For memory testing we used benchtable with the following configuration:
- Motherboard: ASUS P8P67, Intel P67, BIOS 1704
- Processor: Intel Core i5–2500K, Stepping D2
- Memory: G.SKILL F3-12800Cl9F-8GBXL 9-9-9-24 1.50V 2x4096Mb
- Graphics card: MSI N570GTX Twin Frozr II/OC
- Storage drive: Western Digital WD10EADS-65L5B1
- Power supply: Chieftec APS 850W
- Case: Cooler Master Stacker 831 SE
- Microsoft Windows 7 64-bit SP1
- CPU-Z (free) tool
- AIDA64 system information and diagnosis software suite (shareware)
- LinX 0.6.4 AVX edition (64-bit)
CPU basic clock (BCLK) has been set to 100 MHz. The processor operated at default settings — during test the clock raised from 3.3 to 3.7 GHz due to Turbo Boost technology. After stability test run the CPU enabled energy saving feature which resulted in CPU multiplier drop to x16. Core voltage was set automatically. Memory timings were set manually. During all the test Command Rate was set to 1T, except for default settings — DDR3-1600 9-9-9-24. Memory voltage was set to either to 1.5 or 1.65 V.
The processor was cooled by Thermalright Ultra eXtreme 120 Black Edition CPU cooler with one Nanoxia DX 12 fan at 1200RMP. The second fan of same model was mounted on the back panel of the case to blow the hot air out. The ambient temperature during test was around +24°C. Memory modules did not require any additional cooling.
For stability check during overclocking we run LinX 0.6.4 AVX edition (64-bit) ten times with Problem size of 14000.
Memory modules were tested with the following clocks and timings:
Stability check revealed memory malfunction with settings listed below:
- 1600MHz with CL 7-7-7-28 and 1.65 V
- 1600MHz with CL 7-9-8-28 and 1.5 V
- 1600MHz with CL 8-8-8-28 and 1.65 V
All other operation modes were successful to pass stability test. Gathered information on compute power for each setting is presented as graph:
Tested G.SKILL Ripjaws X kit overclocked to about 30% and reached 2133 MHz with 1.65 Volt which is not bad at all. On the other side, LinX says it only gives merely 4% performance boost compared to default settings. Further overlooking required either memory to run at 2400Mhz or enhance CPU BCLK. Both has lead to errors during stability check.
Below you can find screenshots for each successful test passed:
Built in AIDA64 benchmark analyzes memory performance based on Read, Write and Copy operations, as well as measures Memory Latecncy. Such info makes an appraisal of synthetic task memory bandwidth with different settings. All results were tabulated and plotted.
Judging by all results given, we have chosen DDR3-2133 CL9-12-10-28 (1.65V) mode as best memory settings for this kit. Non-considering Memory Write results, the average memory performance has increased up to 16%.
We could not get the memory work fine with equal CL, RCD и RP timings, except for DDR3-1866 CL10-10-10-28 mode. But we have cleared that for better stability it is recommended to set RCD timing as CL+2 (or more). Voltage adjustment to 1.65V brings more chances for successful overclock.
Below you can find screenshots of AIDA64 buil-in benchmark results:
Having unlocked CPU multiplier gave us an opportunity to increase it to x46 thus to overclock the CPU to 4.6 GHz. Combined with tested memory kit at its best setup we managed to decrease memory latency and enhance its performance 8-10% more:
The success of G.SKILL Ripjaws X lies within Hynix chips — memory voltage tweaking gives more room for clock boost while following timings recommendations adds more stability at higher speed.