Overclocking

NVIDIA's Greenlight program has not been.. very conducive to GPU overclocking in recent generations, with an ever-increasing hands-off approach provided thanks to high factory GPU frequencies and limited tools to take advantage of the room left for manual overclocking. In particular, 1st-party tools were sorely lacking from the NVIDIA camp, and board partners instead incorporated a variety of software solutions and quick profiles to make things easier.


Overclocking is not new to TechPowerUp readers no doubt, but everyone will also agree that it can get tedious working in a cycle of frequency increase to stability testing to frequency increase/decrease and more stability testing before finally settling on an optimal overclock that provides a good balance of performance, stability, and heat output as decided by each individual user. More often than not, this number is not universally valid across programs and can even vary from game to game. In such working conditions, users tend to play it safe with a low overclock, rely on GPU Boost, or saved profiles tied to individual applications.


NVIDIA decided now would be a good time to tackle this sometimes vicious cycle of overclocking and introduced the NVIDIA Scanner tool. In a demo shown to the press, NVIDIA provided a one-click overclocking solution based on a two-way feedback loop incorporating NV Scanner API, the test algorithm, and an NVIDIA workload. Think of this as an analogue to in-BIOS stability testing using Prime95 or similar, such as what EVGA recently incorporated in their motherboard BIOS. Fittingly thus, the demo used NV Scanner built into EVGA Precision X1 with screenshot images above showing how the auto-OC works with a test result, as well as the resulting GPU frequency-to-voltage curve generated that is specific to your unique GPU. This process goes through the entire curve, as permitted by the manufacturer and/or board partner, so while there remain constraints as before, the overclocking process does seem to be simpler. We will examine this in more detail in our upcoming review to see how well NV Scanner does compared to manual overclocking.


Providing a software tool for overclocking is all fine and dandy, but the hardware meant to be overclocked has to keep up as well. In a big departure from the standard OEM design, NVIDIA's new Founders Edition RTX cards employ a cooler design that will be more conducive for overclocking, while also providing for a potentially quieter and cooler experience at stock clocks. Indeed, the use of a large vapor chamber heatsink with direct contact to a heat-pipe-embedded base plate and two large axial fans for heat dissipation is more akin to board partner solutions for cases that do not starve for airflow or are server racks. To no surprise then, NVIDIA's internal tests show the RTX 2080 is cooler and quieter across the board compared to the NVIDIA Founders Edition GTX 1080 from yesteryear that employed a single, smaller blower fan and a smaller heatsink.


NVIDIA claims even more changes to further facilitate overclocking, with a factory BIOS that has a higher power limit as well as cleaner power delivery on the actual PCB design resulting in smaller fluctuations in power sent from the VRMs. While we will have to wait until our review is out to better examine the power delivery on the Founders Edition card, NVIDIA did show a graph encompassing various example VRM designs ranging from a 1-phase to an 8-phase design with the RTX 2080 power delivery essentially showing itself to be better in terms of handling core current. We must note here that the number of phases in power delivery by itself does not mean much without also knowing the exact components used, so take this snippet for what it is—marketing talk.

Overall, NVIDIA has done well to provide both an improved software and hardware basis for overclocking support from their own coffers. We don't know yet if the moniker for Turing, "Build for Overclocking", is valid, but there is enough at this point to be optimistic.