Free Web Hosting Provider - Web Hosting - E-commerce - High Speed Internet - Free Web Page
Search the Web

  OverClocked DDR Modules PC3500
     
Computer Memory Upgrades

Apple Computer Memory Live Newsfeeds

Computer Memory Products

Apple G5 DDR 400 Computer Memory

Computer Memory Upgrade Links

Overclocked DDR Computer Memory

BIOS Guide Computer Memory

BIOS Tweaks Computer Memory

Cisco Memory Pricing

Sun Memory Pricing

Micron Original Computer Memory

Apple Mac Memory

Add URL Post Offers Computer Memory

Computer Memory Contact Us

 


4003
Overclocking DDR Memory: Truths and Misconceptions & RAM Wars: Return of the JEDEC

Overclocking DDR Memory: Truths and Misconceptions
A Buyers Guide

Written By: Joey C. aka Chong345


Introduction:

With today's technology changing so quickly it is very easy to get behind. But when you hear about the fastest memory or CPU you think to yourself, "I have got to have it." Well, in truth, when looking at purchasing DDR based memory, there are a lot of things that you have to take into consideration. I am writing this article due to the fact that I frequently browse hardware forums and sites and one of the more notable questions that comes up all the time is, "What is the fastest memory?" or "What kind of memory should I buy?" People typically reply back and say what the faster modules are, or their opinion of different types of memory. What a lot fail to say are the other interesting points that one needs to know in order to maximize his / her system's performance, and at the same time spend money wisely.

DDR memory is available in a wide variety of different speeds at the moment. Here is a short list of some of the more common types:

PC1600 - DDR200 MHz (100x2)
PC2100 - DDR266 MHz (133x2)
PC2700 - DDR333 MHz (166x2)
PC3000 - DDR366 MHz (183x2)
PC3200 - DDR400 MHz (200x2)
PC3500 - DDR433 MHz (216x2)
Many manufacturers produce these modules, Powerline Memory, Kingston, Samsung just to name a few. Each of these companies produces memory that is "rated" at a certain speed. They stand by their product and claim that the memory will run as fast as its rated speed. As I stated before there are many other factors though. For instance, if you are using a DDR333 motherboard, then your motherboard definitely supports DDR333 or PC2700. Anything higher than PC2700 you will have to overclock to get it to run at that speed. To put it simply, just because you buy PC3500 doesn't mean that you will be running 433 MHz (DDR). You have to overclock to that speed because the current standard only supports up to PC2700. To overclock your memory you must raise the front side bus which will typically overclock your AGP/PCI slots, CPU, and DIMM slots. Selection of memory will be directly influenced by the components in the rest of your system.

The main component that you need to take into consideration is the motherboard. It will have to be able to handle the speeds that you want from your memory. So the first thing to think about when purchasing DDR memory is, "Will my motherboard be able to handle the speeds that I want to run at, and does it provide options in the BIOS that I will need to overclock my memory?" I would suggest reading up on your motherboard and finding out the results that others have had. This is key and will greatly aid in deciding which memory you should purchase.



Choosing Memory For AMD Systems:

Let's start with the AMD platform. When picking DDR memory for an AMD based system you need to be really picky about what you buy. There are very few AMD based motherboards at the moment that can run at DDR400 speeds. Most support 333 MHz but some are just better overclockers than others. The good thing about the AMD platform is that you can unlock the multiplier on the chip. The multiplier x FSB (front side bus) gives you your CPU speed. For instance 133 FSB with a chip that has a 10 multiplier would give you 1,330 MHz (133x10) or 1.33 GHz. Since you are raising the FSB to overclock your memory, your CPU will come into play. It will be getting overclocked and at some point it will not be able to overclock any higher. This could stop you from maximizing you RAM. So by unlocking the CPU and lowering the multiplier, you could run the processor at a speed you know it can handle but overclock the FSB even higher to maximize the memory bandwidth. So let's say you don't want to overclock your CPU but you want to get your memory to run at its maximum, which just happens to be 400 MHz (200FSBx2). Using the same CPU as the one in the first example and considering you have a good motherboard, you could use the 6.5 multiplier and run your CPU at 1.3 GHz (200x6.5). Issues that this high of a FSB are mainly PCI/AGP clocks as they are now running out of specification which cause instabilities, video distortions, or glitches (AGP tearing), hard drive issues, and the list just goes on. Luckily most 333 motherboards provide a 1/5 divider which allows you to put your AGP/PCI closer to spec. At 133 FSB on a 333 motherboard your AGP/PCI clocks run at 66/33. This is uses a divider for the PCI. At 166 FSB you could use the 1/5 divider to bring your AGP/PCI back to spec, but anything higher will make you run out of spec again. So when trying to run at speeds over PC2700 you might run into issues with your AGP/PCI clocks. In short your video card, hard drives, soundcard, etc. will determine how high you can go as well the memory.

