After extensive Haswell "testing" and research I thought I might share a few discoveries with OCAU.... I agree with the general consensus that once you go above 1.28v you're really pushing the limits of conventional AIO solutions like the H80, and realistically you won't see much of a gain (+200mhz max) unless you plan to delid and use CLU/CLP or go direct die. Therefore Jazper's quick and dirty tips 1. Get an AIO solution, do not bother with custom water unless you already have it or want it for vanity value or to cool multiple components. My recommended units are the Thermaltake pro water 2, the H80i or H90 - h100 and above are not necessary for the below instructions. 2. Really you won't see much gain in terms of top end overclocks with high priced board. (est: 100mhz) 3. Set voltage to 1.28v, uncore to 39x and core to 46x, try booting and test it. If you test it and it's ok, then try 47x, if not, try 45x. do not change your voltage. 4. Once you have found a stable multiplier take your uncore to 45x, then drop it by 1x per attempt until you reach stability. After all this, you'll have your chip running at about 95% of it's potential on air cooled water. Any higher will generally require delidding. Notes on delidding (research only, not practice- yet, I may consider direct die in future): 1. CLP and CLU (and IX) are all corrosive when applied directly to the die, in time your chip will degrade, removal of any of these will leave an aftermath effect on the chip (eg scoring, or other damage) 2. It is a common myth that delidding temp differences are due to the gap between the core and the heatspreader - I've seen reports of people using TIMs other than CLP/CLU and their temp differences are not that far off stock. If the temp difference was due to the spacing, then this would not be the case. My guess is the difference in thermal performance is because the chip coating is eaten away due to the corrosive nature of these TIMs, and by its nature it is a thermal inductor, so minimising it reduces the thermal resistance General notes: 1. Power (wattage) goes up when temperature goes up, regardless of voltage - Hypothesis: High quality OC boards are likely to handle Haswell at higher temperatures with higher overclocks (+100-200mhz) than lower grade boards 2. Batch number does give a vague idea of overclocking potential. What I've read suggests the following (at the aforementioned settings): L30xBxx chips usually hit 4.2-4.3 ghz L311Bxx Chips usually hit 4.3-4.5ghz L312Bxx Chips usually hit 4.4-4.6ghz L313Bxx Chips usually hit 4.4-4.7ghz While this is what my research and personal testing shows, as always YMMV.