Could Memory Really Be Holding You Back?
Every few years we write a story showing how most applications get the most performance out of ordinary memory. Today, "ordinary" means DDR3-1600. Recently, something changed though. We observed noticeable speed-ups in at least one game after changing a few memory settings. Since we still had the system from FX Vs. Core i7: Exploring CPU Bottlenecks And AMD CrossFire set up, we thought it'd be a perfect time to follow our CPU-oriented gaming analysis with a version more intently focused on memory.
The things we said about frequency and timings six years ago still apply today. CAS latency is still measured in cycles, DDR3-2133 cycles are still twice as quick as DDR3-1066 cycles, and so DDR3-2133 CAS 10 still has the same response time as DDR3-1066 CAS 5. Moreover, DDR3-2133 continues to offer twice the bandwidth of DDR3-1066.
Fortunately, chasing value doesn't mean you're stuck at DDR3-1066, and bandwidth-seekers are no longer stuck at CAS 10. Cheap DDR3-1600 and mid-priced DDR3-2133 CAS 9 are now the norm.
Combining those developments in memory technology with the availability of quad-channel platforms, we have the perfect chance to compare frequency, latency, and bus width. That is, we can make such a comparison without resorting to unrealistically-handicapped configurations.
We're armed with an overclocked CPU and a pair of high-end Radeon HD 7970s in CrossFire to alleviate bottlenecks. So, how much gaming performance can we extract from a few changes in memory configuration?