSB processors. The new design, production, and test requirements created challenges for semiconductor companies and memory module suppliers. Each PC100 SDRAM module required key attributes to guarantee full compliance, such as the use of 8ns DRAM components (chips) that are capable of operating at 125MHz. This provided a margin of safety in ensuring that that the memory module could run at PC100 speeds. Additionally, SDRAM chips must be used in conjunction with a correctly programmed EEPROM on a properly designed printed circuit board. The shorter the distance the signal needs to travel, the faster it runs. For this reason, there were additional layers of internal circuitry on PC100 modules.

As PC speeds increased, the same problem was encountered for the 133 MHz bus, so the PC133 standard was developed.

RAMBus DRAM (RDRAM)

Developed by Rambus, Inc. and endorsed by Intel as the chosen successor to SDRAM. RDRAM narrows the memory bus to 16-bit and runs at up to 800 MHz. Since this narrow bus takes up less space on the board, systems can get more speed by running multiple channels in parallel. Despite the speed, RDRAM has had a tough time taking off in the market because of compatibility and timing issues. Heat is also an issue, but RDRAM has heatsinks to dissipate this. Cost is a major issue with RDRAM, with manufacturers needing to make major facility changes to make it and the product cost to consumers being too high for people to swallow.

DDR-SDRAM (DDR)

This type of memory is the natural evolution from SDRAM and most manufacturers prefer this to Rambus because not much needs to be changed to make it. Also, memory makers are free to manufacture it because it is an open standard, whereas they would have to pay license fees to Rambus, Inc. in order make RDRAM. DDR stands for Double Data Rate. DDR shuffles data over the bus over both the rise and fall of the clock cycle, effectively doubling the speed over that of standard SDRAM. Due to its advantages over RDRAM, DDR-SDRAM support was implemented by almost all major chipset manufacturers, and quickly became the new memory standard for the majority of PC’s. Speeds ranged from 100mhz DDR (with operating speed of 200MHz), or pc1600 DDR-SDRAM, all the way to current rates of 200mhz DDR (with operating speed of 400MHz), or pc3200 DDR-SDRAM. Some memory manufactures produce even faster DDR-SDRAM memory modules which readily appeal to the overclocker crowd.

DDR-SDRAM 2 (DDR2)

The latest DDR-SDRAM technology to hit the market for PC’s has become known simply as DDR-SDRAM 2 or DDR2. It features several advantages over conventional DDR-SDRAM (DDR), with the main one being that in each memory cycle DDR2 now transmits for 4 bits of information from logical (internal) memory to the I/O buffers. standard DDR-SDRAM only transmits 2 bits of information each memory cycle. Because of this, normal DDR-SDRAM requires the internal memory and I/O buffers to both operate at 200MHz to reach a total external operating speed of 400MHz. Due to DDR2’s ability to transmit twice as many bits per cycle from logical (internal) memory to the I/O buffers (this technology is formally known as 4 bit prefetch), the internal memory speed can actually run at 100MHz instead of 200MHz, and the total external operating speed will still be 400MHz. Mainly what all this comes down to is that DDR-SDRAM 2 will be able to operate at higher total operating frequencies thanks to its 4 bit prefetch technology (e.g. a 200mhz internal memory speed would yield a total external operating speed of 800mhz!) than DDR-SDRAM. Currently, this is the memory standard on most new motherboards.

Computer memory is one of those areas that causes quite a bit of unnecessary confusion for newbies. Memory is a fairly straight-forward thing and, for the most part, there is not a whole lot to consider. There are, though, a few terms and concepts that are thrown around, and I will seek to explain them here.

Memory Speed

There are various specs used to specify the speed of a memory module. Older memory types have their speeds measured in terms of access time, or the time it takes to pull a piece of data from memory and get it onto the memory bus. This is usually measured it terms of nanoseconds (ns). A typical access time would be 50-60 ns. Beginning with SDRAM, speed was measured in terms of cycle time, or the minimum amount of time between memory accesses. Typical cycle times for SDRAM are 12, 10 and 8 ns, with faster speeds also seen.

A wait state is a command which orders the CPU to pause for one clock cycle in order to wait for the memory to do something. It is basically a waste of time, because the processor is just sitting there and doing nothing. Older systems sometimes used 2 or 3 wait states, while later ones used only 1. A zero wait state is best, because that means the processor is not waiting around for the slower memory. The concept of wait state was devised to allow systems to support older memory types. Most systems that use wait states (except the really old ones) have the wait state setting controlled in the motherboard’s CMOS, usually in the advanced settings section.

