Bright lights, lunches and brighter speeches: the spectacle that accompanies a GPU release is as visual and logical as an argument between AMD and Nvidia fanbois conducted whilst quaffing Coke and Mentos.
What matters instead to the informed enthusiast (which arguably includes us given that we weren’t invited to this gorram release) is how the specifications pan out and what performance they give. It’s here the Green team have managed to both fail to surprise and somehow still succeed at delighting. Titan X’s final specs have been confirmed as the 8 billion transistors running at 1002MHz Base clock, 96 ROPs, 3072 cores and 12GB of GDDR5 memory over a 384-bit interface that we expected. This makes for what is a GTX 980 with another half added, but it adds up to a glorious piece of hardware. I mean…. just look at it.
The silicon and metal fully unveiled today translates into performance that, depending on your point of view, either needs no pageantry or managed to turn the entire 2015 Game Developers Conference into a giant launch event. More than one GDC demo was confirmed to have been running on the Titan X, including virtual reality simulations and the Unreal Engine 4 ‘Kite’ below. We’ll eventually be along with our own analysis of how fast this puppy is once the crate (hopefully with accompanying plasma torch) unloads from the dock. But for now know we are looking at a surprisingly quiet 250w card which that gives a roughly ~30% increase over the ‘980 at 1080p and 40% to 60% at 4K resolutions whilst carrying an AU $1500-$2000 price tag.
Perhaps the most serious caveat for anyone willing to take the Titan X related plunge is AMDs hotly rumoured R9 390X. Development of the Red Team’s new supercard is running a few months behind Nvidia’s but the leaks and rumblings so far suggest that it will edge out Nvidia in the performance takes. Unlike its competitor the 390X also has a genuine streak of innovation to it, via integrating of Micron’s ‘High Bandwidth Memory’ memory. HBM will completely replace GDDR5 by physically stacking four layers of memory, on top of their memory controller: the attendant jump in bandwidth from this 3D integration will be very, very significant. Given memory is one area where progress has lagged in recent years (hands up who doesn’t have 5.5-7GHz of GDDR5 in their GPU) this also hints at the possibility of genuinely revolutionary 4K performance.
Not that progress will slow or end here of course. One day in the second half of 2015 we’ll move past these supercards to entirely new designs built on 20 nanometer (nm) silicon. Nvidia has already confirmed that HBM will feature on its next-gen Pascal GPUs and today went into a little more detail with a technology called ‘NVLink’. This is designed to, in some configurations, replace PCI express whilst allowing for a higher bandwidth connection between GPUs. The Green team also alluded that NVLink will also allow 8x SLI, something that in reality will have more to do with the upcoming DirectX 12 removing the current four physical GPU limit. With AMD due to release their 20nm ‘Caribbean Islands’ series the second half of 2015 will prove to be flurry of new ideas and GPU technologies.
For now though, there is a new high water mark in the GPU world and it is undoubtedly the Titan X.