SK Telecom Doubles Capacity of ‘Titan’ Supercomputer for ChatGPT-like Language Model

Print Friendly, PDF & Email

SK Telecom of South Korea has boosted the power of its “Titan” AI supercomputer by roughly 100 percent by increasing its capacity to 1,040 NVIDIA A100 GPUs

A story on the Korean news site Pulse stated that SKT announced on Sunday the news on the supercomputer, which powers of the company’s artificial intelligence model, called AIDAT. The reason for the enhancement: to implement and launch a Korean-language chatbot on the model of OpenAI’s ChatGPT.

Titan was installed 2022 and ranked no. 92 last November on the Top500 list of the world’s most powerful supercomputers. It’s an HPE system running on the Apollo 6500 platform, powered by AMD EPYC CPUs and running the NVIDIA-Mellanox HDR Infiniband interconnect. With the added GPUs, Titan delivers more than 17.1 petaFLOPS.

According to Pulse, “SKT commercialized the country’s first commercial version of its self-developed super-large Korean-language model GPT-3 with a beta launch of AIDAT in May of last year. The company plans to strengthen knowledge-based conversation services by applying the ChatGPT model within this year.”

The enhanced power of SKT’s supercomputer will help improve AIDAT’s conversation flow to nearly a human level, said Kim Young-joon, leader of the company’s AIDAT Team, according to Pulse.