Each time we had approached the launch date for Intel's new 925X/915 chipsets, we ran into delays. The rumors seemed to be consistent that there were still issues with DDR2 memory for the new platform, and almost everyone with whom we talked had described the difficulties with getting the new DDR2 memory working as it should.

Against that background, it is positively amazing that we find such a broad selection of DDR2 memory just a couple of weeks after the launch (finally) of the new Intel architecture. While we received the first samples of DDR2 memory over 3 months ago, it is clear that development has continued over the past few months. We had already found that DDR2 533 performed about the same as fast DDR433 in our 925X/915 launch review using the Intel D925XCV. While this is not the 1:1 to DDR533 that many hoped for, it is still an excellent improvement in performance over the early results with DDR2. However, Intel motherboards are not traditionally geared toward pushing memory to even higher performance than specification, and we also had to contend with the limitation of the Intel 10% overclock lock that is engineered into the 925X/915 chipsets.

However, with the Abit AA8 and Asus P5AD2 known to have found ways around the 10% lock, it was time to take a closer look at the performance of DDR2. Other 925X motherboard manufacturers, such as Gigabyte and MSI, are also rumored to have broken the 10% lock. With a little more open route above a 220FSB setting, it now made sense to look at the current state of DDR2 memory performance. In particular, we were interested in seeing how far current DDR2 memory would need to develop to reach the magic DDR667 that will be needed for the next speed ramp of the Intel architecture within the next few months. The answers may surprise you.

Specifications: DDR2 Memory

Considering the delays and false starts with DDR2, we were surprised to receive DDR2 modules from 8 different manufacturers for our first DDR2 roundup. The DDR2 from Corsair was even rated at DDR2 667, which caused us to believe that the Corsair would stand head and shoulders above the crowd. We also know of other DDR2 667 samples that are on the way from OCZ and others that will soon be announced. For memory that was barely working a couple of months ago, it certainly appears that DDR2 is maturing much faster than anyone expected.

The 8 DDR2 memories that we received were all rated at or near the Jedec standard for PC2-4300 of 4-4-4-12 timings. Jedec standards for PC2-5300 (DDR2 667) call for 5-5-5 timings, but our Corsair DDR2 667 was rated at the faster 4-4-4, at 667, which already shows improving timings for DDR2.

Memory	Number of DIMMs & Banks	DIMM Size Total Memory	Rated Speed SPD Timings	Rated Voltage
Corsair XMS2 5300 PRO	2 DS	512MB 1GB	DDR2-667 4-4-4-12	1.8V
Crucial PC2-4300U	2 DS	512MB 1GB	DDR2-533 4-4-4-12	1.8V
Geil PC2-4300	2 SS	512MB 1GB	DDR2-533 4-4-4-11	1.8V
Kingmax DDR2-533	2 SS	256MB 512MB	DDR2-533 4-4-4-12	1.8V
Kingston PC2-4300 Value RAM	2 SS	512MB 1GB	DDR2-533 4-4-4-12	1.8V
Micron PC2-4300U	2 DS	512MB 1GB	DDR2-533 4-4-4-12	1.8V
Mushkin DDR2 533	2 SS	512MB 1GB	DDR2-533 4-4-4-12	1.8V
OCZ Premier PC2 4300	2 SS	512MB 1GB	DDR2-533 4-4-4-11	1.8V

The SPD timings were closely bunched around the 4-4-4 Jedec standard, and all the tested memory was rated at the standard DDR2 voltage of 1.8V. Unlike the standardized DDR memory, we received a mixed bag of single- and double-sided memory, even in the same memory capacity. Memory based on Micron chips, like the Micron, Crucial, and Kingmax, were rated 256MB single-sided and 512MB double-sided. Those based on Samsung, like the OCZ Premier, or Elpida, like the Kingston Value RAM, were 512MB single-sided. Double-sided 1GB modules based on these memory chips are therefore likely to appear in the very near future.

We also received 1GB DDR2 memory modules from OCZ, but these early samples were based on Samsung chips. OCZ has decided recently that Elpida chips will be used for their 1GB DDR2 modules, so we eliminated the 1GB DIMMs from our roundup, since they are not representative of production DDR2. While 1GB modules have remained rare and expensive in DDR memory, you will likely find 1GB memory widely available from many DDR2 manufacturers in the very near future. Several manufacturers also plan to introduce 2GB DIMMs in DDR2.

