System setup and notes, issues
We had some 'fun' getting these modules to A, work and B, work in a way that allowed us to provide comparable benchmark results. We have now got what we consider to be the best results for comparison, given the circumstances we were presented with. The system setup table will enlighten you some more.
|Motherboard||ASUS M2N32-SLI Deluxe (nForce 590 SLI)|
|Chipset driver||ForceWare 9.34|
|Graphics cards||2 x Asus Geforce 7900GTX 512MB (SLI)|
|Graphics drivers||Forceware 91.31|
|Hard drive||Seagate 160GB SATA 'II' (ST3160812AS)|
|Optical drive||Sony DW-Q30A|
|PSU||FSP Epsilon 600W|
|OS||Windows XP Professional SP2|
|Memory configuration 1||Memory configuration 2||Memory configuration 3|
|Memory||2GiB Corsair TWIN2X2048-8888C4DF||2GiB Corsair TWIN2X2048-8500C5||2GiB Corsair TWIN2X2048-8888C4DF|
|Memory speed||4-4-4-12 @ DDR2-1111.4MHz||5-5-5-15 @ DDR2-1072MHz||4-4-4-12 @ DDR2-1072MHz|
We initially set out to test Corsair's new RAM on our i975X Conroe platform. We started off with an ASUS P5W-DH Deluxe (975X+ICH7R) motherboard, but simply couldn't get a stable system above a memory frequency of 1088MHz. We tried different BIOS versions, to no avail. We then attempted to use Intel's Bad Axe, but it didn't have a multiplier for the memory frequency that would allow us to run the clocks we wanted without massive FSB overclocks.
In the end we had to move onto an ASUS based AM2 platform, one cited by Corsair in their testing, the M2N32-SLI Deluxe. We recently reviewed this board alongside another ASUS model, for those who might be interested in some nForce 590 SLI action.
Once on AM2, we still had a few hoops to jump through. EPP simply didn't work properly for us. We wanted 1111MHz, 4-4-4-12-2T timings at 2.4V, but the CPUOC profiles in the M2N32-SLI's BIOS didn't quite give us that. At CPUOC Max, we could get said frequency and timings, but VDIMM would be 1.8V and the CPU would run at 278MHz x 12. The resultant 536MHz CPU overclock was simply too much for the system, which wouldn't even POST at this.
The simplest way for us to run the PC2-8888 modules at 1111MHz was to manually adjust CPU multiplier and HT reference clock. However, we couldn't run PC2-8500 at the exact same CPU frequency and FSB (nor would we have been able to using EPP, for that matter). In the end, we attempted to get close to specified memory speeds with changes to FSB and CPU frequency that were as small as possible. The above table shows what we chose to use.
In light of these discrepancies, performance differences doesn't rest entirely with the RAM, but we can still paint a performance picture for you.
Note that in our tests, we ran the Dominator modules at 1111MHz and 1066MHz. Right now there's no other module available that's advertised to run at 1111MHz, so we added the 1066MHz results to show how the Dominator modules compare to Corsair's TWIN2X2048-8500C5's with tighter timings.
A quick note on competing modules: Currently there are none, although Geil's and OCZ's websites list similarly specced products. However, we've not seen any available, or been offered samples yet.
We've run performance, overclocking and temperature testing on these modules to cover all the areas in which the Dominators should... dominate. For performance testing, we used the following tests:
- ScienceMark Memory Bandwidth
- ScienceMark Memory Latency
- HEXUS DivX Encoding
- Far Cry v1.33 1024x768 - No sound
- Quake 4 v1.04 1024x768
So, both synthetic and 'real world' benchmarks for you, which should show us just how much benefit the PC2-8888 modules can provide.
For overclocking testing, we followed the method used in our TWIN2X1024-8500 review, where we pushed the RAM as far as we could to get 10 minutes of stable Memtest86+ at three different memory timings.
For temperature testing, we calculated the temperature delta between the modules and ambient, with and without Dominator Airflow for both products. We aimed a CHY 110 infrared thermometer at shiny parts (i.e. stickers) of the modules, giving us the most reliable readings. Corsair's PC2-8500 modules seemed to have an uneven heat distribution; we took readings from the hotter part. The PC2-8888 modules seemed to have a more even temperature across them.