Nvidia Titan X Pascal review

[Administrator]

With AMD's focus elsewhere in the graphics market, Nvidia's GeForce GTX 1080 has held an iron grip on the GPU top-end, offering a phenomenal level of performance, in exchange for a hefty price-premium. The new Titan X Pascal ups the ante still further: even more power, and an even higher price of entry. The bottom line is clear - the new Titan offers often staggering levels of performance, but we are looking at a 2x increase in price over GTX 1080 for a maximum boost of around 30-33 per cent in frame-rates.
We can argue about the Titan line's value proposition, but it's a moot point - as a brand, it's Nvidia's equivalent to Intel's Extreme Edition processors and obviously, they wouldn't make them if they couldn't sell them. And of course, the Titan serves as a preview of sorts, a precursor for the more mainstream products to come. There was a time when the Maxwell-powered Titan X was similarly unattainable owing to its $999 ticket price. Then GTX 980 Ti came along, offering nigh-on identical frame-rates with a third of the price lopped off. And now, that same level of performance is available at just $380 in the form of GTX 1070, with a 100W power efficiency saving on top for good measure.
In there here and now, Titan X Pascal sees Nvidia iterate on its previous strategy in terms of the make-up of its big chip. In many ways, the new Titan is essentially an upsized version of its existing GTX 1080, adding 50 per cent to everything in the core processor architecture. For example, memory bandwidth rises from 320GB/s to 480GB/s, with the interface expanding from 256 to 384 bits. Memory allocation itself rises from 8GB of GDDR5X to 12GB. ROPs are increased from 64 to 96. However, the CUDA core and texture unit count doesn't see quite as much of a linear jump - we go from 2560 to 3584 shaders in Titan X and from 160 to 224 texture units. It's now been confirmed that the fully enabled GP102 actually has 3840 shaders on the 12bn transistor die - 256 CUDA cores are disabled in order to get more workable chips from the production line.
Read more…


More...