Patriot Memory Viper 3 Intel Extreme Masters 16GB (2x8GB) 1866MHZ RAM Kit Review - Test Setup and Overclocking
CPU: Intel i5-3570K
CPU Cooling: Cooler Master TPC 812
Motherboard: Gigabyte Z77X-UD3H
OS HDD: OCZ Agility 3 120GB SSD
Secondary HDD: Kingston HyperX 3K 240GB SSD
OS: Windows 7 Professional 64-Bit
If there is one thing every PC builder knows it is that every part does not always work perfectly, especially when it comes to RAM. We dealt with this during our testing of the IEM DIMMs as one of them was DOA, but Patriot's RMA process was a breeze and we had a replacement kit in no time. This appears to be an unlikely issue as the hand-tested DIMMs must have seen some damage elsewhere down the line.
Our testing methods will be fairly simple. We're first going to test the RAM at default settings (optimized settings for the motherboard), what you'd get if you didn't touch a thing after installing the sticks: 1600 MHz, timings 9-9-9-24. Then we used Intel's XMP (Extreme Memory Profile) to auotmatically overclock the RAM, which resulted in a speed of 1886 MHz and timings of 10-12-10-30. Lastly, we manually overclocked the modules to 2018 MHz using 10-10-10-28 as timings at 1.625V. The XMP profile increased the speed by uping the timings sightly, the easiest and most reliable way to overclock RAM. With a faster bus and faster timings in our overclock profile, we should see a decent performance increase in our benchmark testing.
Out-of-the-box specifications tend not to push DIMMs, and performance is generally fairly consisten across manufacturuers and OEMs. The only difference then in terms of performance gains is what can be realized through overclocking. We'll make a setting to see how highly they can clock, and test to see what the maximum stable settings will be. Memory performance will vary substantially from build to build, so don't expect ours to be necessarily the beset you'll find, although we'd expect more consistency with the hand-tested Intel Extreme Masters kits.
We will be hold the CPU frequency constant at 3.8GHz, disable SpeedStep, Turbo Boost, and power-saving features on the i5-3570K, as the speed of the memory is directly tied to the processor frequency. For default and XMP, the voltage settings for the processor and memory were left as an "AUTO" setting, (which remains at the kit's default 1.50V). We increased the voltage to 1.625V to get a stable overclock at 2018MHz. We'll then run several synthetics to gauge the performance of the memory, including AIDA64, SiSoft Sandra, and PCMark 7.
These modules were tested by Patriot to run at 1866 MHz with 10-10-10-30 timings, which is the low end of overclocking RAM. We used Intel's XMP to overclock safely and automatically, as many consumers would do if they wanted a little more performance without any significant risk. Manual overclocking certainly yeilded more performance, but it took some tweaking to get there. After tweaking the timings and voltages iteratively until we found a good, stable overclock, we arrived at 2000 MHz with 10-10-10-28 timings. We had to bump the voltage to 1.625 V to achieve optimal stability, but this is a small bump from the 1.5 V the sticks default at. This is a great little boost, as the 2000MHz threshold is rather difficult to obtain. We're begining to see more kits at 2133MHz and a few even higher than that, but 2000 MHz isn't bad, and from our research others are consistently seeing similar overclocking headroom.
As a quick aside, a note on memory performance and how adjusting the frequency versus timings comes into play. Think of it as a race, where one racer is twice as fast as the other. The slower one gets to go first (lower latency, less wait time) but runs less fast. The faster runner (higher clock speed), has to wait a little bit longer to start running (higher latency) but catches up and passes with greater speed (higher clock). Now lets, for simplicity's sake, assume they arrive at the hypothetical finish line at the same time, both racers have the same relative speed. So with this analogy we can see that overall RAM speed isn't remarkably simple, and is a "tango" between clock speed and latency. Generally speaking, the faster runner will finish first, but often to increase that speed, he must take longer "rests," which somewhat offsets his gains. This is what we see by the diminishing return in memory overclocking.