With the new Intel Core i7 platform, memory makers are required to make new kits because it's using triple-channel, which only slightly enhances performance. More important is that Intel is limiting the memory voltages to 1.65V. While it was commonly seen as reasonable to offer fast memory with quite insane voltages, now all memory makers have to select better chips and finally power-consumption will come down too.
In our coverage of Intel's new platform we have tested some memory settings, but our BIOS did not allow us to use 1600MHz due to its beta status. Meanwhile Intel was working hard and has improved the BIOS with new options and increasing the memory multiplier to 14.
The overall problem with faster memory on this platform is, it requires to overclock the CPU cache, so it needs higher voltages, which also will increase power consumption dramatically. Today we'll find out if faster memory does the i7 any good.
Intel DX58SO "Smackover" (provided by Intel)
Intel Core i7 965XE (provided by Intel)
Thermalright Ultra 120 eXtreme 1366 (provided by Thermalright)
Qimonda 3GB Kit PC3-8500U (provided by Qimonda)
CL7-7-7-20 CR1 @1.50V
Corsair 6GB Kit PC3-12800 TR3X6G1600C8D (provided by Corsair)
CL8-8-8-24 CR2 @1.65V
Jetway Radeon HD3870 (provided by mec-electronics)
Seasonic S12-II 500W
Samsung Spinpoint F1 500GB
SilenX iXtrema Pro 14dB(A) (provided by PC-Cooling.at)
Cooler Master Stacker 831 Lite (provided by Cooler Master)
Intel provided a 3GB PC3-8500 memory kit which was working without problems. With the Corsair 6GB kit you will have advantages when you are using a 64bit OS. We still test with Windows XP SP3 32bit and our benches do only use up to 1.5GB.
CPUz shows us the SPD of every memory module and gives us a nice comparision. The Corsair kit does of course support XMP, which is Intels answer to nVidias EPP. Of course we, the customers, are left in the dark about many advantages of EPP or XMP. It's time that all chipset makers sit on a table and make a standard for all, so we can use our modules on any board regardless of chipset.
Corsair features quite tidy timings with CL8-8-8-24 but CR2, which does not influence performance.
As usual we check out the theoretical bandwidth with Everest:
This test shows up to 20% performance increase for the Corsair kit, but how does that relate to realworld applications?
The benches are the usual stuff, so we need not to explain them to you. The most astounding thing happened while testing the Qimonda memory. We were able to clock it to 1600MHz, of course with relaxed timings. The only drawback was, it was able to work stable with hyperthreading disabled, but the only bench were it would matter is FarCry2. This game does not show any advantages if you enable hyperthreading. As usual the x264 bench will suffer with hyperthreading enabled, so it's disabled by default. Please note the Corsair memory is always running with faster timings compared to Qimonda, so this comparison is a bit unfair, but as you can see, the numbers are very similar.
The performance that comes out of this triple kit is less than we hoped for. While the synthetic bandwidth test shows up to 20% performance gain, it does not translate in real life applications. The maximum gain we see is only a mere 1.5%. In our previous tests with the Core 2 platform we saw some more gains, but it seems the integrated memory controller is hindering any further performance gains.
The modules are quite nice and the heatspreader is well built, but as you can imagine, there is no need for any headspreader at this voltages. The good thing is that modules stay cools at all times.
We cannot justify any high speed modules for the Core i7 platform. This is a pricey platform anyway, and the performance gain is marginal. Also you need to overclock the internal L3 cache clock which will require more power for the CPU.
This 6GB kit can be bought for €198. As you can see prices for DDR3 went down considerably. On the other hand you can buy a 2GB 1066MHz CL7 module for just €30, so three of them will set you back only €90, if you can live with a marginal performance loss.