CeBIT & World Premiere - OCZ FlexXLC PC2-10000! - Testing setup, explanation
Author: Grga Curkovic
Date: 23 Mar 2007

You can?t do much with the modules but overclock them and find out how fast they run at different latencies, so that was what we did. To find out how much headroom there is, we assembled a testbed of the following:

- EVGA nForce 680i SLI
- Intel Core 2 Duo E67000
- EVGA GeForce 8800GTX
- Western Digital RaptorX
- OCZ GameXstream 850W

We ran a batch of tests on the default settings, and of course had no problems. We decided to increase the latencies to 5-5-5-15 2T and try to find the highest stable frequency. After several hours of trying and fine tuning the settings we managed to get the memory to work stable at 1283MHz. We were able to post way over 1300MHz, but the system wasn?t completely stable. With 1320 we were able to boot to windows and the system seemed to work stable, but memtest reported a small number of errors after 20 minutes. We had to go down all the way to 1283MHz to get memtest running without errors. All this was done on 2.4V. We could probably get more out of these modules if we further increased the voltage, but we feel that running the modules higher than 2.4V 24/7 would significantly shorten their lifespan. Since most users won?t sacrifice lifespan for some extra performance, we felt it wouldn?t be fair to test at higher voltage. 1283MHz is a very nice number, way more than any available FSB can push through.

The next step of our benchmarking was finding the highest frequency for 4-4-4-12 2T settings. The voltage stayed unchanged at 2.4V and we started the quest. Again, it took hours, but the result was satisfying. The top memtest stable frequency was 1125MHz. The system wouldn?t even post at 1140, but was rock stable at 1125MHz. This indicates that the modules could be capable of speeds up to 1400MHz running 5-5-5-15, but since the chipset only officially supports speeds up to 1200MHz we believe we were limited by the chipset, not the modules. The FSB:MEM ratio in both cases was set to 1:1 (actually 1:2) and we the CPU was kept somewhere within 2800-2900MHz range so the results measured can be compared. We were able to confirm all of these settings by doing a couple of long memtest runs (over an hour) which was a tight work time-wise because we were doing this all the way through the night?