Nvidia surprised the crowds at its annual GPU Technology Conference last night with the announcement of a new top-end graphics card, the dual-GPU GeForce GTX Titan Z.
Featuring a pair of Kepler GK110 chips, the Titan Z offers 5,760 CUDA processing cores both running at full speed. Each has 6GB of GDDR5 video memory, for a combined total of 12GB - a figure more usually associated with the company's Tesla accelerator boards than its GeForce consumer GPUs. Sadly, beyond the usual claims that it's the world's fastest graphics card, Nvidia did not share full specifications at the event beyond promising that both GPUs are clock-linked, meaning there'll be no bottlenecking involved.
Nvidia was also quiet on thermal design profile (TDP) at the unveiling, but with the single-GPU GeForce GTX Titan on which the Titan Z is based drawing 250W it's hard to imagine that the company has found a way to jam two GPUs onto a card without a major increase in power draw. That said, Nvidia has claimed that the card will be 'cool and quiet rather than hot and loud, promising low-profile components for a triple-slot design and ducted baseplate channels to reduce air turbulence and therefore noise levels. Huang also claimed that a triple-Titan Z setup, for those that could afford such a thing, would draw around 2KW in total - suggesting a 500W+ TDP if you allow for other system components.
One thing Nvidia was willing to share, surprisingly, was the price: the card will launch in the US at a wallet-emptying $2,999. Compared with the company's existing GeForce line-up, that's a seriously high price tag - but with eight teraflops of floating-point performance, and Jen-Hsun Huang tellingly describing it as the card for those 'in desperate need of a supercomputer that you need to fit under your desk,' it seems that despite its GeForce moniker the Titan Z is being positioned as an alternative to the Tesla accelerator board family.
Want to comment? Please log in.