Another difference in DDR2 compared to DDR is the last memory timing number, or tRAS. As we have demonstrated in our DDR reviews, Intel 875/865 chipsets have performed best at the fastest tRAS settings that could be used on the motherboards. This was normally a tRAS value of 5 or 6. DDR2, on the other hand, has a specified tRAS of 11 or 12 in most cases. Does tRAS matter on DDR2 or have the performance rules changed with DDR2 memory timings?

tRAS and DDR2

It made little sense to test 8 new DDR2 modules at memory timings that were less than optimal for best performance. We were able to verify quickly that the traditional CAS, Ras-to-Cas delay, and RAS Precharge timings all improved performance as timings were lowered. However, tRAS at 11 or 12 is a significant departure from the best timings that we have seen with DDR on the Intel platform - it is more like the 9 to 11 optimal tRAS seen on current AMD chipsets/processors.

To determine if 11 or 12 is the best tRAS timing for DDR2, we first ran Memtest86, a commonly used memory tool that measures memory bandwidth from a custom bootable OS. Memtest86 showed absolutely no change in memory bandwidth in the range of 4 to 15 when memory timings were fixed at 4-4-4 and only tRAS was varied.

To further check our Memtest86 results, we ran a complete series of tests with Aida 32, which is also available for free download on many shareware sites. The Aida 32 memory test module is very useful because it reports both memory Read and memory Write operations as separate results. We ran both tests and also added a Total Memory Bandwidth to offer a better comparison of the test results.

Micron PC2-4300U - 4-4-4-tRAS
tRAS	Aida 32 Read	Aida 32 Write	Aida 32 Total
4	5316	2328	7644
5	5264	2338	7602
6	5289	2341	7630
7	5289	2326	7615
8	5302	2340	7642
9	5303	2349	7652
10	5314	2344	7658
11	5303	2344	7647
12	5289	2336	7625
13	5303	2339	7642
14	5264	2334	7598
15	5289	2275	7564

As you can see clearly, over the full range of tRAS 4 to tRAS 13, there is virtually no difference in memory performance when tRAS is adjusted. While there is a slight falloff at tRAS of 14 and 15, it is still minor. We can conclude that a tRAS setting of 4 to 13 makes no difference in memory performance. Since higher tRAS settings usually mean better memory stability, we can well understand why a 12 setting is standard. For our memory tests, we chose a value of 10, which represents the upper mid-range of best performance and stability. However, we did find that some high density (1GB) memory modules that were not included in our tests performed with greatest stability at a tRAS setting of 11 or 12. Overall, we had no difficulties at tRAS 10, but there is no performance penalty when using 11 or 12, if it improves the stability of your DDR2 memory.

Performance Test Configuration

With DDR2 performance testing, we had the task of assembling a new test bed for our memory tests. It is still very early in the 925X/915/DDR2 release cycle, so test bed choices may change as we find even better performing components. The DDR2 test bed represents the most stable and overclocking friendly choices that we have yet tested in the lab with the 925X chipset.

All test conditions were as close as possible to those used in our earlier DDR memory reviews. We are, however, dealing with a new PC architecture in 925X/LGA 775/DDR2, so the test bed requirements were often quite different.

INTEL 925Z Performance Test Configuration
Processor(s):	Intel Pentium 4 LGA 775 2.8GHz (800MHz FSB)
RAM:	2 x 512MB Corsair XMS2 5300 PRO (DS) 2 x 512MB Crucial PC2-4300U (DS) 2 x 512MB Geil PC2-4300 (SS) 2 x 256MB Kingmax DDR2-533 (SS) 2 x 512MB Kingston PC2-4300 (SS) 2 x 512MB Micron PC2-4300U (DS) 2 x 512MB Mushkin DDR2 533 (SS) 2 x 512MB OCZ Premier PC2 4300 (SS)
Hard Drives	Maxtor MaXLine III Serial ATA 250GB 7200RPM drive 16MB Cache, NCQ
PCI/PCIe Speed	PCI Fixed at 33, PCIe and SATA Not Fixed
Bus Master Drivers:	Intel Chipset Driver 6.0.0.1014
Video Card(s):	nVidia 6800 GT PCI Express, 256MB, 1024x768x32
Video Drivers:	nVidia 61.45 Beta
Heat Sink Fan:	Thermaltake Jungle512 (CL-P0037)
Power Supply:	OCZ 520ADJ PowerStream 520 Watt
Operating System(s):	Windows XP Professional SP1
Motherboards:	Abit AA8 DuraMAX (925X) with 1.03 Beta BIOS

New Test Components

Processor

While we report the CPU used for memory testing as the 2.8E LGA 775, the processor is a 3.6ES LGA 775 set to a lower 14 ratio. This particular LGA 775 CPU has demonstrated the ability to run 4.0GHz at near default voltage. This allows a range of available FSB at the lowest 14 multiplier to about 285 or DDR2 760. The production 2.8E LGA 775 has the same 14 multiplier and overclocking results should be comparable on motherboards that have developed methods to bypass the 925X/915 10% overclock lock.

