RAM is one of the most crucial component in a PC. Sadly, it is also the most
neglected one. It comes in different makes, with different latency, bandwidth;
and now-a-days RAM catering to the budget conscious people; and to people who
want to overclock their RAM to attain maximum performance. But with DDR 2 RAM
available, and tweaked to the max, what's the next generation technology that
would replace it? Also what about the static RAM and RAM used in servers, where
are they heading? We look at the available RAM options, what you need to know
about them, alongwith the upcoming technologies that would replace current
generation RAM.
Static or dynamic RAM?
RAM is categorized mainly as Static or Dynamic RAM. The major difference between
the two is that dynamic RAM needs to be refreshed many times per second by the
processor or it loses all the stored data, whereas static RAM retains data even
it is not refreshed by the processor. The major benefit of static RAM is that it
doesn't create overhead on the CPU, by eliminating the need for refreshing
regularly. Up till now we have seen various avatars of both dynamic and static
RAM to fulfill the needs of the growing IT sector.
|
The commonly known dynamic RAM are Fast page mode RAM
out RAM
Dynamic RAM
function as cache memory, and Quad data rate RAM (a form of static RAM), used in
switches and routers. What's coming? Well there are a couple of new
technologies, any of which can become the mainstream component for powering
future generation of computers.
The key trend in RAM over the past 4-5 years has been a huge increase in data transfer rates |
RAM for Communications
QDR SRAM
years and provides data rates above 200 MHz . The SRAM provides data throughputs
of 11Gbps as they use separate input and output ports for read and write and
these ports have separate differential clocks allowing for four words transfer
on each clock cycle. Various products from Cypress and Samsung are available.
QDRII+ RAM was announced last year along with DDR2+. These RAM have high
performance architecture and operate at speeds upto 500 MHz, which is more than
twice of what existing products offer. SRAM delivers a bandwidth of 72Gbps while
using a 165 pin FBGA (fine pitched ball grid array) package. These are used
mainly in the next generation of switches and routers to enhance communication
capabilities by giving the density and performance required to expand network
capabilities.
The return of RAMbus
RAMbus Incorporated is the manufacturer of RDRAM, which at one point of time was
considered a replacement for SDRAM. It formed an agreement with Intel to be the
primary memory technology of all Intel products. However, RDRAM was not able to
make a mark for itself due to various problems such as high latency, heat
emission and high cost of manufacturing. So, RDRAM faded away, but RAMbus did
not die and has come back with XDR DRAM.
Elpida's XDR RAM chips use 90nm process technology to provide bandwidth of 8 Gbps |
This new memory module has a high performance architecture based on the
RAMbus RDRAM. It improves on the high latency problem of RDRAM, significantly
improving bandwidth while reducing the number of ICs required. The XDR devices
will feature programmable on chip termination, adaptive impedance matching,
dynamic request scheduling and zero overhead refresh. Also, XDR uses Octal Data
Rate transfers to transfer eight bits of data per 400MHz clock cycle which
results in 6.4Gbps bandwidth speed for a 16bit bus instead of 3.2Gbps bandwidth
provided by a DDR RAM for a 64bit bus. Samsung and Elpida are already
manufacturing XDR RAM and now Qimonda has joined in to mass manufacture XDR RAM
for game consoles, set top boxes and PC graphics market. The Playstation 3 has
shipped with 256 MB module of XDR RAM onboard to achieve maximum performance,
whereas RAMbus has started development on XDR2 RAM with which they are targeting
8.0GHz data rates, enabling a single DRAM device with 16 Gbps of peak bandwidth.
Successor to DDR2
DDR3 SDRAM is the one that is the hot favorite of many and touted as the leader
of the next generation of RAM. DDR3 is based on 90nm fabrication technology and
would be using 'Dual gate' transistors to reduce leakage of current, leading
to almost 40% less power usage in comparison to DDR2. As DDR3 modules are
expected to transfer data at a clock rate of approximately 800 MHz instead of
400 MHz, the clock rate of DDR2, it is expected to attain up to 1600 MHz single
clock bandwidth as compared to 1066 MHz bandwidth of DDR2 RAM. Also the prefetch
buffer of DDR3 is 8-bit wide in comparison to 4-bit buffer available in DDR2,
which leads to double bandwidth and decreased heat production. A newly
introduced automatic calibration feature for the output buffer enhances the
ability to control the system timing budget during variations in voltage and
temperature.