As you can see when running an AMD machine and trying to maximize memory, there are a lot of other issues that that you have to deal with. Major concerns are the AGP/PCI clocks and the CPU speed. Also to really maximize the RAM it's nice to have voltage options up to 3.2 volts on the VDIMM such as the Epox 8K3A. I like AMD machines myself but they just cannot seem to overclock the memory like the Intel Platforms do.



Choosing Memory For Intel Systems:

Picking RAM for an Intel machine is not any easier than picking it for an AMD rig but you can overclock easier in some aspects and harder in other when compared to AMD. Unlike the AMD, Pentium 4 multipliers cannot be unlocked. So you can only overclock the RAM by upping the front side bus and the rest of the system. The good thing is that on an Intel platform you can lock the AGP/PCI clocks in at 66/33, this is a very nice feature that I like. You don't ever have to worry about if your video card, sound card, hard drive, etc. is giving out. Other things like we discussed earlier like voltage options, etc still will play a part in deciding process.

Okay since we cannot unlock the multiplier what can we do to make the ram run faster? Well luckily Northwood P4's are very good overclockers to begin with. They are able to achieve high FSBs on good air-cooling. Also there are options on the Intel platform motherboards such as RAM ratios. For instance, I have an Abit IT7, let's say I can overclock my CPU to roughly 160 FSB with good air-cooling using a P4 1.6A. This would mean the chip is at about 2.56 GHz and the RAM is set at 1:1 (FSB:RAM) so I would be running it like the AMD rig. It would be 160x2 so it would be 320 MHz DDR but my IT7 has a 3:4 memory ratio. So at 160 FSB the memory would be at 213x2 which would be 426 MHz! This is great for people who want to use PC3500 and run some really great memory speeds. Now you run into issues again though. Let's say you have really good cooling for instance. Something that lets you overclock to say 190 FSB. If you were to use a 3:4 memory ratio the memory would be at 506 MHz (253x2). Let's also say that you have some PC3500 that you want to use, well there are not many sticks of RAM that can run at 506 MHz. But if you were to use the 1:1 memory ratio then you memory would be doing 380 MHz. You would be underclocking you memory below its rated specifications. So what do you do? Well you could get a new chip with a bigger multiplier that will overclock to 160's FSB with serious cooling such as a 2.8 with its 21 multiplier. Or you could run your CPU lower than what you max out at so that you can use the 3:4 divider. This is the exact spot I am in at the moment. In my situation I can run my CPU at 193FSB but I have to use the 1:1 ratio. I don't want to go back to a lower FSB because I like my CPU to be maxed out. So what am I going to do? I am going to get a chip with the larger multiplier so I can have super fast CPU speed as well as memory speed.



Conclusions & Other Insights:

So you see there is a lot to Intel platforms as well as AMD. You have some good benefits and some bad for each. You have to choose your RAM wisely considering what kind of chip you have and what FSB you can run. I've seen some people get in a situation where their CPU was a bad overclocker so they could not max out the RAM, so they got extreme cooling and were able to overclock the CPU a lot more but got stuck in the situation that I am in.

When purchasing memory you have to find the perfect balance of everything if you want your components to be running at their full potential. This is called the sweet spot. Both AMD and Intel machines have a sweet spot where you have the perfect balance of memory bandwidth as well as CPU speed.

Now that we have covered the AMD and Intel platforms separately by what each machine has to offer and how it affects your choice, now let's talk about the last thing that applies to both. Memory timings are just as important as overall speed. Some people buy PC3500 and expect it to run at 3500 speeds and forget that not only is there a rated speed, but every stick of RAM has its own memory timings. For instance, some are CAS 2.5 3-6-3 2T and some are CAS 2 3-6-3 1T. These timings will influence how well you can overclock the memory as well. Since this is not a overclocking guide I will not go into what the timings mean but I just want the buyer to keep in mind that his memory is rated to run at a specific speed at specific memory timings, anything faster and you are overclocking your memory just like any other component, which mileage can vary.





Article Info
RAM Wars: Return of the JEDEC
Created:
April 1, 2003
By:
Bruce Gain

Summary:
DDR is becoming the predominate RAM memory for high-end and mainstream PC applications. The basic mechanisms of RAM memory are detailed, and memory units of measure are outlined and explained. Double channel DDRAM and DDRII are discussed. Also, Intel's divorce with Rambus and the implications add credence to JEDEC.