Motherboard

In the first report of manufacturers finding ways to circumvent the Intel 10% overclock lock, we found that the Asus P5AD2 was able to reach a top overclock of 248. This is currently slightly below the memory timing of 250 required to run memory at DDR2 667, at the standard 3:4 Intel memory ratio for DDR2. Further testing for an upcoming 925X motherboard roundup revealed that the Abit AA8 was capable of reaching 258 FSB setting with a stock PCIe video card and SATA hard drive. The Abit is capable of reaching even higher overclocks with an IDE hard drive and/or PCI video because we have found the current limitation to be a result of a non-locking PCIe/SATA bus speed. Abit is working to resolve this issue, if it is resolvable with the 925X chipset, and an updated BIOS with a fixed PCIe/SATA clock should be capable of even higher overclocks. Because the Abit could reach DDR2 667, it was used as the motherboard for memory testing.

Power Supply

The Power Supply for our new test bed is the remarkable OCZ PowerStream 520, which you will see in a future article by Kris Kubicki, the AnandTech power supply guru. While on the expensive side, the OCZ PowerStream 520 has individually powered rails, like the Antec TruePower units. The rails are individually adjustable with LED readouts for special tweaking needs. PowerStream features high wattage, beefed up connectors for video and hard drive, SATA connectors, and both the 24-pin ATX connector required for 925X/915/server and a 20-pin adapter for a standard 875/865/Athlon system. This may be overkill for an average end user, but these are excellent features for a memory test bed where overclocking, power-hungry processors, and power-hungry video is the norm.

Heat Sink Fan

The Thermaltake Jungle512 (CL-P0037) is one of the first third-party LGA 775 heat sinks to appear in our lab. It appears to do a decent job of cooling with the demands of the LGA 775 at overclocked speeds.

Benchmarks & Test Settings

We ran our standard suite of memory performance benchmarks, plus we added two new benchmarks to our test suite. Quake3 is an excellent game for memory benchmarking, since it is very sensitive to memory timings, but it is an older Open GL game that is becoming a bit long-in-tooth for benchmarking. While we have continued to run and report Q3 results, we have added the more recent Return to Castle Wolfenstein - Enemy Territory to our game benchmarks. RCW-ET has also proven sensitive to changes in memory timings and it is also a stock benchmark in current video testing.

We have also added the synthetic memory benchmarks in Aida 32. This freeware memory test reports both memory read and write results, allowing a more in-depth look at the operations that comprise memory bandwidth. We have also added a Total Aida 32 result to provide an easier comparison of total memory bandwidth.

The following settings were tested with all eight DDR2 memories in the roundup:

1. 800FSB/DDR2-533 - the highest stock speed supported on 925X/915 motherboards.
2. 1000FSB/DDR2-667 - a standard speed rating that represents a future memory speed for DDR2 on the Intel platform
3. 1032FSB/DDR2-686 - the highest setting currently usable on the test motherboard with PCIe Video Card and SATA Hard Drive

Since the Abit AA8 is presently limited to a 258 setting for FSB, we were not able to test the highest memory speed with the DDR2 memory modules. This is a current limitation of all 925X motherboards, since the Abit reaches the highest speeds that we have found so far on a 925X motherboard. The Asus P5AD2 is a very close second, with a highest FSB setting of 248. We will continue to monitor the development of overclocking capabilities on 925X motherboards, and we will update Highest Memory Speeds in an upcoming article as soon as we can find a BIOS or motherboard that can fully test the high-speed capabilities of DDR2.

Corsair XMS2 5300C4PRO

Corsair is well known as a producer of Enthusiast memory, so it came as no surprise that Corsair is first to market with DDR2 memory rated as the next standard of PC-5300 or DDR2 667. The timing of Corsair's introduction, however, was a bit of a surprise, just two weeks after Intel's introduction of the new DDR2 technology. With the difficulty of getting DDR2 to market, DDR2 667 was not expected to appear so quickly - especially at the aggressive 4-4-4 timings rather than the JEDEC 667 timings of 5-5-5.

Test DIMMs were a matched pair of XMS2 PRO with the activity LEDs.

We first saw the LEDs in our review of Corsair XMS4000 PRO and again, in the review of Corsair DDR400 2-2-2 that was recently introduced. Corsair uses the PRO designation to indicate LED activity lights on the memory.

While Corsair did not reveal the manufacturer of the memory chips used in their new XMS2 667, we suspect that the memory is based on specially binned Micron DDR2 533 chips, which were introduced about two weeks ago. The performance across the board matches very closely to Micron and Crucial DDR2 memory tested in this roundup.

