Introduction :

While there have been many reviews on DDR2, I fear early reviews for the new standard were unconsciously prejudicial. Many clankingly presumed DDR2 would; "pick up, where DDR1 left off," in speed and therefore performance. Truth is DDR2 is faster, so much so motherboards must run the higher speed modules on a divider, the asynchronous results of which were disappointing compared to its predecessor DDR1 due to high latencies associated with the speed.

In order for the memory we’re testing today to run at speed and synchronously with the processor's FSB, we would have to raise the CPU FSB speed to 336MHz. Even in an extreme overclocking environment this would be asking a lot from any CPU.

Luckily, motherboard makers such as Asus can help us reach this goal. It seems Intel programmed a 14x multiplier at boot up in their newest CPU’s, and with an overclocking BIOS such as Asus’s “Lock Free" , you are able to change the multiplier on Pentium 4 CPU's from 530 to 570J (S775), as well as Prescott for S478 from 3.2E to 3.4E effectively transforming them into P4 520's or 2.8E's.


Source: TomsHardware

Our test today with Corsair’s DDR2 modules will focus on getting up to speed with this new standard (DDR2 667), and to exemplify any performance differences running the system off the lower multiplier, and attempting a straight overclock.

The benefits of DDR2 are slowly beginning to emerge as more and more End-users begin to purchase systems and familiarize themselves with this new standard. Unfortunately current PCIe motherboards are limited in their BIOS options and it's been quite the debacle. A conundrum exists in the given ability to lower the multiplier for higher FSB speeds, yet to there's no option for locking the PCIe graphics bus. The good news dispelling initial claims there was an overclocking ceiling at just 10%, can be safely laid to rest. I commend Intel Fabs, as .09-micron yields of both 478 pin and S775 BGA packaged P4's has been outstanding. I've had three of these processors including the Socket 478 3.0E, S775 P4 530 and P4 550, all overclocked a minimum of 600Mhz without a need to increase Vcore. With relatively small Vcore increases all have ran at or above 4.0GHz. Of the three, I'm most impressed with my P4 550 which easily overclocks beyond its native 3.4GHz (200FSB) to 4.0GHz (240FSB) without any increase in Vcore.

Problems begin to occur at 250FSB, however; this is primarily attributable to the absence of a PCIe lock. With the PCIe bus unlocked, at 260FSB the graphics card is running from a 120MHz bus speed. Prima facie these figures may appear benign, and while the new PCIe bus leaves Sideband Addressing by the wayside overclocking the graphic card this high can become problematic, eventually damaging the card. The inability to lock the PCIe bus becomes pertinent as we try to run our DDR2 667 at specified speed. While it's exciting to have such high speed memory, it's equally agitating if you’re unable to exploit the memory's full potential due to seemingly “simple” BIOS and motherboard deficits.

Onto our test setup ->

Test system setup :

Liquid3D's Test Setup
CPU	Intel PIV 550 SL7J8 Malay
Mainboard	Abit AA8 (BIOS 17)
Memory	Corsair Twin2X1024 5400C4PRO
Video	ATI X800 XT (256MB Cat 4.10)
Cooling	Water cooling kit from Alphacool with NexXxos XP water block
Other	Maxtor Diamond Max 9 Plus SATA 150, 120GB OCZ Powerstream 420W TTGI USA TT-561T2-BK

Benchmarks and Software used:

Windows XP SP2 with:

SiSoft Sandra Memory Bandwidth-Buffered

Futuremark 3DMark 2001SE(build 330)

AIDA32

Hexus PiFast


System testing:

For our first screenshot, I simply ran the system (memory) at default speeds, allowing the Abit BIOS to determine settings. In the Advanced Chipset menu I chose the SPD command (Serial Presence Detect) as I believe this to be most stable with recent Intel chipsets especially those running PAT. Included in the screenshot is the monitoring software Abit EQ, indicating temps, voltages fan speeds, etc. All voltages remained at default throughout, except where indicated otherwise



It's disappointing to see the memory running so far under its rated speed; unfortunately limitations in the AA8 BIOS prevent higher FSB speeds. While 260FSB was attainable using the 14x multiplier, any attempt to lock the PCI-Express bus results in a “no-post.”

From the default screen, I've provided the maximum performance screen which resulted from running the 14x multiplier at 260FSB. This required a Vcore increase from 1.38V (default) to 1.535V, surprisingly however; VDIMM nor NB voltages had to be adjusted beyond their default 1.80V for DDR2, and 1.50V for the North Bridge chipset.