SDRAM is synchronous, meaning it is tied into the bus speed of the system. This means that the memory must be fast enough to work on the system you intend to put it on. SDRAM does not use wait states for this reason. The memory, then, must be fast enough for the system, taking slack into account. It is really this reason why SDRAM was created in the first place: to make memory that could keep up with the system. For older systems, EDO RAM does just fine. At the 66MHz speed, EDO was a dream, as that is what it was really designed for. It was soon found that EDO RAM worked just fine at even higher speeds, such as 75MHz or 83MHz. SDRAM was designed mainly to operate with stability at bus speeds such as 100MHz or 133MHz. A majority of users today are making decisions between PC-100 and PC-133 memory. In short, PC-100 is designed for a 100MHz bus speed while, obviously, PC-133 is designed for the 133MHz bus speed. Both modules can typically be ran faster than this in overclocking situations, but this can lead to unstability of the PC. So, these ratings are considered the “stable” rating, or the speed at which the memory can run without any stability concerns.

Determining Memory Speed

It can prove challenging to determine the speed of a memory module simply by looking at it. Unless there is a sticker of some sort on it from the distributor, you are pretty much stuck with cross-referencing part numbers with catalogs to determine just how fast a module is supposed to be. Sometimes, though, you can determine the access time or cycle time (depending on the memory types) of a module by looking at the numbers on it. For example, on EDO memory, adding a -6 or a -7 to the end of the part number usually means 60ns and 70 ns access times respectively. Wiuth SDRAM, a -10 means 10ns cycle time.

Nanosecond were used before the PC100 standard was defined. With the standard, you pretty much knew what you were getting with a particular spec, so tracking the actual cycle times was irrelevant.

CAS Latency

CAS Latency is a measure of latency of a memory chip. CAS stands for Column Access Select. Basically, it is a measure of how long it takes from when an initial READ command is sent to memory to when the first piece of the resulting data is output. The measurement is done in clock cycles, so a CAS Latency of x means that a READ command sent to memory at clock cycle c will result in data output starting at clock cycle c + x.

CAS Latency is very closely tied with the system bus speed you are using. On faster bus speeds, data is flying by the memory at a faster rate. This pushes the memory even harder. But, as mentioned above, all memory chips have an access time, even SDRAM. But, that “S” in SDRAM means that is is synchronous, running at the speed of the system bus. So, regardless of the access time of the memory, the bus pushes even harder at high bus speeds. So, at higher bus speeds, it may be necessary to make use of wait states anyway in order to make the memory is capable of operating. But, wait states, on SDRAM, are done in the RAM itself rather than actually having the CPU waiting around. So, it may be necessary to increase the CAS latency on higher bus speeds in order to maintain system stability. This will control how often the CPU will deliver a command to make the memory wait the specified number of clock cycles for the memory to begin output.

In short, CAS3 is the standard latency for memory modules, because it is cheaper to manufacture. If the CAS spec is not mentioned or defined, it is probably CAS3. CAS2, though, is a faster memory module. The latency is less, and this leads to faster application speeds. As you might expect, such memory is typically more expensive. If you are using CAS2 memory and it is enabled as such in the BIOS, you might notice a bit of a speed increase. Even with CAS2 memory, though, it could be necessary to choose CAS3 in your CMOS in order to make the memory stable in a higher bus-speed compter. 

2-clock vs. 4-Clock

Two types of SDRAMs are the 2-clock and the 4-clock. Structurally, they are the same, but they are accessed differently. A 2-clock SDRAM module is set up so that each clock cycle accesses two chips on the module. A 4-clock SDRAM setup accesses 4 chips per clock cycle. To choose what kind to get, you must look into the motherboard’s documentation. 4-clock modules seem to be the popular choice.

Serial Presence Detect

Some SDRAM modules have a special EEPROM chip on it that holds information about the SDRAM module, such as speed settings. The motherboard then queries this chip for info and makes changes in the settings to work with the SDRAM. Basically, this allows the SDRAM module and the chipset to communicate, making the SDRAM more reliable on a larger number of motherboards. Some motherboards require this feature. You will have to look at the manual, once again. If your board requires it, make sure you have it, because SDRAM without this won’t work.

When choosing SDRAM for your computer, you need to know your motherboard and get exactly the type it requires.

