Corsair XMS2 5400UL DDR2-667 scales to gigahertz

CORSAIR'S RAM is known for its unique heat spreaders, some with LED or LCD panels, and outstanding overclocking performance. The new XMS2 5400UL DDR2-667 low-latency RAM is no exception; while the official specs give it 3-2-2-8 performance at 675 MHz, some overclockers managed to push the pair of these sleek black DIMMs towards 1.2 GHz, well into DDR3 territory, on Nvidia Nforce SLI chipset.

How about 'classic' Intel 925XE and other similar chipsets? Will the speed benefits of faster DDR2 be seen there too, knowing those chipsets don't really have great timing choices for memory - at least on Intel reference boards.

Benchmark
So, I put up a configuration with P4EE-3.46 GHz on Intel 925XECV board, with both Micron 3-3-3-8 and Corsair 5400UL 3-2-2-8 memory, both running in DDR2-533 mode in sync to the 1066 MHz FSB of the CPU. First, using the older BIOS rev, I set the RAM to DDR2-667 first with recommended 3-2-2-8 speed settings. After the restart, the system didn't boot at all. So I installed the newest BIOS (last month's rev), and, guess what, there were no more DDR2-667 settings at all - I could only run the memory in DDR2-533 mode.

So I tried the fastest settings BIOS would allow: 2.5-2-2-6 - it didn't work, again. After few tries, only 3-2-2-8 worked even at the DDR2-533 mode.

According to Corsair, the CL 2.5 setting for DDR2 in Intel's BIOS is useless anyway, as there is no such setting in DDR2 spec in the first place...

Now, I ran the memory well below its true potential - so how did it run? A quick round of Sandra 2005 benchmarks answered that question.

Sandra: 3-3-3-8: 3-2-2-8
Integer MB/s: 5607: 5740
FP MB/s: 5593: 5754

So, we do get around 3% extra net bandwidth (out of 8.5 GB/s gross bandwidth on dual-channel DDR2-533). Not bad for just two minor latency changes. Sadly, this is also the maximum I could get out of 'high-end" 925XE reference board without overclocking it - lets check the Nvidia and Intel 955X in the next round! µ