Nvidia presents new gpu interface and architecture




Nvidia is planning to build, so GPUs and CPUs to share data as it is now. Five times as fast a new, faster interface in its GPUs The company in 2016 will also bring a new series of GPUs on the market, under the name Pascal.

The current PCI Express 3.0 interface, which is used as an interface for stabbing a GPU in a PC, according to Nvidia a major bottleneck. The NVLink interface should solve that, as Nvidia has announced at a press conference. NVLink is five to twelve times as fast as PCI Express 3.0. Caches between the CPU and the GPU remain gesny nized.

The data can slide back and forth more quickly as Nvidia is especially important for supercomputers. The new interface allows the company says supercomputers with 1,000 petaflops possible, some 50 to 100 times faster than today’s supercomputers.

When the new interface will be used is not yet known, but it is certain that this will not happen with Nvidia GPUs using the Maxwell-architecture. The first GPUs with that architecture were last month announced .

In addition, Nvidia’s 2016 GPU’s release with a new architecture, called Pascal. This architecture must be equipped with memory that is installed, allowing a GPU with thousands of bits can work simultaneously true that now more than hundreds of bits. Above the GPU That principle, baptized by Nvidia 3d memory, also ensures smaller high-end GPUs: during the keynote showed the CEO of Nvidia Pascal a card the size of a credit card.

interconnect nvidia


In: Technology & Gadgets Asked By: [15500 Red Star Level]

Answer this Question

You must be Logged In to post an Answer.

Not a member yet? Sign Up Now »