Introduction

This year will bring a radical change in the kind of memory you will buy and how you will buy it. The death knell has begun to sound for SDRAM while DDRAM has become a standard memory device. A bevy of new memory capabilities, such as dual channel DDR, will make life that much more interesting.

On the supplier side, the ramifications of Intel's recently-announced divorce from Rambus and its willingness to work with the Joint Electronic Device Engineering Council (JEDEC), the industry standards body, are beginning to bear fruit. Now, PC performance is largely contingent on what RAM is in the DIMM slot, to the same extent as the CPU and graphics processor, for gaming or any other performance-oriented applications. Meanwhile, Rambus is suddenly out of the picture.

In this article, we will look at the role that RAM plays in relation to the CPU, how RAM works, and what DDR, DDR400, and DDRII are all about.

SDRAM DDR RDRAM EDO DDR2
2002 55% 39% 5% 1%
2003 13% 81% 3% 3%
2004 8% 83% 2% 9%
2005 5% 58% 2% 35%

Marketshare of PC RAM types. (Source: iSuppli)

DRAM basics

The principal means of evaluating memory performance is in terms of cache latency, which is measured by the time it takes the CPU to fetch or retrieve data from memory. The RAM segment of memory, in fact, represents only one of three temporary memory caches through which data signals must pass on their way to and from the CPU and the main memory on the hard disk. Waxing philosophical, Dean McCarron, an analyst for Mercury Research in Cave Creek, AZ, noted: "By the time the processor goes out to the main memory, a bunch of bad things can happen."

Along the way through the L1 or L2 caches, a signal needed by the processor can drop from one of the caches and back out to main memory during which time the processor runs idle, which adversely affects latency. All said, the shorter the amount of time the processor waits for a response, the lower the latency. To see an example of how latency affects performance, turn down latency in your PC's Bios and compare performance with and without it.

During the last few years, efforts to improve cache performance and latency have been largely focused on the processor and chipset link (the front-side bus) and the memory system bus between the memory and the chipset, with the bus speed representing a potential bottleneck. A 533 MHz front-side bus on a Pentium 4 talking to PC-133 (133-Mhz) SDRAM represents a significant differential, for example, between the front-side speed and the RAM's bus speed.

It is possible to configure a system that has a higher memory bus speed but a slower latency, though it would perform worse than something that was slower with better latency. As a rule of thumb, the front-side bus frequency of the processor should be no less than one-fifth of the processor frequency. This basically means that the cache can handle things up to that point after which the front-side bus becomes saturated. Using Intel's high-end chipsets as an example, Intel will come out with new bus speeds once they get past the 5x ratio. But with Celerons, which Intel sells on megahertz, the bus/ processor ratio is reportedly as much as 13 to one.

Irrespective of technical specifications, some memory modules do not perform to specification. At the billion dollar fab units where memory is made, RAM devices that do not test out at 333 Mhz might be sold instead as 266 Mhz devices. Among the vendors and distributors around the world, memory is also often mislabeled. A RAM device with a 266-Mhz device may in fact run at a slower clock speed. If several bits go bad, stability will suffer, especially in intensive applications such as overclocking.


"Powerline's 512 PC3500 DDR OverClocked Modules, Provide An Awesome Gaming Experience..."
Doug Perkins- customer
  Latency
A DRAM memory array can be thought of as a table of cells. These cells are comprised of capacitors, and contain one or more 'bits' of data, depending upon the chip configuration. This table is addressed via row and column decoders, which in turn receive their signals from the RAS (Row Address Strobe) and CAS (Column Address Strobe) clock generators. In order to minimize the package size, the row and column addresses are multiplexed into row and column address buffers. For example, if there are 11 address lines, there will be 11 row and 11 column address buffers. Access transistors called 'sense amps' are connected to each column and provide the read and restore operations of the chip. Since the cells are capacitors that discharge for each read operation, the sense amp must restore the data before the end of the access cycle.



The capacitors used for data cells tend to bleed off their charge, and therefore require a periodic refresh cycle or data will be lost. A refresh controller determines the time between refresh cycles, and a refresh counter ensures that the entire array (all rows) is refreshed. Of course, this means that some cycles are used for refresh operations, and this has some impact on performance.

A typical memory access would occur as follows. First, the row address bits are placed onto the address pins. After a period of time the RAS signal falls, which activates the sense amps and causes the row address to be latched into the row address buffer. When the RAS signal stabilizes, the selected row is transferred onto the sense amps. Next, the column address bits are set up, and then latched into the column address buffer when CAS falls, at which time the output buffer is also turned on. When CAS stabilizes, the selected sense amp feeds its data onto the output buffer.

