NHN selects Graphcore IPU-PODs

1 min read

South Korean cloud service provider NHN has announced that it is to use Graphcore IPU-PODs in upcoming AI compute projects, citing a number of significant advantages over GPU-based systems.

NHN highlighted the scalable design of the IPU-POD, including the separation of AI accelerators and servers, allowing highly flexible configurations.

It contrasted this capability with GPU-based systems, where the two elements are packaged together in a fixed ratio.

Graphcore’s approach allows customers to choose their own server configuration, and tailor the number of servers to IPUs to suit the particular AI workload, delivering maximum compute performance and efficiency.

In the deal signed by the two companies NHN and Graphcore will collaborate on a range of joint projects, while also working to develop AI infrastructure systems including HPC (High Performance Computing), integrating with NHN Cloud systems, and OpenStack.

Kim Dong-hoon, executive director of NHN Cloud Business Group, said: “I’m happy that NHN and Graphcore – both leading innovators in their fields – have created this opportunity to collaborate. We will work to lead the domestic AI and machine learning development ecosystem through our continued co-operation”.

The NHN partnership is the largest, publicly announced commercial agreement for Graphcore in Asia, to date.

Graphcore is working with both Atos and Megazone to deliver the NHN deal. The partners are providing expert support on this, and a range of other projects across their respective regions.