It has been a very long wait to see NVIDIA’s next generation Pascal GPU finally hit the streets for consumers to get their hands on. They aren’t quite available to ship right now, though the various retail channels are getting ready to open the gates and let loose the GPUs of war. The reviews, however, are out and are mostly positive about the performance increases that are being seen.
Pascal is a winner, and even hits the performance targets set
In NVIDIA’s announcement they said that the GTX 1080 would be 20% faster than the Titan X. That figure turned out to mostly be true, and in some cases exceeding that by a fair margin. We have to understand that 20% isn’t necessarily a lot, however, and only constitutes a few frames-per-second in most cases. That is still a significant generational jump on its own. The internal IPC improvements represented in the new architecture, which has been years in the making according to Jen-Hsun Huang, are substantial enough that it’s faster while also lowering energy consumption.
Inside isn’t necessarily that large of a difference from Maxwell, but the inclusion of better compression methods, faster memory and a better way that it preempts what you’re doing with both rendering and compute, makes for a much faster experience. It also doesn’t hurt that the 16nm FinFET process allows for a more dense configuration that doesn’t leak nearly as much power. Even though there are less CUDA cores, they’re used far more efficiently here with Pascal. And can clock higher than before with appropriate cooling.
One issue seems to have cropped up, however, in that SLI is now limited to only two cards unless a special tool and help from NVIDIA themselves is used. That’s an interesting move and one that doesn’t necessarily make sense. Though yes, more than two GPUs in SLI is not the most common setup, why the special access in order to use it? The new connector is only capable of supporting two cards, though it should simply be noted that performance scaling won’t be quite as high, nor as sweet, if three or four cards are put together. But locking it down is a strange decision. Our cards haven’t quite arrived yet, but when they do you’ll see a great review that goes beyond the normal selection of games and benchmarks to show you how it performs in a variety of environments.
Thus far, it’s definitely the fastest single GPU out there. With a 20-30% improvement in most games over the GTX 980 Ti. DirectX 12 performance is strange, though, with Hitman seeing a decrease and Ashes of the Singularity seeing an increase. We’ll be sure to investigate that fully as well. In the meantime, here’s some slides.