Try 32M please, with process affinity to core 1+2. Im interesting how it scaling compare Pinnacle Ridge in 32M. On HWBOT are few results, but seems not good (like Pinnacle Ridge is better clock to clock in 32M).