Price, quality, and quantity… three very important factors that guide you in the selection of your computer system\’s physical memory. The \”price\” part is now at an all time low. Just checking at www.pricewatch.com, I found a 256 meg stick of generic memory for all of $16.00. You can\’t get a better \”bang-for-your-buck\” product that will provide your computer with as much of a dramatic increase in system performance as an upgrade in RAM will. Now the big question… Who to buy from?

Two highly recognized names come to mind -  Crucial and Mushkin.

Click to see Picture of Modules

Crucial has long been regarded as \”the memory experts\”.  It\’s hard to dispute this, as they are the largest DRAM manufacturer in America and one of the top three in the world. A division of Micron, Crucial offers over 73,000 upgrades for more than 13,000 different computers.

Mushkin, on the other hand has always been a favorite for overclockers and their memory needs. Mushkin hand picks their memory chips and selects only the very best to put on their modules so that you are guaranteed to get the performance printed on the module - and then some. Mushkin is focused more on quality than quantity.  It is reflected in the high performance of their memory modules.

This review compares modules from both companies, testing their performance, tolerances, and price differences, ultimately showing that you get what you pay for.

Many of these actions involve making changes to the CONFIG.SYS file. Remember, you want to keep a backup of the last CONFIG.SYS that your system last worked correctly with. So, backup yours before making any of these changes. Most name their backup copy CONFIG.BAK.

Here are some things you could try:

• Thin out your AUTOEXEC.BAT and CONFIG.SYS file. Many times these files call up programs that are simply not needed or not there. These lines can be removed. Better yet, just add REM to the beginning of the line you want to take out. This makes it a \”remark\” and the computer will not execute that line.
  
• Use the HIMEM.SYS file. At the top of CONFIG.SYS, add two lines: \”DEVICE=C:WINDOWSHIMEM.SYS\” and \”DOS=HIGH,UMB\”. This will call up HIMEM.SYS, a program that loads DOS into high memory, or that first 64K of memory, that all DOS programs fight over.
  
• Use EMM386.SYS. This program enables DOS to load drivers and other automatically loaded programs into the upper memory while conserving conventional memory. To use it, add the following to CONFIG.SYS right after the HIMEM.SYS line: \”DEVICE=C:WINDOWSEMM386.EXE\”. There are a couple parameters you should add to this line. To disable expanded memory, which hardly anybody needs, add \”NOEMS\” to disable the EMS buffer. To disable the Monochrome Video Area, add \”I=B000-B7FF\”. This disables the monochrome area that is used by the really old DOS programs that were monochrome. Today, with everything in color, this is just a waste of 32K of conventional memory.
  
• With HIMEM.SYS and EMM386.SYS working, it gives you the ability to move drivers and programs that would usually reside in conventional memory up to the upper memory. To do this, you simply add HIGH to the lines loading up the drivers. For example, in the CONFIG.SYS file, a driver will be loaded by \”DEVICE=\”. To load this driver into upper memory, you call it up by \”DEVICEHIGH=\”. In AUTOEXEC.BAT, you can place a program in upper memory by adding \”LOADHIGH\” to the line that runs the program. Most drivers can be moved to upper memory, including CD-ROM, mouse, etc.

Taking this into account, a typical CONFIG.SYS may look something like this:

DEVICE=C:WINDOWSHIMEM.SYS
DEVICE=C:WINDOWSEMM386.EXE NOEMS
I=B000-B7FF

DEVICEHIGH={Your System Drivers Go Here}
DOS=HIGH,UMB
FILES=100
BUFFERS=40
BREAK=ON
LASTDRIVE=Z

I\’ll just make a note here that this page pretty much refers to Windows 3.x users. Windows 3.x still depends on the conventional memory. Windows 95 has the ability to take full advantage of all of the system memory, so these tips don\’t have much impact.

The packaging is simply the entire makeup of a unit of memory, such as the SIMM or DIMM. Since the memory chips themselves are way too small, they must be combined and put onto a medium that can be worked with and added to a system. So, designers took the memory chips, placed them on a small fiberglass card, and created the memory module.

There are several different package styles which one may see for memory:

• DIP (Dual In-Line Package) - This is the old classic “chip” package of memory modules, the kind with small pins undrneath that are plugged into pin sockets. While this design led to the ability to remove as required, it also led to the issue of broken memory pins. This type of package is only seen on old systems (such as those in the 286 era and before) and old video cards.


