3DMark, Unigine Heaven and Overclocking
Let's look at a set of tests from more standard benchmarks like Unigine Heaven and the new 3DMark benchmark.
I consider these tests to be somewhat of a "best case" for all the cards in our comparison. We aren't using our frame capture system, we aren't measuring frame latency, nothing like that; I think this should give you an idea of graphics performance if each vendor had the best result for each game.
In 3DMark the GeForce GTX 1080 Ti is a couple of points faster than the Titan X, as expected, which is only made impressive when you consider the launch price points of those two products ($1200 vs $700).
Under Unigine Heaven, the same result is demonstrated. The GTX 1080 Ti is slightly ahead of the Titan X but is 44% faster than the GTX 1080!
I didn't have much time to spend tweaking the new GTX 1080 Ti but I was able to use the latest EVGA Precision X software to push our sample to its limits. Overclocking and the process of doing it remains unchanged for the GTX 1080 Ti compared to previous products. We adjusted the power target to 120% and the TDP limit to 90C and started ramping up the clock speed offset to see how high we could get while remaining stable.
Click to Enlarge
This screenshot was from early in our stable overclock settings and as the temperature on the GPU rises the clock speed settles in around the 1960 MHz mark. I was able to push our GeForce GTX 1080 Ti to a +150 MHz offset which bumps the base clock from 1481 MHz to 1631 MHz and the "typical" Boost clock to 1732 MHz. In practice, our long term gaming clock speeds were noticeably higher than that as you'll see below.
Clock Speed Consistency
Looping Unigine Heaven for 10 minutes to warm up the GPU for testing, I used GPU-Z to capture the clock speed and temperature of the GPU in real time to see how consistent the clock speeds were at both stock and overclocked settings.
Clock speed stability of Pascal at stock settings continues to be lower than I would expect from a company in its position. NVIDIA is slightly aggressive with target clock speeds and voltages ath the rated power targets, and that results in the variance that you see here.
Once overclocked, things are not only faster, but considerably more consistent. This is likely because we have increased the power limit.
With everything at stock, the average frequency in this test over a long period was 1725 MHz, about 65 MHz faster than the stock settings of our Titan X last year. When overclocked with a +150 MHz offset and increased power target, the average bumps all the way to 1963 MHz! This is short of the 2.0 GHz clock rate that NVIDIA showed at its launch event, but I have a sneaking feeling they were running at 100% fan speeds in those demos. All considered, this is an impressive result for the new flagship GeForce product.