Test Results: Corsair XMS2 5300C4PRO

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Corsair XMS2 5300C4PRO (DDR2 667) - 2 x 512Mb Double-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	3-3-3-10 1.85V	371.4	INT 3196 FLT 3232	INT 4944 FLT 4945	107
667DDR 1000FSB	4-4-4-10 1.85V	458.0	INT 3867 FLT 3898	INT 6170 FLT 6148	86
686DDR 258FSB	4-4-4-10 1.85V	471.7	INT 3979 FLT 4022	INT 6322 FLT 6305	83

Corsair XMS2 5300C4PRO (DDR2 667)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.4	5451	2280	7731
667DDR 1000FSB	97.5	6627	2770	9397
686DDR 258FSB	100.3	6836	2878	9714

Since there is no motherboard support at this time for a standard DDR2 667 memory, Corsair has programmed the SPD to 4-4-4-12 timings at DDR2 533. These are actually the rated timings for DDR2 667 performance. It is clear that XMS2 667 performs significantly better than 4-4-4 at 533, performing with faster 3-3-3 timings at 1.85V with no effort at the 533 speed. Corsair easily met the DDR 667 4-4-4 specifications at DDR2 667, at the same low voltage of 1.85V.

It is not surprising that the Corsair 667 had no problem running at the highest speed that the Abit AA8 currently supports at DDR2 686. Since this was accomplished at the same 4-4-4 timings and 1.85V, it certainly is logical to expect that the XMS2 667 is capable of even higher speeds than we can currently achieve on 925X motherboards.

Crucial PC2-4300U

Crucial is basically the retail sales arm of Micron, one of the world's largest manufacturers of memory. The Crucial DDR2 test modules were, therefore, essentially a retail version of the same Micron engineering samples supplied with the Intel 925X/915 launch kits from Intel. This allowed verification of the performance and also gave us a very good idea of the normal variations that you might find in memory modules based on the Micron memory chips.

Test DIMMs were a matched pair of double-sided 512MB DDR2 DIMMs without heat spreaders. The Crucial PC2-4300 appeared identical to the Micron PC2-4300 that we also tested.

Not surprisingly, Micron memory chips were used in the Crucial DDR2 533 DIMMs.

These are basic production DDR2 Crucial dimms, and they are not specially binned for top performance. Crucial has also hinted that they will soon be producing an even higher performing version of DDR2 memory based on specially selected higher speed micron DDR2 chips.

Test Results: Crucial PC2-4300U

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Crucial PC2-4300U (DDR2 533) - 2 x 512Mb Double-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	3-3-3-10 1.9V	372.6	INT 3185 FLT 3244	INT 4949 FLT 4952	107
667DDR 1000FSB	4-4-4-10 1.9V	459.4	INT 3766 FLT 3833	INT 6167 FLT 6158	86
686DDR 258FSB	4-4-4-10 1.9V	472.4	INT 3997 FLT 4024	INT 6315 FLT 6318	83

Crucial PC2-4300U (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.6	5437	2280	7717
667DDR 1000FSB	97.5	6639	2738	9377
686DDR 258FSB	100.5	6839	2916	9755

It was no surprise that CorsairXMS2 667 performed without incident at DDR2 667 and the highest Abit AA8 setting of DDR2 686. However, it was a complete shock that basic production Crucial DDR2 533 also ran without incident at DDR2 667, at the same aggressive 4-4-4-10 timings, and at the highest speed DDR2 686 with the same timings and modest voltage settings.

Geil PC2-4300

Geil has been very active in introducing a full range of Enthusiast memory over the past year. This was even clearer when Kingston and Geil were the first two manufacturers to supply DDR2 memory for testing.

Test DIMMs were a matched pair of single-sided 512MB DDR2 DIMMs with lightweight aluminum heatspreaders. The Geil aluminum heatspreaders appear just as effective as others in this roundup, but they are much lighter than some other solutions. Geil also includes the temperature-sensitive label on the DIMM to report heat-sink temperature. While a good idea, the labels are all but hidden away when memory is mounted and they are, therefore, not very useful for truly monitoring memory temperature. They are nice to look at and talk about, however, and screen-based temperature monitoring will probably make its way to Enthusiast memory, eventually.

We do not know what memory chips are used by Geil in their memory, and the performance in our memory tests was unique compared to other tested memory.

Test Results: Geil PC2-4300

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Geil PC2-4300 (DDR2 533) - 2 x 512Mb Single-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	4-4-4-10 1.8V	363.4	INT 2756 FLT 2828	INT 4854 FLT 4846	108
667DDR 1000FSB	5-4-5-10 2.0V	449.6	INT 3347 FLT 3422	INT 6016 FLT 6020	87
686DDR 258FSB	5-5-5-10 2.0V	458.0	INT 3280 FLT 3387	INT 6130 FLT 6128	85

Geil PC2-4300 (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	77.5	5226	2059	7731
667DDR 1000FSB	96.6	6466	2303	8769
686DDR 258FSB	98.3	6857	2515	9102

Geil is an early DDR2, so we somewhat forgive the fact that across the board, it required slower timings than other DDR2 in this roundup. The memory bandwidth is also at the bottom of performance in almost every test. This is more likely the result of the early DDR2 chips used by Geil, and we are confident that current production is likely using updated DDR2 chips with better performance.

