Game on with Kepler graphics cards

David Wright describes what he sees as the advantages of Nvidia's Kepler architecture

Nvidia introduced its Kepler architecture this year. Apparently, the four graphics cards in the series are still in allocation as the company cannot produce them quickly enough. In my view, they promise the best performance in the market, and the best per-watt performance of any GPU available.

The Kepler platform was developed with efficiency in mind, including a redesign of the streaming multiprocessor to improve performance per watt. Nvidia has also added a feature that increases clock speed dynamically. This should improve performance within a card's power budget.

Kepler architecture can equip such cards with hardware-based frame rate metering, a technology that aims to reduce video frame stuttering. With two GPUs on board, frames can be handled nearly simultaneously, which should make for smoother overall graphics.

Moreover, the Kepler architecture defines the GPU building blocks and will be the basis not just for a single desktop chip but for a family of designs that can be deployed in high-performance desktop PCs, supercomputers, mobile devices, workstations and datacentre kit.

As a partner of Nvida board partner Asus as well as Palit, we believe Kepler architecture will rapidly go mainstream in the second half of this year.

Apple has said it is switching back to Nvidia graphics for its 15in MacBook Pro model, which will come with its so-called Retina Display. Such displays require significant graphics processing power, so in my view this bodes well for Nvidia with regard to AMD discrete GPUs and integrated GPUs.

I believe Nvidia is also poised to take a large share of the new Windows RT tablet market. Microsoft's new Surface tablet, using Nvidia cards, will help drive growth within the enterprise market.

Nvidia also has GRID GPUs that could help gaming providers harness cloud computing. Here, the new Kepler GPU architecture can help reduce power consumption by simultaneously encoding up to eight game streams.

Sophisticated games could be rendered in the cloud and encoded on the GPU, rather than the CPU. This allows servers to run more game streams at once. More power can be directed at the high-end games, and less at the low-end games. Server latency may also be reduced, by capturing and encoding a game frame in a single pass.

David Wright is a product manager at VIP Computers