Nvidia Corp.’s latest graphics processing units based on “Maxwell” architecture are rather good overclockers, but only to a certain degree. According to Vince “K|NGP|N” Lucido, a professional overclocker, without extreme methods of cooling (such as use of liquid nitrogen, chiller or other), most “Maxwell” GPUs just do not work at frequencies beyond certain limits even with ultra-high voltages, which is why in many cases it is useless to just amplify GPU voltage supply.
In a bid to overclock a microprocessor or a graphics processing unit, one needs to cool it down properly and increase voltage because operation at higher clock-rates requires additional power. Since higher power consumption means higher heat dissipation, requirements of better cooling increase with every additional 0.1V. Some previous-gen Nvidia GPUs could handle extreme temperatures and therefore could be overclocked to record frequencies even with air or water coolers. However, everything is different with “Maxwell”. The new graphics processing units from Nvidia require both additional voltage and exotic cooling methods to be overclocked to extreme frequencies.
What it means for end-users and makers of graphics adapters with fine coolers is that they will be unable to overclock “Maxwell” GPUs beyond certain points no matter what voltage they apply.
“With GeForce GTX 980 [GM204] and later GPUs, including GeForce GTX Titan X [GM200], the scaling on air/water has all but almost gone,” said Mr. Lucido in a forum post. “I would say about 95 per cent of all [Maxwell-based graphics cards] no matter [from] what [graphics card] brand, do not scale with more voltage than 1.25v-1.275v at temps warmer than 25°C or so. There is no magical BIOS that can effectively remove this.”
Very interesting, probably has something to do with density?