Memory Specs
Memory Specs

1987 FPM 50ns
1995 EDO 50ns
1997 PC66 SDRAM 66MHz
1998 PC100 SDRAM 100MHz
1999 RDRAM 800MHz
1999/2000 PC133 SRAM 133MHz (VCM option)
2000 DDR SDRAM 266MHz


In the early days of mainstream PCs in the 1980s, memory was a big deal, with 512K being the standard, while cutting edge enthusiasts would send e-mail and play Tetris (not at the same time) with a combination of Intel x386 processors that had as much as 64 Mbit of RAM.

Then and until the advent of SDRAM, the time it took the processor to send and receive data from RAM was measured in nanoseconds (ns). Fast memory speeds ranged from 80ns to as quick as 60ns.

Megahertz and Bytes-per-second transport speeds did not become the standard of measure until the advent of SDRAM. Today, memory and its bus architectures live and die by both capacity, measured in MBytes, and speed, which is measured in the amount of of data that can be accessed and sent by the memory module, expressed in Bytes (bits multiplied by eight) per second. Equally important is the front-side bus speed between the CPU and memory, which is also measured in MHz and Bytes/s.

An issue of potential confusion is that memory modules are still labeled using nanoseconds as a unit of measure for memory performance. Beginning with DRAM, instead of access time, nanosecond units have measured clock cycle times. DDRAM will likely continue this legacy.

Accord Memory Service, Inc., a leading memory module producer, offers the following table indicating the method for determining speed equivalencies between MHz and ns ratings.

STEP 1
MHz = 1 million clock cycles per second:
66
100
133

STEP 2
Multiply by 1 million to get total clock cycles per second:
66,000,000
100,000,000
133,000,000

STEP 3
Constant: 1 billion nanoseconds per second: 1,000,000,000
1,000,000,000
1,000,000,000

STEP 4
Divide nanoseconds per second (from Step 3) by clock cycles per second (from Step 2) to get nanoseconds per clock cycle:
15
10
8

Buy Powerline's OverClocked DDR 433MHz Gaming Module Here

Enter GDDR2 SDRAM

Why are graphics heavyweights ATI and Nvidia jumping ahead of everyone else, including JEDEC, with their own proprietary versions of DDRII? Because as the GPU has gained ground in the PC performance hierarchy, so have the requirements of SGRAM memory, or video memory.

Indeed, the graphics processor world has seen many great advances during the past few years -- many a result of the slugfest between ATI and Nvidia to offer the best graphics card capable of matching the best games out there (anyone who has played Halo or Unreal with a Radeon or GeForce video card online late past midnight, only to call in sick at work or skip school the next day, knows this). In many respects, after a certain clock-speed threshold has been reached, the distinguishing factors of the GPU have become more crucial for graphics performance; and, in order to match the GPU's processing capabilities, ultra-high performance memory must be on the menu as well.



Micron is well aware of this, and along with Samsung and Hynix, has begun developing graphics DDRII memory prior to the finalization of a graphics DDRII standard by JEDEC, which is slated for this summer. This memory, now known as GDDR2, will do a lot to accommodate the powers of ATI's latest Radeon and Nvidia's GeForceFX, as well as other vendors' goods, such the latest devices from Trident or S3 with 1-Ghz of bandwidth.

DDR's double-fetch-per-cycle capability for graphics has helped, but in a recent patent filing Micron has indicated that a burst operation is required, which Micron defines as an operation retrieving a given number of data stored at sequential locations within the memory.

Additionally, GDDR2 will include a memory array that is addressable by even and odd word addresses. The logic circuitry with a burst increment mode will access the array starting at an even word address, and a burst decrement mode will access the array starting at an odd word address.

Within the scope of Micron's burst mode, the memory device has a burst increment mode when starting at an even word address, and a burst decrement mode when starting at an odd word address. Data increments are counted, so that the second data word retrieved is still from the same memory location as the first data word as addressed by the logic circuitry. When starting at an odd word address, the device counts down (decrements), so that the second data word retrieved is still from the same memory location as the first data word as addressed by the logic circuitry.

But, as if GDDR2 were not enough for the GPU when it now looks as if DDRII for CPU chipsets may not see wide-scale application before 2005, Micron, Infineon, Elpida, and Hynix says they will have samples of GDDR3 this year.

