OCZ Technology recently released their DDR2 PC2-9200 Flex II series kit that consists of two 2GB modules. OCZ rates these particular modules at DDR2-1150 with timings of 5-5-5-18 at 2.10V on higher end P35 and X38/X48 motherboards. Besides the impressive speed ratings, these modules feature a revised thermal management system that consists of a new heatspreader with dedicated cooling channels directly over the ICs along with two 1/4” ID barbs for attaching a liquid cooling setup. The kit contains 3/8” and 1/2” barb adapters, four-way manifold barb, and 40” of plastic tubing for those who want to cool the memory a different way.

The cooling channels and manifold are made from aluminum so a user with copper devices in their water cooling system will need to use an inhibitor to ensure the differing metals stay at détente during operation. Due to the size of the heatspreaders, the modules cannot be placed side-by-side, thus limiting memory capacity to 4GB on most boards.

In practice, we found utilizing air-cooling was just as effective for reaching our maximum clock speeds as using chilled water, even though temperatures were up to 6C lower when chilled. The primary reason for this is the maximum voltage guaranteed by OCZ is 2.15V. The ICs are from PSC and OCZ highly bins these particular chips to ensure DDR2-1150 capability on supported boards. As such, any voltages over 2.10V in testing did not result in any additional speed increase or timing decrease. In fact, our maximum voltage utilized at DDR2-1200 was 2.08V on the ASUS P5Q Deluxe board.

We are going to cut to the chase with today’s sneak peak and will only be presenting our maximum stable clock results with the PC2-9200 Flex II 4GB kit on the P5Q Deluxe board sporting the new P45 chipset. During testing for our 14 module, 11 different suppliers 2x2GB DDR2 roundup, we had a couple of kits that stood out from the rest; this kit was one of them from a clocking standpoint.

We also had success with running this kit at low voltages up to DDR2-900 (1.7V at 5-4-4-10), but will save those results and others for the roundup. Today’s preview will just answer the maximum clock question and ensure that OCZ’s DDR2-1150 claims are indeed true. Our system setup consists obviously of the Flex II kit, Intel E8500, ASUS P5Q Deluxe, WD 640GB HD, a couple of optical drives from Sony, and our lab favorite Zotac 9800GTX AMP! Video card. Cooling our E8500 at 4.3GHz on a 24/7 basis was not going to occur with the retail heatsink, so we employed the CoolIT Systems Pure CPU cooler, which surprised us by keeping our CPU significantly cooler and quieter than our standard heatsink.

Let’s take a quick look at our maximum clock results today.

Our primary focus today is finding the maximum stable overclock of this memory on the P45 chipset, a chipset that will soon become Intel’s mainstream part, replacing the excellent P35 in most market sectors. We have tested this memory on several different platforms ranging from the AMD 790FX to the Intel P35/X38/X48. Our advice for getting the maximum speed out of this kit is to use it on very capable boards such as the DFI LANParty UT P35 T2R, ASUS X48 Rampage Formula, or the Gigabyte GA-MA790FX-DQ6.

Honestly, our upcoming 4GB roundup will show that DDR2-800 at 4-4-4-12 or DDR2-1066 at 5-5-4-15 will offer nearly the same application performance as this memory or others at DDR2-1100+ speeds. A user buying a $170 kit like this can count on top benchmark results, absolute best application results by a few percent at most, or the comfort of knowing this memory kit is highly configurable from a timing and voltage selection from DDR2-800 up to DDR2-1200 on the right board. If that is not interesting, then a dozen or so 4GB DDR2-800 kits priced in the $75~$110 range will more than suffice for a majority of users.

We set up our P5Q Deluxe board based on how most enthusiasts would overclock the platform for maximum performance, yet still retain 24/7 stability. While we could have overclocked the CPU to 4.7GHz and the memory to DDR2-1225, the stability of the system was only good for that all-important CPU-Z or SuperPi 1M screenshot. Instead, we took the path of optimizing our system for best possible clocks and performance at reasonable voltages and temperatures. Our FSB rate of 480 was extremely stable with a CPU multiplier at 9x that ensured an attainable/attractive CPU speed and gave us several different options for memory dividers to reach the elusive DDR2-1200 goal with a 2x2GB kit.

Zotac 9800GTX on the ASUS P5Q3 Deluxe DDR3 Motherboard...