• SOJ (Small-Outline “J” lead) - This is a more modern type of package often found on memory SIMMs or sometimes with BIOS chips. It is similar to the DIP package, but is designed for surface mounting by having the leads protrude from the side of the package, but bent down under the package in the shape of a “j”.


• TSOP (Thin, Small-Outline Package) - This is also a surface mounting package, but it requires a very small space, and is thus used on PCMCIA cards, as well as in notebooks and on some video cards.


• BGA (Ball Grid Array) - A newer packaging method in which the chips are attached using small balls of solder underneath the chip. The advantage to the manufacrturer is that they are cheaper to make, allow more capacity, allow better heat dissipation and better electrical performance. There have been various variations of this packaging style: Fine-BGA, Tiny-BGA, etc. , all various advancements off of the BGA standard. This type of package is most common today in heavily-used memory modules, including Rambus.

Finally, engineers put the chips on SIMMs, or Single Inline Memory Modules, or DIMMs, Double Inline Memory Modules. SIMMs use a 32-bit memory bus whereas DIMMs use a 64-bit memory bus. These cards are placed in a socket on the motherboard, like a card in a slot, then latched in. This design eliminated problems of the past, and made upgrading memory a breeze due to providing standardization which other manufacturers could rely on.

SIMMs

SIMMs come in two sizes, 30-pin and 72-pin. The 30 pin SIMMs usually came with small amounts of memory (smaller than 8MB). They are not used now, being mainly used in earlier 486’s and older machines. The 72-pin SIMMs were much more popular, and were used on many motherboards until SDRAM came into the picture. Although you will occasionally see 72-pin SIMMS still in use, it is usually only if you are opening up an old system.

For pinout info: see 30-pin pinout or 72-pin pinout.

SIMMs come in both single sided and double sided designs. This refers to whether the SIMM has memory chips on one side of the SIMM or both. Usually, 1, 4, and 16MB SIMMs are single sided. Other sizes are double sided. Some double sided SIMMs are actually two single sided SIMMS back to back, where they are wired together within the fiberglass module. These designs operate a little different electrically, explaining why some boards only use SIMMs of certain sizes.

But, the world of SDRAM is not cut
and dry. Standard SDRAM is great for “older” boards. Now,
with the release of BX motherboards, and the Super 7 boards, standard
SDRAM begins to cause problems. Why? Because even though it was
originally said that SDRAM could go up to 100MHz, it really couldn’t.
In fact, some SDRAM even got unstable at the 83MHz bus speed.

Enter PC100. Basically, PC100 is
SDRAM which meets a certain specification to work with stability at
100MHz. This SDRAM usually operates at 10ns, although some is created
that is faster. Since the only qualification for PC100 is the ability
to operate at 100MHz, there is no rule as to the access time. 10ns is
the minimum speed for stability at 100MHz. some companies advertise
PC100 faster than this, say 6ns, but, a lot of times you will find
this to be inaccurate.

All PC100 is not equal. While it all
operates at 100MHz, when you get into higher bus speeds than that, the
high-quality stuff starts to stand out. The reason is that the latency
rating of the higher quality stuff is lower. The latency is a
measurement of how long it takes for other hardware to return data to
the RAM. The lower the latency rating, the better the chip, and the
faster it will operate.

The most common, and cheaper, type of
SDRAM chip uses GL or G8 chips. The “GL” or “G8″
will be seen on the actual SDRAM chips on the memory circuit board, so
you will know what you’re looking at. The GL’s use a CAS latency of 3,
which is pretty standard. The better stuff uses “GH” chips,
which uses a CAS latency of 2.

To operate at 100MHz or 112MHz bus
speeds, almost any of this PC100 will work. But, bump it up to 133MHz,
and you’ll need to get the better GH SDRAM with a CAS latency of 2.
Only with this will you get stable operation at such high front
side bus speeds.

Along with high quality PC100, one
must take notice of the printed circuit board on which the chips are
mounted. The quality of these boards, for the most part, is measured
in the amount of layers. You can equate this to thickness. Obviously,
the thicker the material of the board, the less chance you have of
damaging it, the longer it will last, and the less electrical noise
you will get. so, the more layers the better. Run-of-the-mill, cheap
SDRAM often used good quality chips, but the manufacturer would cut
corners by using lower quality PCB’s(Printed Circuit Boards). Often
they would use 4-layer PCB’s. Well, part of the PC100 spec is a
minimum of 6-layer PCB. this ensures a higher level of quality. But,
some manufacturers use even better PCB’s, such as 8-layer. Pay
attention to this. The more layers the better.

