Nvidia has launched the Tesla K40 GPU accelerator, which it clams is the world’s highest performance accelerator. The K40 provides double the memory and up to 40 percent higher performance than its predecessor, the Tesla K20X. Computational performance has been measured at 4.29 Tflops single-precision and 1.43 Tflops double-precision peak floating-point performance.
The Tesla K40 features 12GB of GDDR5 memory and 2,880 Cuda parallel processing cores. It is based on the Kepler compute architecture.
The accelerator has dynamic parallelism which enables GPU threads to dynamically spawn new threads; this enables users to process adaptive and dynamic data structures. The K40 features PCIe Gen -3 interconnect support, which moves data twice as fast as PCIe Gen-2 technology.
The product also features GPU boost technology which converts power headroom into a user-controlled performance boost. The company states that this will enable users to unlock the untapped performance of a broad range of applications.