Sure, we could have set up our system to show a maximum stable 540FSB but with a 6x multiplier that would have resulted in a ridiculous “overclock” setting. A 124MHz overclock and lightly loading the MCH in order to do a high FSB is not really overclocking in our opinion. We did not go to an extreme here either, as additional cooling and voltages would have put us out of reach of the target market for this memory. In the end, we settled on a 9x480FSB, 1.475 VCore for 4.32GHz, 2.08 VDimm for DDR2-1200, 5-5-5-15 base timings, Vmch at 1.38V, tRD set to 8 but with all phases pulled in for an effective tRD of 7, and the balance of board settings at Auto.

We found this particular BIOS was better suited to auto settings in most cases, especially with Clock Skews and GTL-Ref settings. ASUS still has some fine-tuning to do, but this chipset has made us wonder about the future of the X38/X48 chipsets at this point. Just to make things interesting, we clocked our Zotac 9800GTX to 801MHx GPU, 1200MHz Memory, and 2002MHz Shader clocks. Our initial benchmark results are just for sanity purposes today with 3DMark Vantage and Everest utilized. Our particular settings passed a variety of benchmarks from Crysis loops to PCMark Vantage loops to OCCT runs to encoding a couple of terabytes of movies to ensure stability over the course of several days.

One last item before we get to the obligatory screenshots: while we appreciated the product samples that OCZ sent us, our test results were with a retail kit purchased at Newegg. Our new testing methodology includes the random purchase of retail kits from various sources for audit procedures. This is not to imply or confirm any problems with OCZ and other suppliers; we just want to ensure the results we experience with retail product are the same as you will experience. However, just from the viewpoint of manufacturing variations, there will always be a difference in results, sometimes better and sometimes worse. In case you are wondering, our retail kit performed 4MHz higher than the sample kit.

3DVantage Results

DDR2-1200

DDR2-960

Our 3DVantage tests completed at High settings with a final score of 3935 with DDR2-1200 and 3924 with DDR2-960 settings using the Zotac 9800GTX. The scores are basically even with the GPU result going to the DDR2-1200 setting while the CPU Score was slightly higher on the DDR2-960 option.

Our Everest results showed a wider disparity with the DDR2-1200 setting having an advantage in the Read, Copy, and Latency numbers by a measurable difference. The DDR2-960 setting operated at a 1:1 memory ratio compared to 4:5 for the DDR2-1200. At DDR2-960, memory timings were 5-4-4-12 at 1.88V with tRD (performance level) set to 12. Comparatively, at DDR2-1200 the memory timings were at 5-5-5-15 at 2.08V with tRD set to 8. Both tRD values were set manually with all phases pulled in resulting in actual tRD values near 11 and 7 respectively.

Our tRD values were arrived at after numerous hours of stress testing and represent 24/7 stability results. We were disappointed in the tRD settings at DDR2-960; even on a significantly more demanding 1:1 ratio as we expected numbers near 9 which probably would have resulted in similar throughput and latency numbers. We are still working with ASUS on tuning the BIOS and additional tuning of the clock skews, GTL-Ref, and other sub-timings have resulted in a tRD setting at 10, but we have not completed stability testing yet. However, initial test results are closer to the DDR2-1200 settings.

Game/Sandra Results

We ran a few additional tests based on two of the more popular games along with Sandra Memory Bandwidth numbers. Our results speak for themselves, as DDR2-1200 is just slightly faster at the 1280x1024 resolution and basically even at 1680x1050. Although not reported, the 1920x1200 numbers are identical, as this setting becomes truly GPU dependent. This is not surprising as most current game engines are much more dependent on the GPU and to some extent the CPU for improvements in frame rates rather than subsystems like memory or storage.

Our Sandra numbers show a 20% advantage in the unbuffered test and a 24% improvement in the buffered test. We typically see a translation in these advantages (also true for Everest) in applications that are truly memory sensitive; however, even those differences are typically within a few percent of each other. Benchmarks like 3Dmark 2001SE, SuperPi, older games like Battlefield 2, and WinRAR will typically show the greatest differences compared to typical applications.

Quick Thoughts

In the end, users purchasing this memory will typically be those looking for any performance advantage and this memory delivers on OCZ’s promise of the highest performing 2x2GB DDR2 memory kit available. While the ability to reach and maintain a stable DDR2-1200 setting was impressive to us, the more impressive aspect of this memory is its ability to run at 4-4-3-10 timings on 1.80V at DDR2-800 or 5-5-4-12 at DDR2-1066 on 1.95V. Really, the greatest strength of this memory is offering a multitude of timings and reasonable voltages up to its DDR2-1200 ceiling in a 4GB package. Pardon the pun, but this memory really is flexible. We will see just how much in our next article.