So, if you find yourself buying a
Super 7 or BX motherboard, you should pick up some PC100 SDRAM with
it. The older stuff will work, but, without PC100, you are stuck with
your new board’s slower bus speeds.

PC-133
Basically, PC133 SDRAM
is another implementation of the same old SDRAM.  It’s basically
the same SDRAM from the days of the LX Chipset, the Pentium II 333MHz
processor, and the 66MHz bus.  The only difference between PC133
SDRAM and the others, is that the PC133 has a lower latency
than PC100 and PC66 SDRAM, which means it can run on a faster bus.

If you don’t already know,
PC133 SDRAM can run stably on a 133MHz bus, just as PC100 ran stable
on the 100MHz bus, and PC66 ran stable on the 66MHz bus.  PC133
SDRAM increases the total bandwidth available to the processor from
the memory, because it runs faster.  That is because it raises
the speed limit, so to speak, on the road between the Processor and
the RAM.  

Sometimes it easy to think of
the lines data moves between two computer components as roads. 
The road between the SDRAM and a current processor, like the Pentium
III, is 64bit, which can be thought of as a 64 lane highway. 
With older PC100 SDRAM, the speed limit on that road was 100MHz, which
means that during a second, 100 million bits moved though each lane on
the highway.  That’s 6.4 Billion bits, and as we all know, 8 bits
= 1 byte.  That means, that with older PC100 SDRAM, the processor
could get a maximum of 800MB per second.  With PC133 SDRAM, the
speed of that road is increased to 133.33 million bits on each lane
per second.  That translates into 8.533 bill bits.  Using
the same math above, that means the processor could get a maximum of
1060 MB per second (1.06GB) from the SDRAM.

More data, of course, means
better performance.  Your games will run faster, business
applications load faster, and even Windows boots faster.  Only
problem is that the performance increase isn’t all that much, and most
of the time it’s hardly
noticeable.  Possibly with new types of SDRAM which are trying to
compete with RamBus RAM, users will see a
much higher performance increase.

Double Data Rate
Well, there really is not much to say on this topic,  because
the topic is rather cut and dried. DDR RAM is Normal SDRAM that sends data both on the rising of
the clock cycle, and the falling of the clock cycle.  

Twice the sending
of the data, twice the data sent. Where standard 100 MHz SDRAM has an estimated 800 MB/sec data
transfer rate for a theoretical maximum, DDR is, not surprisingly,
twice that. No, we
don’t actually see that much bandwidth, but that is their
theoretical maximum (64 bit X 100 MHz = 800 MB/s). DDR SDRAM would be 1600 MB/s. Its just faster, and it will be cheaper than Rambus RAM, and
its currently supported by quite a few motherboard manufacturers.

It’s called Rambus DRAM. 
It works on a pretty basic principle.  The more data the CPU gets
from RAM, the better.  Rambus DRAM achieves just that, by
increasing the bandwidth from the DRAM to the CPU.  They do this
in two ways.  One way is by increasing the speed of the RAM, as
they do with the CPU, called clock multiplying.  They basically
multiply the front side bus by a certain number to achieve the speed
of the Rambus DRAM.

There is a problem when it
comes to Rambus.  When Intel increased the bandwidth, they also
decreased the latency, the speed at which the DRAM can get data from
it’s inner most depths, and working toward the CPU.  Think of the
data going two and from the CPU as a Interstate Highway.  With
SDRAM, there are only 4 lanes, but the speed limit is 100km/hr. 
With RAMBUS, there are 8 lanes, but the speed limit is only
50km/hr.  In effect, on the average, RAMBUS doesn’t help, nor
hurt, performance at all.  There wouldn’t be a problem with that,
but there is.  The price for RAMBUS DRAM is approximately five to
ten times more expensive than SDRAM.

Intel’s new chipsets, that
support their newer Pentium iii Processors only support RAMBUS. 
But, there is a solution out there that will make your old SDRAM look
and act like RAMBUS to the computer.  It is a daughter card that
fits into the RIMM (stick of RAMBUS) slot.  Because it’s a data conversion
card, and not just an interface conversion card, it will slow down performance. 
Mainly, because of the cost, RAMBUS is falling out of favor with a lot
of people, causing them to flock for a chipset that support’s Intel’s
newer processors, the Apollo Pro 133A.