Despite the fact that this is single-sided memory running older chips at poorer memory timings, the Geil PC2 4300 also ran without incident at DDR2 667 and at the highest speed DDR2 686 that the Abit AA8 could achieve. While the Geil memory did require a bit more voltage to reach 667 and 686, it ran all of our tests without a problem at both 667 and 686. Things are certainly looking good for 667 speed on DDR2.

Kingmax DDR2-533

Kingmax may not be as familiar to US readers, but Kingmax is a very large supplier of memory products in Asia. You will also find Micron chips in many Kingmax products, and that is the case with the Kingmax DDR2-33.

Test DIMMs were a matched pair of single-sided 256MB DDR2 533 DIMMs without heat spreaders. We would have preferred Kingmax to supply 512MB samples for testing, but these 256MB modules will provide a rough idea of the performance penalty, if any, with 256MB single-sided modules. You may recall in our Searching for the Memory Holy Grail - Part 2 article that single-sided modules performed more poorly than double-sided on the 875/865 chipsets.

As already mentioned, the top-performing Micron memory chips were used in the Kingmax DDR2 533 DIMMs.

Test Results: Kingmax DDR2-533

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Kingmax DDR2-533 (DDR2 533) - 2 x 512Mb Single-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	3-3-3-10 1.85V	369.4	INT 2955 FLT 3038	INT 4903 FLT 4903	107
667DDR 1000FSB	4-4-4-10 1.9V	454.6	INT 3484 FLT 3605	INT 6076 FLT 6077	86
686DDR 258FSB	4-4-4-10 1.9V	466.7	INT 3555 FLT 3661	INT 6245 FLT 6247	84

Kingmax DDR2-533 (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.2	5424	2177	7601
667DDR 1000FSB	97.1	6608	2600	9208
686DDR 258FSB	99.7	6754	2645	9399

Since Kingmax uses the same Micron chips as Micron, Crucial, and Corsair, the performance should be at the top of the charts. Generally, the Kingmax is one of the better performers in this roundup, but the 512MB total memory and/or single-sided configuration holds Kingmax back from the top of the charts. Since Crucial and Micron had no trouble with 667 and 686 memory speeds, it is not a surprise that Kingmax performs at the same timings and low voltages.

Kingston PC2-4300 Value RAM

As one of the largest retail and OEM memory suppliers, Kingston produces a huge amount of DIMMs for the world market. As with other finished memory makers, Kingston uses chips from chip manufacturers like Micron, Samsung, Elpida, and others. They then assemble the memory from selected or custom PCBs, and market the finished memory product in the retail and OEM channels.

Kingston supplied a pair of Value RAM for the DDR2 roundup. Value RAM is among the most reasonable of the Kingston memory offerings, and they are not specially selected chips such as those used in the Kingston HyperX series, for example.

Test DIMMs were a matched pair of single-sided 512MB DDR2 PC 4300 DIMMs without heat spreaders.

Kingston uses the high-density Elpida chips in their DDR2. This means that they can also easily produce a double-side 1GB version using the same Elpida chips.

Test Results: Kingston PC2-4300 Value RAM

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Kingston PC2-4300 (DDR2 533) - 2 x 512Mb Single-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	4-3-3-10 1.8V	367.2	INT 2877 FLT 2952	INT 4886 FLT 4890	108
667DDR 1000FSB	4-4-4-10 1.9V	456.1	INT 3524 FLT 3560	INT 6064 FLT 6084	86
686DDR 258FSB	4-4-4-10 1.9V	465.8	INT 3650 FLT 3693	INT 6232 FLT 6235	85

Kingston PC2-4300 (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.0	5353	2008	7361
667DDR 1000FSB	97.3	6543	2392	8935
686DDR 258FSB	99.5	6796	2494	9290

The Kingston Value Ram was an average performer in our tests. We do not know whether this is a result of the single-sided design of the 512MB modules or the result of the Elpida chips performing somewhat poorer than the Micron chips used in Crucial, Micron, Kingmax, and Corsair. The performance is about what you would expect from a bargain DIMM, except for the fact that even the Kingston Value ran without incident at DDR2 667 and at the highest speed DDR2 686, at the same timings and modest voltage settings. The Kingston required slightly slower 4-3-3 timings at DDR2 533 than the DIMMs with Micron chips, but the timings and required voltage at 667 and 686 matched the best in our roundup.