This is impressive bandwidth at 6.4GB/s and begins to show DDR2's true potential. This is just 15MHz overclock from the rated speed of these modules, and at just 1.80V.

Let’s see how it the memory does in our benchmarks ->

Sisoft Sandra:



In the graph above the 260FSB benchmark was run using the 14x multiplier option. Unfortunately without the ability to lock graphics bus, or change the FSB:DRAM divider there still exists a restriction on performance. Regardless, Intel's decision to leave the 14x multiplier available compliments DDR2's higher speeds.

3DMark2001SE:

In graphics performance there was at least a 4K improvement between the default speeds of 3.4GHz (200FSB) to 4.2GHz (250FSB) in 3DMark2001 scores. I was able to run all 3D benchmarks up to 250FSB, and using the 14x multiplier to 260FSB, while I've tried and test all Futuremark benchmarks, it was 3DMark2001 which reflected memory performance increases best.



And at the other end of the spectrum, at 250FSB



And not to exclude performance using the 14x multiplier at 260FSB



AIDA 32:





As is painfully obvious from the PiFast results, DDR2’s Achilles heel is its high latencies. Making matters worse P4 Prescott core architecture pipelines are deep enough to warrant knowledge of Speleology. DDR2 will not begin to come into its own until FSB increases are implemented, such as in Intel's 1066 925EX. Until that time, appropriate implementation of FSB:DRAM multipliers will be critical.

Onto our conclusion ->

Thoughts and Conclusion:

The question as to whether DDR2 provide any real-world performance increase at this time must be viewed in respect to the PCIe system as a whole. DDR obviously has more life in it; otherwise nForce4 motherboards would be implementing the DDR2 standard.

Given the fact they'll offer Dual graphics capabilities, perhaps DDR's relatively low cost was a consideration as many will be purchasing two graphics cards, however; my contention in nVidia has made a pragmatic decision pure and simple. DDR's low latency, abundant availability, and superior performance where latency intense applications are concerned such as graphics, prevailed.

While it's true that speed and bandwidth is in DDR2's favor, and will continue to favor the new standard, tight latencies do make for some excellent graphic performance. My recommendation then, if you’re looking for pure speed with tons of bandwidth, the Intel i915X/i925XX supporting PCIe, DDR2, and 1066FSB make an excellent alternative.

Ever since I abandoned my AMD system in the latter T-Bred B days, for an Intel P4, my reasons were primarily frequency based. Yet Intel's Hyper threading technology, which despite any competition from AMD in the form of HyperTransport, has been a resounding success, and a technology I've come to appreciate. I came back to AMD when the nForce3 250Gb chipset was released, purchasing what was then AMD's fastest processor, the A64 3400, for S754. Although the system did excel in its graphics performance, it was wanting in its multitasking ability and overall system speed.

At that time the A64's on-die MCH (Memory Controller Hub) was single channel, and consumed much of the processor's energy, which created a substantial amount of heat. This severely limited the A64's ability to overclock from higher multiplier, requiring PC Enthusiasts to drop the FSB, and exploit performance from much lower speeds. Since Intel released their .09-micron Prescott for Socket 478, they've most certainly had the overclocking advantage, easily surpassing 4.0GHz with simple air cooling.

So in conclusion: DDR2 is obviously hamstrung by the platform for which it was designed. Not nearly as much as PCIe, but the standard is no where as effective as it could be. While this is a memory review, given the circumstances one cannot separate memory from the system on which it depends, especially when in its infancy.

So while I feel Corsair has produced an outstanding product which performs on par with most of their line, I'd be remiss in advising End-users to go out and buy an entirely new system just to try DDR2. There are, however; options where one can reap the bandwidth and high-frequencies of DDR2 without having to invest in an entirely new system, including “BTX” PSU.

For example ECS offers what may be one of the most versatile motherboards ever, their 915G-A. A $100 Socket 775 board which offers PCIe, AGP-express, DDR, DDR-2, Serial-ATA, Gigabit LAN, USB 2.0, and on-board 8-channel audio, all based on the Intel i915X chipset. This would allow you to try DDR2, until PCIe graphics card GPU's are truly PCIe native, while taking advantage of the more stable, cooler running Socket 775 Prescott (editor’s note: AGP Express is not suited for gaming, it’s basically an AGP 1x slot).

While PCIe/DDR2 technology continues to experience growing pains, Corsair has done an excellent job with what is arguably the fastest memory on the market. At the time of writing, Corsair Twin2X1024 5400C4PRO can be purchased for $365 at Monarch Computers.

I'd like to thank Anna and Joe at Corsair for setting us with the memory modules.

Questions/Comments: Forum thread