We won't go over the details of the differences between DDR2 and DDR3 in detail, since it's been thoroughly covered a large number of times already. But we would like to comment that comparisons directly between DDR2 and DDR3 should be approached with caution, as the architectures are quite different, and the way that the clocks and latencies interact is also different. Clock speeds are higher in DDR3, and thus you may, if more familiar with DDR2 latencies, think that the timings in DDR3 chips are high.
This is a good segway for us to a quick comment on memory performance, and how timings and clock speeds interact to provide memory performance. Think of it 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 twice as long to start running (twice as high latency) but catches up 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.
With Bulldozer on the way for the 990FX platform, Patriot wanted to ensure that the best possible cooperation between the RAM and the new platform was achieved. AMD tends to be a little pickier with memory, and so Patriot brings their latest of what ought to become a robust offering AMD-tuned memory kits, in order to maximize the clocks that can be obtained from your memory. These memory improvements can also help squeeze the most out of a CPU overclock.
Our tests involve first testing the claimed default timings, in the case of our current Gamer 2 modules, they claim timings of 7-7-7-20, so we will set these timings, and run stability and benchmark tests to ensure proper operation, and measure performance.
The procedure will to (try) to boot into windows, and close any non-essential processes which are running, as to best isolate the RAM and "open" its potential. We will do this for each setting we make for the memory in the BIOS.
The stock timings of 7-7-7-20 were accepted without any trouble whatsoever, and we didn't have to change any other settings in the BIOS of our Gigabyte GA-990FXA-UD5 motherboard to get it to do so. We stress tested the settings for stability using MemTest86 and Prime 95, and we didn't observe any signs of instability with these settings. Many budget RAM modules won't even hit their specified timings, so this is a good start, as an assurance of quality.
Next, we were hoping to hit something in the 1800+Mhz range, and weren't successful in doing so in multiple BIOS settings and combinations on our Gigabyte GA-990FXA-UD5. So we bumped it back to 1600MHz and began ticking up the CPU Frequency. The highest we were able to boot into stable, and survived Memtest86 and Prime 95 stress and stability tests was at 1664MHz. At 1664MHz, we were then able to lower the timings to 9-9-9-29, which is much better than the AUTO setting's 11-11-11-29. This really isn't too bad of an overclock for a 1333MHz kit, but we were hoping for something in the 1800MHz range, like we were originally shooting for. This leads us to a few things we should keep in mind with these results:
-We are using an early model/BIOS motherboard in the 990FX platform which is still rather infant
-The 990FX platform is currently a bit limited due to the lack of the bulldozer chips it was intended for, and will open up further RAM OC support
-We have an early bin RAM module, and they may improve over time as well
Another potential area of overclock improvement is that we did not push the voltages as hard as we could as we had thermal concerns in our test bed during testing (high ambients), and we will try to see if the RAM can be pushed to the 1800MHz+ range with higher voltages and update our results when we get a lower ambient condition.
As we can see, we saw some pretty good gains in our SiSoft Sandra benchmarking, increasing from 12.0GB/s to 12.4GB/s, and a pretty healthy increase across the board.
We also ran AIDA64 for memory read and write benchmarks, and the comparison is in the table below:
|1333MHz 7-7-7-20 Timings||8655MB/s||7416MB/s||10625MB/s||49.5ns|
|1664MHz 9-9-9-29 Timings||9405MB/s||7675MB/s||11325MB/s||47.3ns|
The increases are also fairly pronounced in the AIDA64 benchmark as well. It's really not a bad overclock at all from a 1333MHz stock kit, and certainly allows for a nice boost for your money initially. Other chipsets may prove better for the Gamer 2 AMD Black Edition until Bulldozer arrives, and we will perform a new test when we get a Bulldozer chip in hand, which will hopefully open up the juvenile 990FX platform up.
With the Bulldozer chips on the way, it is also likely we will see a fresh crop of AMD-designed RAM modules, and hopefully they will lead to better overclocking potential that the Bulldozer chips are promising.
We will further test these modules when Bulldozer arrives, and try some more extreme voltage changes and higher clocks when ambients decrease.
It is also worth noting that it has been shown that clocks over 1600MHz in RAM starts to reach the point of diminishing returns. That is, as you go up, the linearity of the performance gains to continues to decrease, and you see less performance gains as the clock speeds increase, so being able to easily reach 1664MHz is definitely nice in that regard.