In case the developing message is not clear enough, every memory that we have tested so far in this roundup is able to run at DDR2 667 and DDR2 686 with complete stability. In almost every case, the 667 timings required are 4-4-4 instead of the slower JEDEC 5-5-5 standard. The voltage required for 667 and 686 operation is almost a modest increase to 1.9V from 1.8V in most cases.

Micron PC2-4300U

As one of the largest memory manufacturers, you will likely find Micron DDR2 memory used by OEMs and Micron chips used by many manufacturers in producing DDR2 memory. Micron is not really a memory retailer, but they do sell Micron products through their retail arm, under the Crucial brand name. That means these same DIMMs are available from Crucial for direct purchase.

The Micron PC2-4300U were included with Intel's launch kits for the 925X/915 as representative of DDR2 memory. They appear identical to the Crucial PC2-4300U included in this roundup and the performance is virtually the same.

Test DIMMs were a matched pair of double-sided 512MB DDR2 DIMMs without heat spreaders.

Are you surprised that Micron memory chips are used in the Micron DDR2 533 DIMMs?

While the Micron PC2-4300 are engineering samples, they appear to be the same as Crucial retail PC2-4300U memory. Performance is also essentially the same, confirming that these Micron DIMMs are representative of the retail chips available in the same configuration.

Test Results: Micron PC2-4300U

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Micron PC2-4300U (DDR2 533) - 2 x 512Mb Double-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	3-3-3-10 1.8V	373.3	INT 3209 FLT 3236	INT 4942 FLT 4937	106
667DDR 1000FSB	4-4-4-10 1.9V	459.0	INT 3746 FLT 3835	INT 6182 FLT 6162	86
686DDR 258FSB	4-4-4-10 1.9V	472.9	INT 3980 FLT 3990	INT 6329 FLT 6333	84

Micron PC2-4300U (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.7	5479	2284	7763
667DDR 1000FSB	97.7	6629	2767	9396
686DDR 258FSB	100.6	6837	2871	9708

We have found in this roundup the same Micron chips in Micron, Crucial, Kingmax, and Corsair. These are also the memories that are topping the performance charts in almost every benchmark. While it is still very early in the development of DDR2, it certainly appears at this point in time that the best memory bandwidth is found with memory built with Micron memory chips.

Mushkin DDR2 533

Mushkin is well known as a quality supplier of memory directly to the Enthusiast market. You can order Mushkin memory on-line, and you also find Mushkin appearing as an OEM supplier for memory in some Enthusiast systems.

Test DIMMs were a matched pair of single-sided 512MB DDR2 DIMMs with distinctive purple heat spreaders.

While we are not certain which memory chips were used by Micron in the manufacturing of their DDR2 533, performance and configuration suggest either Samsung or Elpida high-density chips.

Mushkin does not consider these DDR2 533 to be high performance modules. They are the first Mushkin DDR2 memory, and Mushkin will likely provide DDR2 memory aimed at the Enthusiast as the market develops.

Test Results: Mushkin DDR2 533

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

Mushkin DDR2 533 (DDR2 533) - 2 x 512Mb Single-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	3-3-3-10 1.9V	370.4	INT 2936 FLT 3025	INT 4854 FLT 4846	107
667DDR 1000FSB	4-4-4-10 1.9V	454.1	INT 3516 FLT 3574	INT 6016 FLT 6020	86
686DDR 258FSB	4-4-4-10 1.9V	468.0	INT 3615 FLT 3672	INT 6130 FLT 6128	84

Mushkin DDR2 533 (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.5	5342	2135	7477
667DDR 1000FSB	96.9	6568	2601	9169
686DDR 258FSB	99.9	6732	2668	9400

While the Mushkin is an average performer in our performance tests, it joins our other tested memories in running at DDR2 667 and DDR2 686 without much effort. Once again, we find the Mushkin 533 runs at DDR2 686, at 4-4-4-10 timings and at 1.9V. We can only guess the highest speed of any of these DDR2 modules, since they all reach DDR2 686 with reserve timings and voltages for even higher speeds.

OCZ Premier PC2 4300

OCZ is a well-known supplier of Enthusiast DDR memory. However, OCZ is still working on their performance versions of DDR2. We expect to receive a DDR2 667 version in the next couple of days that is presumably based on the same high-performing Micron chips that topped these performance tests.

For this roundup, OCZ supplied their Premier PC2-4300. The Premier line is OCZ's budget memory, comparable to value lines from other manufacturers.

Test DIMMs were a matched pair of single-sided 512MB DDR2 DIMMs with copper-colored heatspreaders.

