NVIDIA’S TITANS

Nvidia’s new Titan X is the fastest gaming card the world has ever seen. But it’s also the most expensive. So, what are these Titan boards about, and can the crazy prices possibly make sense?

Nvidia rolled out the first Titan in early 2013. This was the fastest graphics card a lot of money could buy. At $1,000, it was double the price of a GeForce GTX 680. However, the chip at the heart of the first Titan was GK110, a monster measuring 550mm2, and it did more than just graphics. It was packed with features for general-purpose parallel computing, and formed the basis of Nvidia’s high-performance Tesla compute boards. The follow-up Titan Black was just a fully enabled version of GK110, with all the CUDA cores switched on. However, Nvidia’s approach shifted with the original Titan X last year. That used a new GM200 GPU from the Maxwell family, and ditched the fancy FP64 compute capabilities.

The latest Titan X is more of the same, only based on Nvidia’s new 16nm Pascal architecture and a chip codenamed GP102. It’s pure graphics, not an FP64 compute machine, even if Nvidia says it will be good at neural network computing, and won’t be used for Nvidia’s Tesla compute cards, which get their own P100 chip. It’s also $1,200,

which sets a new high. Where the contradictions come in is that the new Titans are more graphics-focussed, so better value for gamers. On the other hand, the original Titan’s exotic compute capabilities went some way to justifying the price, even though you were never going to use them.

With the latest Titan X, you’re getting roughly 30 percent more performance for double the money of the GTX 1080, itself very expensive. That seems like a bum deal. Especially when you consider, based on our experience of the GTX 1080, that it might not quite even be the single-GPU 4K gaming solution the world has been waiting for.