Multinational technology company IBM has developed several technologies as a direct result of its decision to collaborate with other parties.

This not only provides its customers with good-value systems and products, it is also proving to be an effective business model for the firm, says IBM Systems senior VP Tom Rosamilia.

IBM formed the OpenPOWER Foundation – an open technical organisation for data centre infrastructure – three years ago with five partner firms, including graphics chip maker Nvidia.

The foundation now has about 250 member companies in 26 countries.

“Collaborative innovation is about forming ecosystems and partnerships, which we . . . have pursued over the past few years. We have been forming partnerships in hardware, services and software innovation, which is also the right way to do development.

“Seventy per cent of C-level executives [as reported in the Institute of Business Value survey commissioned by IBM] aim to expand their partner networks in the next few years. It is all about aligning with others and being able to gain partners in industry and even form partial partnerships with direct competitors,” says Rosamilia.

IBM’s partnership with Nvidia and the US Department of Energy has led to the development of the NVLink interface for data centre server systems using the IBM Power Systems architecture.

NVLink enables users to customise and optimise central processing units (CPUs) and system platforms running in Power data centres according to their business needs. NVLink uses the graphics processing unit (GPU) to accelerate applications by incrementally offloading some of the processing requirements from the CPUs to the GPUs.

Currently, Peripheral Component Interconnect Express interfaces have a maximum speed of 16 Gb/s, while the NVLink interface provides 40 Gb/s data transfer speeds between the CPU and GPU.

The NVLink interface, therefore, provides ways for users to accelerate their applications and also enables users to more effectively deploy machine learning, deep learning and cognitive systems.

IBM will optimise its next-generation Power Systems Linux servers for high-performance computing and high-performance data analytics that use Nvidia NVLink high-speed interface to connect Power8 CPUs to Nvidia Tesla GPU accelerators, Rosamilia notes.

“We expect a whole new set of enterprise data analytics and cognitive applications will take advantage of GPU accelerators because of the CPU-GPU NVLink system architecture.”

The IBM Power8 multiprocessor chip can be built by OpenPOWER Foundation members under licence, which is also the first time IBM has allowed this for its high-end processors.

Meanwhile, IBM Power Systems CTO and IBM Fellow Satya Sharma confirms that the firm’s Power9 chips will have NVLink interfaces to talk to peripherals and Nvidia GPUs.

“The idea behind the Power8 and Power9 chip architectures is to ensure that the cores are fed with data and the architecture is focused on meeting data-centric workload requirements, from a data flow point of view,” he says.

Open source ecosystems provide several advantages by enabling all stakeholders to do their own development and to leverage the development of others in the community. It dramatically reduces lock-in and negates the need for companies to negotiate with dominant players in the market, notes Sharma.

IBM has extensive relations with many open source companies and the fact that it has enabled licensed manufacturing of Power8 architecture chips means that it is able to innovate alongside the members of the OpenPOWER Foundation. Its collaboration has also led to the development of the Coherent Accelerator Processor Interface (Capi) technology that enables accelerated computing, emphasises Sharma.

Rosamilia concurs, noting: “Accelerators and acceleration technologies are important for the industry to overcome the tapering of Moore’s Law [which states that the number of transistors that can be placed on a processor will double every two years, and has become increasingly difficult to achieve given current miniaturisation and materials].

“High performance computing, therefore, will be done in new ways in future, and firms building open compute platforms will be able to leverage the acceleration technologies to provide clients with very fast processing power that can support modern and increasingly technically complex business applications and analytics systems.”

Schalk Burger is a guest of IBM at the IBM Edge 2016 conference taking place in Las Vegas, Nevada, in the US.