Developing AI-based translators requires significant time and expense, so adopting high-performance processors to accelerate AI model training is proving effective. The introduction of cutting-edge IPUs is expected to emerge as an effective solution for shortening development times and reducing total cost of ownership.

Artificial intelligence semiconductor company Graphcore announced on the 13th that it has introduced IPU to Twigfarm, a domestic deep learning-based natural language processing (NLP) startup, to improve service innovation and efficiency.

Twigfarm said that by introducing IPU, it has achieved performance that is nearly 10 times better than existing GPUs, and has succeeded in increasing research speed and accelerating service development while significantly reducing total cost of ownership (TCO).

Twigfarm is an AI startup that provides AI-based customized translators for businesses, data de-identification processors, and data verification tools. It is particularly focused on developing customized translators. To reflect customer needs and respond nimbly to market demands, we are conducting research that continuously repeats a series of processes, including data collection, data preprocessing, and modeling improvement.

By introducing Graphcore's IPU, Twigfarm achieved a nearly tenfold increase in training speed compared to existing GPUs, while also significantly reducing costs. The biggest challenge facing many startups is finding ways to effectively utilize relatively limited resources, improving service quality while reducing costs.

He emphasized that Graphcore IPU enables faster learning results at a much lower cost than existing systems, and that this plays a crucial role in accelerating research and shortening the time to market for services.

Twigfarm also stated that with the support of Graphcore's team of experts, they were able to quickly and seamlessly migrate from their existing GPU system to IPU. They added that Graphcore's training, as well as technical support such as code conversion and optimization, minimized the learning curve associated with migrating the framework.

Kang Min-woo, CEO of Graphcore Korea, said, “Graphcore plans to actively seek ways to work closely with domestic AI startups with growth potential going forward,” adding, “Through this, we will contribute to the further growth and development of Korea’s AI ecosystem.”

Baek Seon-ho, CEO of Twigfarm, said, “Graphcore’s IPU system is a groundbreaking solution that solved the cost and time issues that we have been struggling with for a long time,” and added, “We plan to continue our collaboration with Graphcore and expand the application of IPU technology to develop and innovate services.”