ZOTAC GTX 660 Ti AMP! Edition
WHAT WE LIKED:Excellent performance, Bold styling, Low temperatures, Factory overclock, Compact PCB design
WHAT WE DISLIKED:Lack of power upgrade for serious overclocking, Pricing higher than competitors
For anyone who has ever tried overclocking a graphics card, you know that it’s a bit hit and miss, trying to find the right combination core and memory clocks, voltages, and fan speeds to achieve stability. You’ll also know that trying to match the card’s power consumption to suit your needs is nearly impossible. Nvidia’s GPU Boost technology completely changes the rules, and it’s such a significant departure from past approaches, that it’s safe to say this is an entirely new way of doing things for graphics cards.
A bit of background information: if you’ve heard of Intel’s Turbo technology found in their Sandy Bridge processors, then you know that it dynamically adjusts the CPU frequency and voltages, depending on the load applied. It not only optimally applies horsepower when needed, but also conserves energy and reduces temperatures when the CPU isn’t being heavily taxed. And lastly, it essentially provides a turbo boost of sorts, clocking the chip higher than its rated speed when necessary. Turbo is sort of built-in, on-demand overclocking. And this is very similar to what GPU Boost does, and it acts very much the same.
GPU Boost is not only new to Kepler but it’s also hardwired right in and cannot be disabled. It is not a driver or software feature, and is a function of the product itself. You can’t change it, you can’t turn it off. What this also means is that the Kepler GPU acts very much like Intel’s Sandy Bridge chips as well: there is a GPU Base Block (think of this as the traditional core clock) and a Boost Clock (think of this as the automatic Turbo clock).
What happens with GPU Boost is that the card’s tempeatures, voltages, loads, and so forth are continually monitored many times each second, and are dynamically adjusted in the background to optimize the overall performance. The speeds and voltages may downclock, or overclock, depending on the load demand, whether it’s taxing gameplay or simply using a spreadsheet. There will be no wasted power, energy, or heat output.
It’s important to understand that the Base Clock is fixed, while the Boost Clock is a “target”; it is not a guarantee. As a result, there will be some minor variances between each card’s abilities to automatically clock higher, similar to Sandy Bridge CPUs. It’s expected that board partners will use custom coolers on their products, which will ultimately allow higher Boost Clocks. So consumers can not only expect “bonus” performance from Kepler cards in the form of quasi-auto overclocking, but manufacturers will have the ability to determine different Boost Clock targets based on their product profiles.
Also realize that GPU Boost doesn’t negate overclocking; it merely supplements it in a sense. Overclocking a Kepler card will simply increase the Base and Boost clocks. Since they’re not static speeds, think of overclocking on Kepler as just shifting the overhead targets on the card. You’re not achieving particular clocks; you’ll be achieving a range of potential clocks. This will certainly complicate things for the extreme and competitive overclocking community, and we suspect it will take some time for the dust to settle in order to see how much will change on that front.
So, with Kepler and GPU Boost, users can simply install the card and let it do its thing, optimizing your card’s speeds, power consumption, performance, and temperatures automatically, and just call it a day. Or you can tell the card (through Afterburner, for example) that you’ll offer a higher power overhead, in which case the card’s Boost Clocks will be allowed to go higher as well, giving you a higher turbo speed (not unlike a Sandy Bridge CPU).
Suffice it to say that Nvidia is moving in an entirely new direction with GPU Boost. You don’t need to adjust the fan speeds, voltages, or core frequencies to improve performance. You don’t need to worry about finding a stable overclock either. It’s all taken care of now, built directly into the hardware itself. The downside, of course, is that you lose the ability to shut it off, though some would argue that’s not necessarily a bad thing.