Latest technology news
Gordon Mah Ung looks at the new Pascal-based Titan X, which is nothing like the old Titan X
A new Titan has entered the next-generation graphics war. But while Nvidia’s new Titan X Pascal is no doubt the most potent gaming hardware ever released – the Titan X page on GeForce.com is plastered with lofty gaming performance claims – you won’t find too many reviews of the graphics card hitting the streets today. Instead, Nvidia’s focusing more on the graphics card’s use for professional deep-learning AI applications.
The Titan series has always been designed to bridge the gap between the consumer-centric GeForce line-up and pricier Quadro professional cards. The new Titan X Pascal, which gets its name from Nvidia’s cutting-edge 16nm Pascal graphics architecture, delivers 11 teraflops for single-precision floatingpoint performance. But Nvidia’s decision to surprise launch this card during an AI meetup in San Francisco, combined with its touted 44 TOPS INT8 performance – a new deep learning inferencing instruction – shows that the company expects the new Titan X to be used to bolster neural networks and machine learning.
Make no mistake: Nvidia is pushing the Titan as a compute card first and foremost. But most people reading PC Advisor aren’t data scientists. Most people want to know how much ass the Pascal kicks in games.
Titan X specs and gaming performance
The new Titan X Pascal packs in a lot more hardware than the older Maxwell GPU-based processor, as you can see in the comparison chart opposite. It has 3,584 CUDA cores with a 1,417MHz base and 1,531 boost clock. That’s a face-melting half a gigahertz faster
than its predecessor, roughly speaking, and more than 1,000 CUDA cores greater than the ferocious new GeForce GTX 1080, which also uses Nvidia’s Pascal architecture.
The new Titan X is no slouch in the memory department either, with 12GB of next-gen GDDR5X memory clocked at 10Gb/s. It’s connected to the GPU over a 384-bit bus, delivering a blistering 480GB/s memory bandwidth aided by the new delta colour compression system created for Pascal. You read that correctly; it doesn’t have the super-fast high-bandwidth memory first featured in AMD’s Fury cards. Some people expected the new Titan X to come equipped with it after HBM2 made its debut in Nvidia’s powerful P100 supercomputing card. But first-gen HBM is limited to 4GB, and second-gen HBM with higher capacities isn’t expected to be ready until sometime around the end of the year. Fear not, though: 12GB of newer, faster GDDR5X memory definitely won’t be a gaming bottleneck any time soon.
Nvidia’s not providing any additional architectural details about the new ‘GP102’ processor beating at the heart of the Titan X. ROP and texture-unit counts aren’t being disclosed.
So what does that mean in terms of pure performance? Nvidia says the Titan X Pascal will offer 60 percent greater performance than its predecessor. Considering that the GTX 1080 offered 25- to 30 percent greater performance than the last-gen Titan X, the new Titan X will likely offer about 30 percent higher frame rates than the GTX 1080. While we won’t know for sure until we get our grubby paws on the card, there’s a pretty decent chance that the Titan X Pascal will be able to play many of today’s top games at damned near 60fps with high settings at 4K resolution.
If that winds up being true, it’ll be a major milestone. While the original Titan X, GTX 980 Ti and Radeon Fury X can all play games at 4K, doing so often requires graphical compromise and a FreeSync or G-Sync monitor to smooth out a sub-60fps frame rate. Now, however, you can run two Titan X Pascal cards in a single system with Nvidia’s new SLI HB bridge – the hard bridge pictured below.
Despite pushing far more pixels, the Titan X Pascal still draws the same 250W of power as the original Titan X, sipping the juice through 8- and 6-pin power connectors. The two-slot card measures 10.5x4.376in wide, and has the same connections as the GTX 1080: DVI-D, an HDMI 2.0b port, and three DisplayPort 1.4 connections. The new Titan comes with the same vapour chamber cooling with a blower-style fan as the GTX 1080, along with a similarly angular metal
shroud, though the Titan X Pascal’s is all black, rather than silver.
The latest Titan X also packs all the various new features rolled out with Pascal, including simultaneous multi-projection, asynchronous compute enhancements, the Ansel super-screenshot tool, Fast Sync, GPU Boost 3.0, and more.
Price and availability
Though it’s pretty safe to say that the Titan X Pascal will be a beast of a gaming card, we’d recommend – as always – holding off on buying one until you’re able to read indepth reviews of its performance.
If you do decide to pick one up, Nvidia will be the only manufacturer, so you will have to buy the Titan X Pascal from GeForce. com, or get it as part of a custom system via a handful of boutique system builders. In countries where you can’t buy from Nvidia directly, the card will be offered by its partners, but they’ll still be made by Nvidia.
There are two points worth noting. First, the price: the Titan X Pascal is £1,099, or £200 more than the original Titan X’s MSRP. That continues the GTX 10-series trend of being priced higher than their Maxwell GPU-based predecessors. You have to wonder how much the inevitable GTX 1080 Ti, which will likely be more gamer-focused, will cost at launch. Secondly, availability is a question. The high-end GTX 1080 hasn’t stayed in stock reliably since its launch, which in turn has resulted in inflated street prices. (The GTX 1070 and GTX 1060 have had more stock, but are still in high demand.)
Don’t expect Titan X Pascal prices to go sky-high since they’re being sold only by Nvidia, but stocks may very well be limited in the near term. The demand for a £1,099 card will no doubt be lower than for even a £600-plus card such as the GTX 1080.
Regardless, this thing looks like an utter monster. There’s a new Titan in town, and it’ll be the proverbial 800-pound gorilla for a while – AMD’s enthusiast-class ‘Vega’ cards with HBM2 aren’t expected until at least the end of the year, or possibly 2017.
Nvidia’s new SLI HB bridge