The memory chips are Samsung high-density, and OCZ also provided a pair of 1GB DIMMs for testing. We did not include the 1GB OCZ PC2 4300 DIMMs in this review because we were told recently that the production version of the 1GB DIMM would use different memory chips. We saw no point in including DIMMs that would not be representative of the retail OCZ product.

Test Results: OCZ Premier PC2 4300

The full suite of benchmark tests were run at all memory speeds. This includes Quake3, Super PI, Sandra 2004 SP2 Memory Tests, Aida 32 Memory Tests, and Return to Castle Wolfenstein-Enemy Territory. We also ran UT2003, Aquamark 3, and Comanche 4 at every memory speed to verify stability of the reported memory timings. All benchmarks and additional tests had to complete without incident for the memory settings to be considered stable.

OCZ Premier PC2 4300 (DDR2 533) - 2 x 512Mb Single-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
533DDR 800FSB	3-3-3-10 1.85V	370.5	INT 3001 FLT 3030	INT 4897 FLT 4897	107
667DDR 1000FSB	4-4-4-10 1.9V	454.1	INT 3457 FLT 3546	INT 6089 FLT 6084	86
686DDR 258FSB	4-4-4-10 1.9V	468.0	INT 3593 FLT 3655	INT 6266 FLT 6274	84

OCZ Premier PC2 4300 (DDR2 533)
Speed	RCW-ET Radar	Aida 32 Read	Aida 32 Write	Aida 32 Total
533DDR 800FSB	78.4	5409	2081	7490
667DDR 1000FSB	96.9	6572	2575	9147
686DDR 258FSB	99.8	6743	2664	9407

Since the Samsung chips were some of the first working DDR2 produced by the memory manufacturers, we are not surprised to find average performance in the OCZ PC2 4300 results. Having said that, it is important to note that even the older Samsung chips manage DDR2 533 performance at 3-3-3 timings and at 1.85V, and DDR2 667 and 686 at 4-4-4 and a modest 1.9V.

Performance Comparisons

Performance of the eight DDR2 memories were compared at DDR2 533 (PC2-4300), DDR2 667 (PC2-5300) and DDR2 686, which is the highest stable overclock currently supported on the Abit AA8 motherboard. Please keep in mind that the Abit is still the fastest 925X motherboard which we have tested thus far, and that both the Abit and Asus do manage to bypass the Intel 925X/915 10% overclock limiter. Full performance comparisons were made with all benchmarks at all 3 tested DDR2 speeds.

AnandTech has published extensive performance comparisons of DDR memory over the past year. We also recently published DDR/DDR2 comparisons in the launch article for the 925X/915 chipset. If you would like to compare performance results from this DDR2 roundup to DDR performance, you should refer to our extensive library of memory reviews on Anandtech. Recent Memory Performance reviews include:

Intel 925X/915: Chipset Performance & DDR2
New DDR Highs: Shikatronics, OCZ, and the Fastest Memory Yet
The Return of 2-2-2: Corsair 3200XL & Samsung PC4000
OCZ 3700EB: Making Hay with Athlon 64
OCZ 3500EB: The Importance of Balanced Memory Timings
Mushkin PC3200 2-2-2 Special: Last of a Legend
PMI DDR533: A New Name in High-Performance Memory
Samsung PC3700: DDR466 Memory for the Masses
Kingmax Hardcore Memory: Tiny BGA Reaches For Top Speed
New Memory Highs: Corsair and OCZ Introduce DDR550
OCZ PC3700 Gold Rev. 2: The Universal Soldier
OCZ 4200EL: Tops in Memory Performance
Mushkin PC4000 High Performance: DDR500 PLUS
Corsair TwinX1024-4000 PRO: Improving DDR500 Performance
Mushkin & Adata: 2 for the Fast-Timings Lane
Searching for the Memory Holy Grail - Part 2

DDR2 533 Performance

Geil appears to be using early DDR2 memory chips, since the performance of the Geil DDR2 is consistently at or near the bottom of our DDR2 tests at 533. Otherwise, we see a rather consistent pattern developing in our tests of the DDR2 memory. This is probably best illustrated by the Sandra Unbuffered memory tests, where the pattern stands out.

The three 512MB DDR2 based on Micron chips consistenly top the performance charts, sometimes switching positions, but always with Crucial, Micron, or Corsair at the top. This is followed with a 2^nd cluster of the Elpida and Samsung based DDR2 memory and the single-sided 256MB Kingmax, which also uses Micron memory chips.

DDR2 667 Performance

We would never have guessed when we set out to test the current state of DDR2 memory to find that every tested DDR2 DIMM would perform at 667 speed. This applied to older DDR2 memory chips as well as the latest Micron chips, which were introduced about two weeks ago. We no longer need to ask when we will have DDR2 that can perform at DDR2 667 because we now see that all the DDR2 we tested were quite comfortable at 667 speed. The question now is when will Intel decide to release the DDR2 667 capabilities?

