DDR-2 Memory Investigation Guide - Making sense of memory settings

In our latest guide we investigate DDR-2 memory on the Intel Pentium 4 platform and the differences between high bandwidth and timings to figure out which offers the best system performance. What is better - higher clock speed or super tight timings? We discuss the reasons why one is better than the other and prove our claims through a range of different system configuration benchmarks.
Published Wed, Apr 20 2005 11:00 PM CDT   |   Updated Tue, Apr 7 2020 12:26 PM CDT
Manufacturer: none


IntroductionMemory is one of the hottest growing components on the overclockers list of most important requirements. A good CPU is one thing and cooling is another but if your memory can't handle the pace, then you're simply throwing money away.DDR-2 has now started its mass run on the Intel market, with rumours that AMD will start looking at DDR-2 on the Athlon 64 sometime in the near future - but one thing good is it? We have seen in the past that DDR memory running at lower latencies can do a better job then memory running at higher clock speeds with higher latencies. Though this is the general rule for the AMD Athlon XP and Athlon 64 CPU, due to a much more ridged memory controller (especially on the Athlon 64 with the on CPU memory controller), this isn't as much the case when it comes to the Intel camp. Intel Pentium 4 with its longer pipeline tends to respond better to higher bandwidth than lower latencies, at least this proved true 90% of the time with DDR. As we mentioned earlier, DDR-2 is a different story, as some of the key developments have changed, and thus changed the overall bandwidth and latencies of the memory and how does this affect the memory performance of the Pentium 4 based DDR-2 systems? Today we hope to answer thisToday we are putting OCZ's DDR-2 memory onto one of the best overclocking boards available today, the Intel 925XE ASUS P5AD2-E Premium to see just how far DDR-2 can go and what the overall performance pluses and minuses are of the different configurations such as clock speed vs. latency, running memory 1:1 and not and so on.If you're interested about how to setup your new DDR-2 based system properly when it comes to memory, read on and we'll show you as proved by a bunch of benchmark comparisons.

DDR vs. DDR-2 - The Main Differences

DDR vs. DDR-2This is one of the biggest points that separate both AMD and Intel. While we will be talking about both systems here, we won't actually be testing the AMD, simply because it doesn't support DDR-2 and is not likely to for some time.DDR-2 is the evolution of the DDR memory standards. First off let's take a brief look over the two technologies and their main differences. In order to increase the overall data transmission throughput without having to increase the speed of the memory is to actually increase the internal data rate of the cell array. DDR-2 uses two cells running in parallel; each cell uses two bits, making it a 4 bit total cell whereas DDR only has a 2 bit process. In order for the data buffer to keep up, the buffers run at twice the cell array speed. This overall process allows for DDR-2 to run at an effective clock rate 4x that of the actual memory modules. What this means is DDR-2 memory with a 400MHz rating is actually running with cells running at a clock speed of 100MHz which would be 200MHz with regular DDR.The next major addition to DDR-2 is On-Die signal termination. One of the biggest problems with DDR1 was compatibility with other modules in most cases and in a few others with motherboards. This was because of the Signal termination. DDR memory used a series of resistors on the motherboard in order to terminate the bus signal noise on the memory channel. The problem with this is two fold:1) When using modules from different manufacturers, if one module sent too much noise through the bus line to the additional memory module you could get memory instabilities.2) If the motherboard didn't use enough of the right resistors or use ones of inferior quality, the board would cause memory stability problems. DDR-2 actually puts the termination for the bus signals on the modules themselves. This means that modules can be mixed and matched between manufacturers with relative ease. Another great feature is that the signal termination is closer to the memory modules, allowing for less noise on the bus and allowing for faster speeds due to a cleaner signal.

The Latency Issue

The Latency Issue - Timings vs. BandwidthNow we get to what has caused all the uproar with DDR-2 - higher latencies. DDR1 memory modules are well ahead of DDR-2 in terms of latency. DDR operates at 2.0ns, 2.5ns and with super high speed modules 3ns. 2ns are the modules of choice for AMD Athlon users as memory latency really takes a huge wack out of the AMD CPU's however; Intel doesn't follow this general rule. DDR-2 with the double buffer rate has pushed latencies (otherwise known as timings) to 3ns, 4ns and even 5ns. AMD Athlon 64 with its on-die memory controller uses a much faster crossbar to access the memory modules which means that the faster the modules can synchronise with the controller, the better the performance.Due to the traditional FSB nature of the Intel Pentium 4 and the Netburst architecture's higher bandwidth capabilities, latency doesn't play a huge role, as the latency of the FSB alignment is already there, memory modules that can run faster than this alignment simply don't help.The large pipeline of the Intel Pentium 4 requires a huge amount of data per clock cycle in order to keep the Pentium 4 from stalling; this means that the memory has to have a lot of throughput in order to keep the data path happy. Pentium 4 uses an 800MHz FSB for its Pentium 4 with a 1066FSB for the faster Extreme Edition CPU. This means that an 800 FSB P4 can chew up to 6.4GB/s and an Extreme Edition up to 8.5GB/s. This simply is beyond what the DDR standard can produce; however, DDR-2 can do this. DDR-2 running at higher than CPU clock speeds is also possible with Asynchronous bus speeds.Traditionally this has led to latency problems, however, with the nature of DDR-2 and the already higher than traditional latencies, Asynchronous DDR-2 on paper actually would benefit the Pentium 4 previously than before.