Memory installation is a pretty easy upgrade to perform. Most of the work, if you could call it that, comes before you actually do the upgrade - in being sure you get the right kind of memory for your system. For the purposes of this tutorial, we will assume you have already done this.

We will break the tutorial up for the different types of memory.

SIMMs (EDO)

First some info, though. Most computers that use EDO RAM have been retired in favor of newer technology.  If you are lucky enough to still own one, EDO RAM can be a bit trickier to install than more modern kinds of RAM.

Your computer, if it uses 30-pin or 72-pin SIMMs, organizes its SIMM sockets into groups called banks. Some boards say that two sockets make a bank. Some say that one is a bank. Nevertheless, a bank must be full. A half full bank will drive your computer nuts. Also, you can’t mix two different kinds of memory in a single bank. For example, you can’t put a 4MB SIMM and an 8MB SIMM in one bank and expect to get 12 MB of RAM. Also, many systems require you to put the memory in in pairs. Therefore, if you want 32 megs of RAM, you have to stick 2 16’s in instead of one 32.

Here’s a shortcut which is almost always true. An older computer with a 386 or an early 486 chip usually has a 4 socket bank of 30-pin SIMM modules. A later model 486 requires only one socket of 72-pin modules. Pentium machines have two socket banks of 72 pin modules, meaning you must install RAM in pairs. In all of these systems, the bank must be full for your system to operate. Following these guidelines, lets say you want to add 16 meg of RAM to your Pentium machine. You could buy one 16MB SIMM, but this won’t work because you will have a partially filled bank. You must buy two 8MB SIMMS instead, and install them in a pair.

1. Turn off the computer, unplug it, and take off the case cover.


2. Locate the memory slots.


3. Remove the old memory (if applicable). This will entail loosening the little retainer springs on each end of the memory socket until the memory stick can be removed.


4. Install the RAM. To do this, first locate the little notch on the pin-side of the module. This notch will line up with  a notch on the memory socket itself, to ensure proper alignment. Position the module over the slot at a 45-degree angle with the module pins in the slot. Gently rotate the SIMM until it is in an upright position. When it is in an upright position, the retainer springs will snap into place and secure the SIMM.


5. Repeat previous step for all remaining SIMMs you wish to install.


6. Test it. Before you put your case back on, power on your system and make sure it correctly tallies the RAM.


7. Close up the case.

DIMMs (SDRAM, DDR-SDRAM, DDR2)

Now, it’s fortunate that SDRAM and DDR came out, or we’d all have to decipher all that bank stuff in order to upgrade memory. DIMMs are much simpler. Most older motherboards have the 168-pin slots for SDRAM. Most newer motherboards use 184-pin slots for DDR or 240-pin slots for DDR2. Each DIMM slot is a bank, so one can install these types of memory in any combination they want. It is best, if you have several open DIMM slots, to use the lowest number slots first. The slots are numbered, such as DIMM 0, DIMM 1, and DIMM 2. Choose the lesser, unused number. If this is the only module in the system, use DIMM 0.

Some motherboards that support DDR or DDR2 can utilize Dual-Channel DDR.  Dual-Channel DDR allows two sticks of DDR RAM to work together to effectively double the available bandwidth.  Performance gains usually range from 15%-20% over standard single channel setups.  To enable it, you will need to make sure that both DIMMs are identical.  Second, you will have to check with your motherboard manual to find into which slots the sticks need to be installed.  Typically, bank 0 and 3, or bank 2 and 3 are used. If everything matches up and is installed correctly, a message should appear on the POST screen when you boot up your computer confirming that Dual-Channel is enabled.

Some older motherboards have both SIMM and DIMM slots. On these, each DIMM slot is a bank, just like normal. The SIMM slots right next to them are usually paired in a bank, just like the normal Pentium bank setup.

1. Turn off the computer, unplug it, and take off the case cover.


2. Locate the memory slots.


3. Remove the old memory (if applicable). This will entail pressing down on the little ejector clips on each end of the memory socket until the memory stick pops out of the socket. Then you just life it out.


4. Install the RAM. To do this, first locate the little notches on the pin-side of the module. These notches (usually two) will line up with keys on the memory socket itself, to ensure proper alignment. With the ejector clips in the open position, position the module over the slot and begin pressing the module down into the slot. You will need to press down pretty hard. As you press down, the module will sink into place and the ejector clips will close themselves to lock the module into place.