Powerline OverClocked DDR Modules


Measuring modules & Dual Channel DDR 400
Measuring modules

Figuring out how much capacity your memory module has is straightforward enough, but also involves some basic math. Arriving at the number of bits in a module involves multiplying the size of the memory chip by the number of chips in a module. For example, eight 64-Mbit chips are in a 512 Mbit module. But unfortunately, when you read module specs, module capacity is expressed in Bytes. In this case, you have to divide by 8, and are left with 64 MBytes.

SDRAM

DRAM that has a synchronous interface is known generically as SDRAM. This includes CDRAM (Cache DRAM), RDRAM (Rambus DRAM), ESDRAM (Enhanced SDRAM) and others.

SDRAM that meets the standards established by JEDEC (see Return of the JEDEC section) not only has a synchronous interface controlled by the system clock, it also includes a dual-bank architecture and burst mode (1-bit, 2-bit, 4-bit, 8-bit and full page). A 'mode register' that can be set at power-on and changed during operation controls the burst mode, burst type (sequential or interleave), burst length, and CAS latency (1, 2 or 3).

Just a quick note on CAS latency: CAS Latency is a performance-related timing for SDRAM that falls under the latency umbrella. This measurement is the time it takes to strobe in the Row Address, and to activate the bank. When a burst read cycle is initiated, the addresses are set up and RAS and CS (chip select) are held low on the next clock cycle (rising edge of CLK), thereby activating the sense amplifiers on the bank. A period of time equal to tRCD (RAS to CAS delay) must pass, after which CAS and CS are held low (again, at the next clock cycle). After the time period for tCAC (column access time) has passed, the first bit of data is on the output line and can be retrieved (at the next clock cycle). The basic rule is that CAS latency multiplied by the clock speed (tCLK) must be equal to or greater than tCAC (or CL x tCLK >= tCAC). This means that the column access time is the limiting factor for CAS Latency.


During the coming weeks, Intel will launch its Springdale chipset with DDR400, which will represent the first commercially available dual channel memory application.

Dual channel memory is anything but new. It has been used for many years in servers and in Rambus memory and allows the memory controller, as it links to the memory module, to split the data and simultaneously write to one physical module on the right and one on the left.



Springdale will combine DDR400 with an 800 Mhz P-4. At first glance, the P-4 800-Mhz memory bus will be twice the speed of the 400-Mhz DDR400, which would normally slow things down at a 400-Mhz clock speed. Ideally, you would want to match the processor front side bus and the memory bus, so 800-Mhz DDRAM would represent a perfect match in this hypothetical example.

But dual channel DDR400, in theory, offers the same bus speed as single-channel 800-Mhz DDRAM.

Dual channel DDR400 requires two DIMM slots and two modules. The architecture, while offering 6.4 GBytes/s of peak bandwidth, simultaneously splits the back-and-forth signaling with the CPU. The signal from each channel comes from one of the two sockets in the DIMM slots.

Vendors are scrambling to offer chipsets that conform to DDR400. According to Crucial Technology, the following chipsets will support DDR400:

Intel Processors with VIA P4X400, ALI M1681, SIS 648DX, or Intel Springdale;
AMD Processors with KT400, SIS 746DX or 746FX, or Nvidia Nforce2.


Conclusion

The year has reflected slow to flat growth in PC sales while memory sales continue to plummet. But, while this disappoints shareholders and other money-chasing folks, 2003 should prove to be one of the most exciting years ever for gamers and other PC enthusiasts. In just a few weeks, we will see the widespread commercial launch of motherboards with 3-Ghz plus processors and 800-Mhz front-side buses with matching 800Mhz DDR400 memory specs. Already available are Nvidia's GeForceFX and ATI's Radeon 9800, where DDRII used as graphics memory offers a taste of what is yet to come until DDRII becomes available for CPU chipsets.



But this is not a perfect world, to say the least. Now that Rambus is out of the PC picture, what entity will set the standard to beat in the way of memory performance? Even begging the question will rankle many feathers. But then again, maybe relying on the agreed upon JEDEC standards for DDR400 and DDRII will offer more than enough improvements to match chipset advances for years to come. After all, DRAM is only a commodity component.

Further Reading:

DDR400 versus Rambus: Charts and tests of DDR400, DDR333, and DDR266, as well as RDRAM PC800. Also, DDR Performance Analysis gives some interesting memory comparison benchmarks.

"Computer Organization and Design: The Hardware/ Software Interface" by David A. Patterson and John L. Hennessy.

This book is the Bible for many PC hardware engineers and is also on the required reading list for university computer design and computer science undergraduates. The good writing style makes learning RAM and memory concepts more accessible.