The Testing Modules

The Testing ModulesOCZ were kind enough to provide us with two 512MB sticks of their high quality PC-5400 DDR-2 Performance Series memory for our article.
To read more about this memory from OCZ, click here for more information from our recent DDR-2 PC-5400 Performance Memory Shootout.

Benchmarks - Test System Setup and Further Information

Test System SetupProcessor: Intel Pentium 4 3.46GHz (1066 FSB) (Supplied by Spectrum Communications)Memory: 2x 512MB OCZ DDR2-5400 Performance Series (Supplied by OCZ Technology)Motherboard: ASUS P5AD2-E Premium (Supplied by ASUS)Graphics Card: ATI Radeon X700 256MB Operating System: Microsoft Windows XP SP1Drivers ATI Catalyst 5.2 and DirectX 9.0cAll tests were ran with ASUS Hyper Path turned on which is supposed to increase memory performance but in the majority of our tests, it didn't make any difference. All tests with the memory voltage set to 2.2v which is the maximum the OCZ memory will accept and the highest available on the ASUS motherboard.We tested each of the memory settings with Mem Test using the standard 100% coverage test on all unused memory at default to check memory for errors and we had 0 errors. Mem Tests locks up when sticks are overclocked but Mem Test is questionable in this area since the overclocked settings passed all other tests except Mem Test.For our results, we actually ran the FSB, memory and CPU at a range of different speeds and combinations which affect the outcome of the results. We did this to show you which setup manages to gain the best overall performance. In order for you to understand, the following settings were used for each result:- Max Memory Clock / Loose Timings: CPU Speed was 3780MHz with a DDR2 ratio of 3:4 running at 840MHz. FSB rating was 315Mhz with a 12x multiplier - timings were 5-5-5-8- Max 1:1 / Default Timings: CPU speed was 3780MHz. DDR2 ratio was 1:1 at 630MBz. FSB running at 315MHz with a 12x multiplier - timings were 4-4-4-8- 266 1:1 / Best Possible Timings: CPU speed was 3458MHz with a DDR ratio of 1:1 at 533MHz. FSB was 266MHz with a CPU Ratio of 13x - timings were 3-3-3-4- 266 1:1 / Default Timings: CPU speed was 3458MHz with a DDR Ratio of 1:1 at 533MHz. FSB was 266MHz with a multiplier of 13x - timings were 4-4-4-8- Best on Paper Combo: CPU speed was 3720MHz with a DDR Ratio of 3:4 at 826. FSB was 310MHz with a multiplier of 12x - timings were 4-4-4-1

Benchmarks - SiSoft Sandra

SiSoft SandraVersion and / or Patch Used: 2004Developer Homepage: Homepage: It Here
SiSoft Sandra (System ANalyser, Diagnostic and Reporting Assistant) is a synthetic Windows benchmark that features different tests used to evaluate different PC subsystems.
We can see here that the best on paper combo yields the best synthetic results.

Benchmarks - PCMark04

PCMarkVersion and / or Patch Used: 2004Developer Homepage: http://www.futuremark.comProduct Homepage: It Here
PCMark is a multipurpose benchmark, suited for benchmarking all kinds of PCs, from laptops to workstations, as well as across multiple Windows operating systems. This easy-to-use benchmark makes professional strength benchmarking software available even to novice users. PCMark consists of a series of tests that represent common tasks in home and office programs. PCMark also covers many additional areas outside the scope of other benchmarks.
Again the best on paper combo takes the win with the Max Memory clock coming in a very close second.

Benchmarks - PCMark02

PCMarkVersion and / or Patch Used: 2002Developer Homepage: http://www.futuremark.comProduct Homepage: It Here
PCMark is a multipurpose benchmark, suited for benchmarking all kinds of PCs, from laptops to workstations, as well as across multiple Windows operating systems. This easy-to-use benchmark makes professional strength benchmarking software available even to novice users. PCMark consists of a series of tests that represent common tasks in home and office programs. PCMark also covers many additional areas outside the scope of other benchmarks.
Again the best on paper combo shows the way.

Benchmarks - ScienceMark 2.0

ScienceMark 2.0ScienceMark 2.0 is a mathematical program designed to stress the memory subsystems of both desktop/workstation and server environments to determine the read/write latency as well as the overall memory bandwidth available between the CPU and the memory controller.
In both bandwidth and latency tests the on paper combo leads the way yet again.