5. Repeat previous step for all remaining DIMMs you wish to install.


6. Test it. Before you put your case back on, power on your system and make sure it correctly tallies the RAM.


7. Close up the case.

RIMMs (Rambus DRAM)

Some modern motherboards that don’t use DIMMs use Rambus DRAM (RDRAM) with 184-pin slots.  RDRAM is very similar to DIMMs RAM, but the pin configuration is a bit different.  Nevertheless, the same installation procedure for DIMMs applies for RIMMs (note that Dual-Channel operation is not supported).

DDR-DRAM is basically SDRAM which has been modified so as to make it more advanced, and ultimately faster. Double Data Rate Synchronous Dynamic Random Access Memory is the complete wording of the product, but DDR-DRAM is much easier to deal with. To dive you an idea of why it is called this, one must consider the design of SDRAM. On SDRAM, data, along with commands and addresses, are transferred on the rising portion of the clock cycle. With DDR, there is some special circuitry at the data pins that allows it to transfer this information on both the rise and fall of the clock cycle. This gives the memory a doubled speed when compared to SDRAM, this the term “Double Data Rate”.

Data rate is doubled by adding some extra circuitry between the DRAM itself and the data pins. This extra logic produces a data output strobe which synchs the data flow to the external clock speed, and allows the memory to transfer data on both sides of the clock cycle.

DDR-SDRAM and Rambus DRAM have been the two competitors for the memory of the future. Rambus DRAM was seen by most, myself included, as the future. This was caused by the continuing trend that when Intel speaks, everybody listens. Intel said that RDRAM was the future, and as such it was accepted as inevitable. However, RDRAM was very expensive. It was to the point of people needing an RDRAM slush fund, right next to their Roth IRA, in case one needed to upgrade. A number of RDRAM chipsets were produced, primarily for high end servers and whatnot, due to the effectiveness of Rambus. Then we forgot about it. Rambus is still used on many Pentium IV systems, but DDR-DRAM is by far more popular. This is caused by a number of fiscally interesting factors. Primarily, unlike RDRAM, this stuff doesn’t force you to sell your grandmother’s organs to buy some. Its price is comparable to that of the run-of-the-mill SDRAM available at your neighborhood tech kid’s garage computer chop shop.

Speed Ratings

The speeds of SDRAM were typically rated by the bus speed at which they are capable of operating, so PC100 and PC133 SDRAM sticks could operate on 100 and 133 MHz frontside busses respectively. DDR-DRAM, though, is rated a bit differently. Common ratings are PC 1600, PC2100, and PC3200. The rating systems are different, and are based on the theoretical maximum bandwidth of the memory units in megabytes per second. How is this calculated? Well, as I stated already, DDR-DRAM manages this by working with data both on the rising and falling clock ticks. This allows the 100MHz version to effectively run at 200MHz. Math time: 64 bit architecture, running at 2 times the normal rate, on a 100MHz bus. Divide by 8 bits to the byte, and we get a transfer rate of 1600MB/s. In my mind, I would say that that is more than acceptable. Do the same to the 133MHz bus and you get 2133MB/s. Hence, we get PC1600 and PC2100.

This rating system was also used for matketing reasons, no doubt. Rambus is sometimes called PC800, because of the rate at which it operates. Manufacturers of DDR-DRAM can’t go out and use small little numbers like PC133. It’ll give RDRAM the marketing advantage.

When motherboard manufacturers whiched from standard SDRAM to DDR-DRAM, ti gave the market a nice performance boost. But, users craved more, of course. PC1600 and PC2100 were nice SDRAM replacements, but did not give enough of a performance boost at first to make the techie drool on himself. So, newer and faster speeds were released: PC2400 and PC2700. Some of this memory was true PC2400/2700 memory, some sticks were actually overclocked. But, beyond that, there is now PC3000 DDR and some faster than that. Some of these faster units some with copper heat spreaders on them to aid in the dissipation of heat.

One reason that DDR-SDRAM needs a modified motherboard is because of its different package. It comes in 184-pin instead of 168-pin. It also only has one notch in the unit, as opposed to two on more conventional SDRAM.

DDR-DRAM is now the new standard for consumer memory. The new renditions of the technology compete with RDRAM quite well, and it is sure to increase performance on Athlon-based systems as well.