DDR2 686 Performance

Every DDR2 in the roundup not only passed DDR2 667, but also reached the highest speed that we could currently support in our 925X roundup. Every memory reached 686 and it is clear that almost every DDR2 memory we tested has additional headroom.

The frustration with such great memory performance across the board is that we have no idea how fast current DDR2 really is. Does Corsair 667 reach much higher than 667? Do our leading Micron chips start separating into different performance clusters as we reach well above 686 speed? These are all questions that will need to wait until our memory test bed can reach speeds that will provide answers to them.

Conclusion

When we first received DDR2 for testing, we really thought this would be a simple review. Only Corsair and OCZ were talking about DDR2 667 and we assumed that those 2 memories would top everything, if we could find a 925X board to support DDR2 667. The rest would be behind in the pack. OCZ then told us that they planned to devote more time to EB programming of the SPD on their upcoming 667, so our review looked even simpler - Corsair 667 followed by everyone else. That is the problem with assumptions. Test results often turn everything that you expect to find upside down.

The most amazing test results came from all the new DDR2 memories that we tested. Amazing, in that even the cheapest value DDR2 ran with complete stability at DDR2 667, which is the next major speed in the DDR2 highway. In addition, every one of the eight memories that we tested in this roundup ran with complete stability at the highest FSB which we could support on any current 925X motherboard - DDR2 686. Most of the tested DDR2 even accomplished this overclocking feat with much better than standard 4-4-4-10 timings and a modest voltage increase to 1.9V from the default 1.8V.

After this roundup, the message about DDR2 is very clear. As it exists today, DDR2 is much better than we expected it to be. Intel could easily move to DDR2 667 timings tomorrow and there would be a whole range of memories available to support that speed.

Perhaps the other half of the message is that this may well be the reason why Intel implemented a 10% overclock lock on the 925X/915 chipsets. No one likes to give away anything for free, and if we were Intel, we might also not want to give customers DDR2 667 performance for free. We can only applaud Abit, Asus, Gigabyte, MSI, and other creative motherboard manufacturers who found ways to bypass the 10% overclock lock. Without this engineering feat, we might never have known that DDR2 667 was already very much a reality.

The new Intel architecture was just launched 2 weeks ago, which makes the incredible performance of the DDR2 memory even more amazing. It is still early, and there will be even more performance DDR2 introduced by the memory companies that cater to Enthusiasts. Over time, you will likely find even higher overclocks at even more aggressive memory timings. For now, we just hope to find a 925X motherboard soon that will allow us to determine what the fastest speed really is for all the DDR2 tested in this roundup.

Right now, Micron memory chips have proved to be the top performers, and these were found in Crucial, Micron, Kingmax, and Corsair memory. The Kingmax may well be the equal of the other 3, but the performance was hampered by our 256MB DIMMs, so we cannot state that with confidence. The Corsair 667, Crucial 533, and Micron 533 were all at the top in every benchmark, so clearly, these 3 are the top performers. While we applaud Corsair for being the first to bring DDR2 667 to market, the reality is that the 667 performed no better than Crucial or Micron 533 in our tests, at any speed that we could reach. Given the fact that the Corsair 667 will likely cost more, we see no reason (with our current test results) to spend more for DDR2 when Crucial/Micron PC2-4300U matches the top performance in every way.

We are pleased to award our Editors Choice for the top DDR2 memory to Crucial PC2-4300U. It is a top performer and the best retail value in this DDR2 roundup. It is exciting when a product rated at DDR2 533 performs at DDR2 667 and DDR2 686 with ease, but every DDR2 memory that we tested accomplished this feat. However, the Crucial PC2-4300U did this with the widest bandwidth and tighest timings found with any memory that we tested. This same Editors Choice applies to Micron PC2-4300U, which is the same memory for the OEM market. Micron and Crucial are to be congratulated for providing the best performance and best value currently available in DDR2 memory.

Since Corsair XMS2 5300 appears to be based on the same Micron chips, you can expect similar performance. This may also apply to Kingmax DDR2-533; though, we could not verify performance results using 256MB DIMMs.

Please keep in mind that DDR2 686 was the highest memory speed that we could test with our current 925X test bed. We may well find that the Corsair 667 and upcoming OCZ 667 will easily leave the Crucial and Micron in the dust with memory speeds well beyond DDR2 686. If they do, we will revisit our DDR2 test results. For now, however, it is difficult to complain about DDR2 533 that consistently performs at DDR2 667 and beyond with every DDR2 memory that we have tested.