Dominator dominates the DDR2 world

---

Product: Corsair TWIN2X2048-9136C5D
Website: WWW.CORSAIRMEMORY.COM
System Requirements: Motherboard with 240-pin DIMM slots (DDR2)
Price: £274.99

---

THE DDR2 SPEC is definitely getting out of hand, and engineers from companies such as Corsair, GeIL and OCZ are using all of their engineering knowledge to squeeze more juice from the standard which was a failure since day one. Yes, I've been a vocal critic of DDR2 performance for almost three years now and it's quite pathetic that DDR2 gained performance boost over DDR1 when clock doubled, DDR-400 vs. DDR2-800, but for people living in DDR2-533 nightmare land, good performance is still a pipe dream. So, it's no wonder that people have been screaming out to get decent performance parts, and they are ready to pay dearly for it.

So, what do memory vendors have in store? GeIL was first with PC2-9600 MultiSpec kit, but availability was always an issue. Still, that memory worked like a charm at 1.2 GHz with low latencies, so first blood went to the Golden Emperor. Second came OCZ with optional water-cooled PC-9200 modules which clocked even higher, up to 1.28 GHz, as our own Fudo achieved in his review. Now we're testing Dominator modules marked PC2-9136C5, or a guaranteed clock of 1.14 GHz with CAS5 latency, but in conversations with Vivian, we were encouraged to aim even higher and try to crack the 1.33 GHz barrier.

The modules

New Dominators certainly are a head higher than the regular Corsair DDR2-800 modules

We saw the Dominator modules back in June at the Computex Taipei show, and at the time it seemed a unique proposition. Competition did catch up and probably overtook, especially with H2O cooling support, design-wise. but we wanted to check the performance of these modules, not the looks.

Corsair even offers a tri-fan memory cooler option to chill the modules, but we didn't receive the fans and so we cannot judge the amount of noise that those three tiny fans generate. The memory itself consists of prolonged PCB (Printed Circuit Board) with memory chips being placed at the bottom. Each module comes with four heatspreaders, two large ones by each side and two smaller in the middle, which are in direct contact with either chips or the top of the PCB. This reminded us of ASUS dual-PCB design from a while ago and we have to say it works pretty well.

The Dominators come in a pack of 2x1GB. Our inqpressions were pretty positive, but ever since I reviewed those GeIL 1.2 GHz "Rare as hens teeth Edition" modules, I expect even more from companies like Corsair and OCZ. Corsair however, established Enhanced Performance Profile standard in cooperation with Nvidia and EPP, dubbed SLI-Memory by the boys in Green, and these modules show no difference. In any SLI-Ready motherboard, you can easily adjust the memory clocks in correlation with the "SLI-Ready" memory setting.

How times change: nForce Summer Camp combo for 2006: Zalman AM2 nVIDIA edition, Foxconn motherboard and Corsair SLI-Ready memory... well, forget about Dominators here

Nothing good comes without some compromises. Extra height puts some interesting challenges with many motherboards on the market, especially with nForce 590SLI, our chipset of choice for AMD Socket AM2. If you use a motherboard that was made with Nvidia's reference design, such as our Foxconn insertununderstandableandannoyingnamehere and own a Zalman CNPS7000, let alone 7700 or Silverstone/Thermaltake etc. high-end coolers, you're out of luck. The first DIMM slot is unusable, and thus we were forced to forfeit testing on AMD CPUs. Even Zalman CNPS9500AM2, which was launched on the same day as the nF590 was and is not compatible. So, we had to resort to a borrowed EVGA nForce 680i SLI motherboard and see how far these Dominators will go. We hope Nvidia won't screw up PCB design in the next iteration of the nForce chipset for AMD.

Going for the Gold

1.30 GHz clock achieved and benchmarked

As you will see in the graphs below, we wanted to crack the 1.33 GHz sync rate between the CPU and RAM, so we took additional measures in order to achieve the desired clock. Although Kentsfield is known for not showing much cooperation at this FSB clock, we wanted to see if our sample would prove golden.

Since our Asetek VapoChill Micro with maximum performance fan took the EE955 to a stable 4.72 GHz clock, we had enough faith that Asetek would keep the CPU chilled at a modified default 2.66 GHz. The IHS (Integrated Heat-Spreader) on the CPU was polished and coated with Zalman's new thermal paste on the CPU, and we put extra cooling on the memory and NorthBridge chip and turned the machine on. The latest BIOS was also slapped on the motherboard, but this was more in order to achieve safe S-ATA operation and avoiding nasty data corruption which could occur with the 680i chipset.