Benchmarks - MP3 Encoding

MP3 EncodingUsing a program called CD to MP3 Ripper (v1.50), we timed how long it would take to convert our retail copy of "50 Cent - Get Rich or Die Tryin'" from CDA to MP3 format using the bit rate size of 128kbit for the entire album. Obviously lower is better.
Again the on paper combo manages to steal a few extra seconds but the max memory clock comes out in front since the CPU is operating fastest here.

Benchmarks - 3DMark Series

3DMark2001 SEVersion and / or Patch Used: Build 330 SEDeveloper Homepage: http://www.futuremark.comProduct Homepage: It Here
3DMark2001 SE is a part of the popular 3DMark series. By combining DirectX 8.1 support with completely new graphics (including the GeForce4), it continues to provide benchmark results that empower you to make informed hardware assessments.
Here we see that the extra bandwidth of the best on paper combo only gives a very slight lead, nothing that would be noticed in game.3DMark03Version and / or Patch Used: Build 360Developer Homepage: http://www.futuremark.comProduct Homepage: It Here
3DMark03 is the latest version of the highly favored 3DMark series. By combining full DirectX9.0 support with completely new tests and graphics, 3DMark03 continues the legacy of being industry standard benchmark.Please Note: Due to recent events with the 3DMark03 series, we are adding results purely for those who are still in favor of 3DMark03. These results should not be taken too seriously and are only added for interest sakes.
Here the Max memory clock manages to take a slight lead.3DMark05 Version and / or Patch Used: Build 120 Developer Homepage: Product Homepage: Buy It Here
3DMark05 is the latest version in the popular 3DMark "Gamers Benchmark" series. It includes a complete set of DX9 benchmarks which tests Shader Model 2.0 and higher. For more information on the 3DMark05 benchmark, we recommend you read our preview here.
Again the Max memory clock manages to take the lead.

Benchmarks - Doom 3

Doom 3Version and / or Patch Used: UnpatchedTimedemo or Level Used: Custom TimedemoDeveloper Homepage: Product Homepage: http://www.doom3.comBuy It Here
Doom 3 is the latest game to hit our test lab and is one of the most intensive games to dates. With our own custom time demo we are able to give a realistic rating on what kind of FPS you will be achieving.For more information on benchmarking Doom 3 we recommend you check out our extensive article regarding it here.
Doom 3 is the first of our real world benchmark and gives a slight lead to the Max Memory clock.

Benchmarks - Half Life 2

Half Life 2Version and / or Patch Used: UnpatchedTimedemo or Level Used: Custom TimedemoDeveloper Homepage: Product Homepage: http://www.half-life2.comBuy It Here
By taking the suspense, challenge and visceral charge of the original, and adding startling new realism and responsiveness, Half-Life 2 opens the door to a world where the player's presence affects everything around him, from the physical environment to the behaviours even the emotions of both friends and enemies.We benchmark Half Life 2 with our own custom timedemos as to avoid possible driver optimizations using the "record demo_name" command and loading the timedemo with the "timedemo demo_name" command - For a full list of the commands, click here.
In the last of our tests the Max memory clock manages to take the lead.

Final Thoughts

Final ThoughtsWe can see from our test results that it is pretty clear the Intel Pentium 4 is not a latency craver, as its design simply precludes any need for extremely low latencies which the AMD Athlon craves. It is however, a bandwidth hog. Intel Netburst was originally designed to work hand in hand with RDRAM, which for its time managed to introduce some extremely high bandwidth in the order of around 3.2GB/s when DDR was only able to push out 1.6GB/s which should be a pretty good indication of our claims and what our tests proved.DDR-2's higher latency over DDR doesn't make it such a bad choice now that speeds are starting to reach 667MHz and above which is much quicker than the speeds which DDR-2 was first released at 533MHz. DDR-2 is now starting to show its colours, and with Dual Core, PCI-Express now starting to take a chunk of the bandwidth, its certainly a good thing to see that bandwidth is now in abundance.We can see that when we use the tightest possible timings, performance is improved a little in our synthetic tests but when it comes to real-world (games), there are hardly any performance increases, definitely increases which you won't notice.So with everything in mind, when it comes to setting up your system (overclocked or not), we can see that for the Pentium 4 platform series, higher bandwidth is more important than tighter timings when it comes to overall performance. Tight timings are good but are far from crucial for the Pentium 4 - keep this in mind when it comes to buying DDR-2, especially expensive DDR-2.

PRICING: You can find products similar to this one for sale below.

USUnited States: Find other tech and computer products like this over at

UKUnited Kingdom: Find other tech and computer products like this over at

AUAustralia: Find other tech and computer products like this over at

CACanada: Find other tech and computer products like this over at

DEDeutschland: Finde andere Technik- und Computerprodukte wie dieses auf

We openly invite the companies who provide us with review samples / who are mentioned or discussed to express their opinion. If any company representative wishes to respond, we will publish the response here. Please contact us if you wish to respond.
Newsletter Subscription

Latest News

View More News

Latest Reviews

View More Reviews

Latest Articles

View More Articles