G.Skill TridentX DDR3 2666MHz 8GB Memory Kit Review

Once we had set the XMP profile within the BIOS, we booted into Windows and fired up CPU-Z to make sure that all of the settings we had entered were applied by the system. Once we had CPU-Z open, we could then confirm we were running an 8GB kit at 2666MHz with timings of 11-13-13-35-2T.

After we had confirmed our settings, we opened up AIDA64 so that we could take a look at the stock performance of the memory at 2666MHz. With a kit rated at this speed, we’re expecting some good numbers to come out of the memory benchmark test.

Just as we thought, the numbers were strong for a kit of this speed, with 21466MB/s read, 18956MB/s write, 23044MB/s copy and a latency of 33.7ns.

Now that we have a marker to work from, we can start our battle of trying to squeeze as much performance out of the kit as possible, while keeping the timings at their stock values of 11-13-13-35-2T. To do this, we started by increasing our i7 3770k processor to 4.5GHz to eliminate any bottleneck from the memory controller. After this, we could start to increase the speed of the memory and as this is a 2666MHz kit, the next logical step was 2800MHz, but this wasn’t looking hopeful as it’s quite a big jump.

By moving to the 2800MHz divider, it resulted in the system not booting and overclocking failing so we knew that the only way to increase the speed of this memory was via the BCLK and so we started to increase it from its stock value of 100MHz by .5 increments and so we started to do just that. We knew that we wouldn’t get too close to 2800MHz but we were still pleasantly surprised as to the results we got with 100% stability.

We ended up at 2760MHz using a BCLK of 103.5MHz while keeping the timings at 11-13-13-35-2T at 1.65V.

By increasing the speed of the modules, but keeping the timings at their stock values, we were able to push the bandwidth of the modules much further than before and this is reflected within AIDA64. We saw an increase on all tests with speeds of 25310MB/s read, 24270MB/s write, 25776MB/s copy and a latency of 30.3ns.

The next thing for us to do was to continue to push the memory, but this time by setting the timings to auto on our Asus Maximus V test board as this should allow us to push the memory even further, by allowing the board to slacken the timings off slightly. This is where we hit some problems, as our board continued to keep the timings set at 11-13-13-35-2T at 1.65V so we knew we’d have to resort to setting the timings manually.

With this in mind, we went back into the BIOS and started to loosen the timings to see if we could push things past 2760MHz.

We started by loosening the timings off to 11-14-13-37-2T at 1.65V and went back to the overclock we had before at 2760MHz and started to try and push things up from there and went to 104MHz on the BCLK. We knew we had a bit more to go and so went back into the BIOS and increased the BCLK to 104.5MHz which gave us a speed of  2786MHz at the timings we specified at the stock voltage of 1.65V. This was confirmed in CPU-Z as you can see below.

Once we double checked this, we did try to tighten up the timings once again to 11-13-13-37-2T but this resulted in a non-booting system.

We then opened AIDA64 for the last time today to see how the performance scores were reflected and saw that a bit of a mixed bag of results greeted us with the read speed dropping slightly to 25184MB/s but the write speed had increased to 24338MB/s and the copy speed increased over the 26000MB/s mark to 26022MB/s while the latency stayed the same at 30.3ns.

While we could have possibly increased the BCLK further, this would have meant having a lot looser timings and the performance would have then been hindered because of it.