Each platform has it's own peculiarities regarding how to fine tune and get the most out of your hardware. With Ivy Bridge the tertiary timings are crucial regarding booting at high clocks and matter even more when starting to bench. The ASUS bios is as usual well dialed in, this to enhance RAM compatibility. The ROG boards however get an extra treatment. The ASUS engineers did most of the hard work for it's end users by giving us preset RAM profiles. Just select the type of profile that suits your RAM kit and your needs, then start to fine tune from that starting point.
Now let's see what the board's bios sets when we load the Corsair XMP profile. The result is shown below in the in the first untweaked Mem TweakIt screenshot.
As we pointed out with a few tests on the previous page, some scores were slower at 2666Mhz then at 2400mhz. This is partly our fault, as we only change the main timings and let the board dial in the secondary and tertiary timings. We opted for this approach as this is probably what most end users will do too.
Very important for maximum performance are the tRRSR and tWWSR values. For daily usage a value of 5 is fine. However benchers will try to set 4. Take notice that this slight change could require far more Vdimm voltage. Secondly not all RAM kits remain stable with the 4 value selected for both the tRRSR and tWWSR timings.
Command Rate 2T also gives a little bit less performance then when utilising 1T Command Rate. Similar and regularly used is a far tighter TRFC setting. The XMP profile sets 214, but the kit is perfectly stable at 174 or even less. How we slightly tweaked the XMP profile settings in the quest for more performance can be seen in the below screenshot.
Most important changes we made are:
- Locking tRSSR and tWWSR at 5
- Command rate at 1T iso 2T
- Tightening TRFC to 147 iso 214
- Some slight secondary timing adjustments
How does it affect the RAM or oveal system performance ? We redid a few simple tests, which you can easily replicate. First up is SuperPI 32M, almost a 2 second gain and believe me there is more to dial in via the tertiary timings:)
The AIDA Bandwith test shows the amazing performance at the COPY test. All achieved by tightening the tWWSR setting from 7 to 5. Write and Read performance achieve a minimal gain.
The CPU Queen test, which is close to zero bandwith dependant, logically hardly improves. But just look at the PhotoworXX outcome.
The multithreaded Y-Cruncher stress test completes almost 3secs faster. The Cinebench Release 11 scores remain almost identical.
Even though the encoding test seemed already sort of maxed out at 2400Mhz. There's still the option to tweak for some extra performance. Maybe a bit time consuming, yet making a nice difference. Futuremark's 3DMark11 PhysX test increases with 200 points. However the overal score gains far less.