Nvidia announces $3,000 Titan Z graphics card

Two Titan Black GPUs in a single, monstrous product.

Nvidia reveals what must surely be the last hurrah for its outgoing Kepler architecture.

Nvidia has revealed its latest ultra high-end graphics card, dubbed the Titan Z. It packs two of the company's top-end GK110 GPUs onto a single board - the same chip that powers Titan Black, the fastest single GPU graphics card on the market today.

We're looking at a mammoth 12GB of GDDR5 memory (6GB per GPU) married up with a total of 5760 CUDA cores - 2880 per chip. It's unclear right now exactly how fast the chips run: Nvidia's last dual-GPU product - the GeForce GTX 690 married up two GTX 680s onto a single board, but saw clock-speeds reduced in order to fit within a reasonable power/heat profile. Nvidia says that the Titan Z is rated at eight teraflops, while a single Titan Black ranks at 5.1TF, strongly suggestive of a downclock. However, while Nvidia isn't talking clock-speeds, it is known that the Titan Z uses the fastest 7GHz GDDR5 the firm has available.

Nvidia talks up Titan Z as "engineered for next-generation 5K and multi-monitor gaming" but the astonishingly high price-point suggests that even the most affluent gamers may be better served by buying two existing GK110 products (GTX 780, 780 Ti, Titan, Titan Black) and running them in SLI. Instead, we suspect that Titan Z better serves the compute market, where cramming as many high-end GPUs into a PC as possible has a real tangible benefit - a sentiment expressed by Nvidia big cheese Jen-Hsun in this Nvidia blog post.

Warrior robot in disguise? No, it's a CG animation based on the new Titan Z graphics card with its all-new cooling arrangement.

Titan Z appears to be the last hurrah for the company's Kepler line and with Maxwell now in production, Nvidia used its GTC keynote to reveal its plans for 2016. A new architecture - dubbed Pascal - is en route, offering two revolutionary technologies that will see GPU power continue to follow the relentless pace laid down by Moore's Law. NvLink seeks to eliminate the bottleneck between CPU and GPU via a new interconnect with 5x to 12x the bandwidth of the current PCI Express standard, also offering a 5x increase in available bandwidth for multi-GPU set-ups. The second innovation is stacked memory chips, which should offer a huge increase in memory throughput (few specifics were revealed today).

Before Pascal arrives, we'll see more from the Maxwell line in the form of more powerful iterations of Nvidia's successful GeForce GTX 750 Ti, with the architecture also forming the basis for the follow-up to the recently revealed Tegra K1 mobile chip.

Comments (120)

Comments for this article are now closed, but please feel free to continue chatting on the forum!