How they Compare | |||
DDR | DDR2 | DDR3 | |
Clock Frequency | 100/133/166/200 MHz | 200/233/333/400 MHz | 533/677 MHz |
Data Rate | 200-400 Mbps | 400-800Mbps | 800-1600Mbps*2 |
Prefetch Bit Width | 2 bits | 4 bits | 8 bits |
Supply Voltage | 2.5V | 1.8V | 1.5V |
CAS Latency | 2, 2.5, 3 clocks | 3, 4, 5 clocks | 5, 6, 7, 8, 9, 10 clocks |
Driver Caliberation | No | Off Chip Driver Calibration | Self Calibration with ZQ Pin |
Interface | SSTL_2 | SSTL_18 | SSTL_15 |
System Assumption | 4 slots (8 loads) | 2 slots (4 loads) | 2 slots (4 loads)*2 |
However DDR3 does have a disadvantage of having a higher CAS (column address
select) latency. But this can be overlooked, due to the higher bandwidth
provided by DDR3 memory modules. Intel has already announced that it would
develop the architecture to support DDR3 RAM when they become available late in
2007. AMD too has announced a joint venture with SimpleTech Inc, to get ready
for DDR3-based SDRAM, expected to be available by 2008. So, DDR3 has already
established itself as the successor to DDR2.
How much RAM is enough? |
Technically the more RAM you have the better, but then everyone is not willing to spend huge sums of money. So how much is enough? It depends on the OS and the type of work to be done. Even though the minimum required RAM for XP is 256MB, it is suggested that at least 512MB RAM should be installed to achieve optimum performance. If you are a hard core gamer or a power user running CPU/graphics intensive work such as 3ds Max, then a minimum of 1GB RAM is recommended. But for Vista the minimum RAM should be 1GB, whatever version you choose. |
RAM for servers
Up till now, DDR2 RAM has been used in most of the servers, but things are
changing with the arrival of FB-DIMM RAM. It has already been adopted by Intel
for their latest Xeon 5000/5100 series servers, whereas AMD's new processors
based on the K8L architecture will have support for FB-DIMM RAM Even Apple is
using this RAM in its Mac Pro workstations. These RAM use an Advanced Memory
Buffer (AMB) between the memory controller and the memory module. Instead of the
parallel bus architecture used in DRAM, they use a serial architecture
connecting the memory controller and the AMB, which in turn allows for error
correction and more memory bandwidth than DDR2 and DDR3. The architecture also
allows for more memory width without an increase in the pin count; however it
suffers from a higher latency in memory request similar to that in DDR3. But
these provide the required bandwidth needed by servers and thus higher latency
can be overlooked. Additionally, GDDR3 memory is used in current generation
graphics cards to boost video performance. Recently ATI shifted onto GDDR4
memory in their X1950 series of graphics cards. There are talks of GDDR5 memory
being released in 2007, but this is yet to be confirmed from the manufacturers.
Overall, this year could be a defining stage for next generation RAM.
Glossary | |
Access time The average time interval between storage peripheral receiving a request to read or write a certain location and returning the value read or completing the write. Bandwidth The capacity to move data on a bus or a channel to another point in a given time. It is expressed in bits, bytes, or Hertz (cycles) per second. Bus cycle A single transaction occurring between the system memory and the CPU. Cache A small, fast area memory holding recently accessed data, designed to speed up subsequent access. CAS latency The time taken in sending a request to read a memory location by the memory controller and the time taken to send the data to output pins. Clock Speed The rate at which a computer's internal system clock operates. Clock Rate The number of pulses emitted from a computer's clock in one second. The speed of the clock is measured in Megahertz (MHz). |
DIMM Dual In-line Memory Module. FSB Front Side Bus is the data channel connecting the processor, motherboard and RAM. FSB is described in terms of its width in bits and its speed in MHz. Memory Cycle Minimum amount of time required for a memory to complete a cycle such as read, write, read/write, or read/modify/write. Read time The amount of time required for the output data to become valid once the read and address inputs have been enabled; generally called access time. SIMM Single In-line Memory Module. Write time Time expended from the moment data is entered for storage to the time it is actually stored. Prefetch buffer It is a memory cache located on modern RAM modules which stores data before it is actually needed. |