Nvidia launches K40 GPU, strikes partnerships with IBM and others

|By:, SA News Editor

Nvidia (NVDA) has unveiled the Tesla K40, a server/HPC GPU it claims is 40% more powerful than its predecessor (the Tesla K20X) and the first product of its kind optimized for big data/analytics and large-scale scientific use cases.

The K40, which competes against AMD's FirePro and Intel's Xeon Phi lines, comes with 12GB of memory and 2,880 GPU cores, as well as Nvidia's GPU Boost tech.

In tandem with the K40 launch, Nvidia has announced a partnership with IBM to develop "GPU-accelerated versions" of IBM apps running on Big Blue's high-end Power systems (they rely on IBM's Power CPUs). Nvidia's CUDA programming model (allows apps to be optimized for Nvidia GPUs) will be leveraged, as will IBM's Rational software development tools.

IBM has been getting creative in its efforts to grow Power-related sales in the face of plunging demand for traditional UNIX Power servers. Three months ago, IBM announced it would license the Power architecture and related technologies to spur broader adoption, particularly among Web/cloud companies; Nvidia was one of the launch partners. More recently, the company opened up its Power-based Watson A.I. computing platform to 3rd-party developers.

Nvidia has also announced SGI, Cray, and other server vendors will offer the K40, and that the Tesla GPUs will work with Mellanox's (MLNX +1.9%) InfiniBand adapters (often used in HPC applications).

Nvidia's Tesla GPU sales have been growing quickly, helping offset PC and Tegra weakness. The chipmaker has also been targeting the server GPU space with its Grid GPUs (meant for cloud gaming and PC virtualization apps).

Previous: Nvidia hit by Morgan Stanley downgrade