On a platform with near infinite levels of configurability, just how do you demonstrate whether a new GPU really has the power to deliver a quality, native 4K experience at 60 frames per second? In our testing with the new GTX 1080 Ti, we established a very simple test criteria: if the new card's performance at ultra HD matches up to the 1080p prowess of GTX 970 at the same settings, we have a winner. We're perhaps one generational step away from a complete match, but Nvidia's new GPU king gets astonishingly close. In some cases, it's actually even faster - a remarkable turnout bearing in mind the 4x increase in pixel-count.
And for us, that's the comparison that really sets out what this product is capable of. It gives users an idea of 'real world' performance based on an experience many PC gamers can relate to, but there's another important yardstick too - whether Nvidia really has handed in a performance uplift over its last GPU king - Titan X Pascal. Across the years, the tinkering Nvidia has carried out on its 'big chip' Titan products to create consumer-orientated cards has involved halving RAM and cutting CUDA cores, with barely perceptible results. GTX 1080 Ti's cuts are even more of an irrelevance.
In fact, Nvidia has boosted GPU frequency by 50MHz, giving the new card a very slight performance lead over the Titan X Pascal in terms of shader throughput (similar to the Titan, the specced boost clock is way lower than its real-life performance, where it regularly hits over 1850MHz). Memory cutbacks vs Titan this time around are limited to the omission of a single module of Micron G5X RAM, giving the Ti 11GB rather than 12GB. This has the knock-on effect of dropping ROP count from 96 to 88, with a small drop to L2 cache, while the Titan's 384-bit memory bus is cut down to a 352-bit interface instead. Nvidia mitigates this with faster G5X modules with 10 per cent more bandwidth than the Titan's, so even memory throughput sees a very small increase.
0 comments:
Post a Comment