Corsair TWIN3X2048-1333C9DHX G

DDR3 at 1600MHZ! All hail the new era!

It seems like only yesterday we were clamouring for DDR2, despite early benchmarks showing their performance to be less than amazing over DDR. But, as technology marched ever onwards, DDR was superseded by DDR2. DDR3, only now making its way into our hot little hands, will likewise toss DDR2 on to the street where it will beg for a place in anyone’s slot.

This month we welcome Corsair’s entry in the growing DDR3 memory market – two 1GB modules in a paired kit, specced for 1333MHz frequencies with 9-9-9-24 timings on a 2T command rate. The sticks are rated for a mere 1.5V, the stock voltage for DDR3, and one of the strengths of the new standard.

To bench these new sticks, we pulled out GIGABYTE’s GA-P35T-DQ6, as our regular test motherboard is based around DDR2 memory. The DQ6 also happens to be the same board we used to test Kingston’s HyperX DDR3-1375 C7 kit last month. To simulate the 1333MHz frequency of the RAM, we grabbed our Intel X6800 and lowered the multiplier from 11 to nine and wound the FSB up to 333MHz, bringing the effective speed of the processor to 3GHz (effectively simulating an E6850). Next, we manually set the CAS, tRCD, tRP and tRAS latencies to 9-9-9-24 with a command rate of 2T, the CPU to memory divider to 1:2, and the memory voltage to 1.5V.

With the system jury-rigged for DDR3, we booted up to start our benchmarks. Everest read, write and latency tests scored 7575MB/s, 6058MB/s and 67.8ns respectively. Super Pi Mod 1.5, calculating pi to four-million places (SPi 4M), managed 1min 37.719s. Rounding off the test suite with wPrime, the system completed the 32M test in 29.312s.

These results are slower then the scores obtained with the Kingston HyperX DDR3-1375 C7 kit, as was to be expected thanks to the lower timings and slightly higher frequencies of the Kingston kit. It should be noted however that the lower timings showed better scores in areas such as memory read and write and a sizeable decrease in latency response results, with Super Pi seeing a slight impact and wPrime still being primarily influenced by the CPU while running the 32M test.

The next step was to overclock the modules and see what true performance they held. The FSB was raised from the previously set 333MHz to 400MHz, the processor multiplier lowered to seven, and the memory divider left at a 1:2 ratio. While this lowered the CPU speed to 2.8GHz, it pushed the RAM frequency to 1600MHz. The sticks were more than happy to run at the higher speed on their default timings.

The DDR3-1600 frequency is a milestone within the DDR3 ranks and, while in the coming months we will see sticks released at these frequencies and boasting tighter latencies, for an entry-level piece of kit to hit this speed currently is an achievement to be noted.

At 1600MHz, the modules obtained 9201MB/s, 7202MB/s and 58.6ns in the Everest read, write and latency benchmarks. In Super Pi 4M they scored a time of 1min 39.812s and 31.281s in the wPrime 32M benchmark.

As you can see, the extra 267MHz made a massive difference. Further overclocking yielded instability at 1650MHz and not a great deal of extra performance for the resulting flakiness.

With the recent news that the Kingston HyperX PC3-11000 C7 kit has an RRP of $1395, compared to the $735 of Corsair’s here, make it extremely attractive, even with CAS 9-based timings.

At stock, the modules don’t particularly shine next to the HyperX hardware, but when overclocked, the increased bandwidth negates the impact of the higher latencies.