This is all we could achieve. A POST at 1.33 GHz. HDD would not be recognized and it was OC game over

However, even if we didn't achieve 1.33 GHz, stable operation at 1.3 GHz was pretty much enough to keep this setting for entire time of EVGA's board stay in Croatian branch of INQLabs. This is the fastest memory clock we have managed to achieve in DDR2 land, and seeing the performance levels at 2.60 GHz clock only tickles the interest in what kind of performance will be achieved if we clock the CPU at 3.25 GHz with sync clocked memory.

Testing, Testing...
The computer setup for testing was our default INQtest #1 system, with only change being was putting the EVGA motherboard instead of the D975XBX2 "Bad Axe 2". The XBX2 will stay a default part of my INQ testbed since I had to return the motherboard. For comparison sake, we also ran all of the tests on the Bad Axe 2 and we can confirm that the results with memory running at a DDR2-800 clock were identical.

To make long story short:
Intel Core 2 Extreme QX6700 clocked at 2.60 and 2.66 GHz
Asetek VapoChill Micro
EVGA 680i motherboard
2x1GB Corsair Dominator
Corsair HX620W Power Supply
MSI GeForce 8800GTS
24" Dell 2407WFP
250GB Seagate Barracuda 7200.8
Sony BWU-100A Blu-ray Drive
Logitech G15 Keyboard
Logitech MX-510 Mouse

System was running on clean Windows XP Professional SP2a install, Nvidia ForceWare 9.53 for nForce 680i, nTune 5.05.18.00 tweak utility and ForceWare 97.44 for the MSI's baby. The regular army of benchmarks was also provided.

EVEREST Ultimate Edition in its latest iteration shows what happens if you keep the CPU FSB at default

RightMark Memory Analyzer shows better scores when memory is running at default 800 MHz than in any of the overclocked modes. Real Memory Score?

EVEREST shows visible a performance difference if you raise the FSB as well as the memory

As you can see from the test results, if you're leaving the Intel CPU at the default FSB and cranking up the memory speed you will not be awarded with extreme performance or something similar. The whole FSB bottleneck issue Intel has right now brings back memories about probably the second best chipset of all time, although, I could even go that far and say the best due to high level of integration, namely Graphzilla's nForce 2.

The Dual-Channel memory controller inside the NorthBridge chip could offer 6.4 GB/s of bandwidth, but it could never pass the 3.17GB/s mark due to AMD's CPU EV6 front-side bus being a 64-bit single-channel at 200 MHz, so 3.2GB/s max. The winner move in that arena was the speed of memory write, which is pretty much more important than read, memory copy and of course, memory latency.

With Intel's GTL+ the situation pretty much repeated itself. Running memory at god knows how fast won't get you much of performance boost unless you raise the clock of CPU FSB as well. For the sheer purpose of testing, we have raised the FSB to 1.2GHz / 300MHz and 1.3GHz / 325MHz QDR and synced the CPU with the memory clock. The result of this test was pretty expected, when running at 2.60 GHz, our Kentie will eat every score achieved by the 67MHz higher clocked CPU but running on the default 1066MHz FSB / 266MHz QDR.

Belonging to Ripley's "Believe it or not" category is the fact that memory achieved the same scores while being passively cooled as when an 80mm fan was placed above the DIMMs. It turned out that our placement of additional 80mm fan didn't produce any difference in achieved results and Corsair was pretty much right when the company did not ship the tri-fan cooling solution with these modules.

In Short
Corsair pulled this one off the manufacturing lines right at the end of the year and goes in my mark as the fastest memory of 2006 and an excellent start to 2007. Modules were working at 1.3 GHz at "only" 2.38V voltage and posting at 1.33 GHz, sadly our Kentsfield is the problem here, which shows great potential in the design. This memory certainly caught us by surprise because I honestly did not expect this level of performance. Performance at 800MHz was excellent due to low latencies, which go way under the declared 4-4-4-15.

If you're in the market for the highest performing memory on the market, be sure to check this one out. ?

The Good
Works at 14% higher clock
Innovative cooling solution
Works at lower latencies than declared
DDR1 style latency at 800 MHz clock (CAS3, 3-3-9)

The Bad
High Price
Tough availability
Compatibility issues caused by increased height of the modules

The Ugly
It won't work on a dream AM2 configuration, but this is something that you need to be used if you choose modules with extra-height cooling such as the Dominators or PC-9200 from OCZ.

Bartender's Report
INTEL: Five out of Five Pints
AMD: Four out of Five Pints