Creaform Connect Korea ~02.25
반도체 AI 인더스트리 4.0 SDV 스마트 IoT 컴퓨터 통신 특수 가스 소재 및 장비 e4ds plus

SKT, Doubles GPU of Ultra-Large AI 'Adot' Supercomputer

기사입력2023.02.13 09:15

▲SKT's supercomputer 'Titan'

Adot Brain Role GPU, 1040 Units Built
Performance of 17,100 trillion operations per second

SK Telecom's giant AI model 'A. (Adot)' is getting smarter.

SK Telecom announced on the 12th that it has doubled the GPU of its supercomputer 'Titan', the brain of its ultra-large AI model 'Adot'.

SKT has been building and operating its own supercomputer since 2021. This time, it has increased the number of NVIDIA A100 GPUs to 1,040.

SKT emphasized the importance of supercomputer infrastructure that can quickly and accurately process massive amounts of data dealing with billions of parameters in order to drive ultra-large-scale AI.

Titan also supports a performance of over 17.1 petaflops (1,000 trillion calculations per second), or 17,100 trillion operations per second. One petaflop is the speed of 1,000 trillion floating point (a computer's way of recognizing real numbers) operations per second.

It was ranked 85th in the world's supercomputer ranking 'Top 500 (top500.org)' last year. It is the first among domestic telecommunication companies.

Meanwhile, SKT expects that this GPU expansion will enable its ultra-large AI model 'Adot' to perform more sophisticated learning than before.

To drive the super-large AI 'Adot', a model with over hundreds of billions of parameters is being applied to commercial services. It is explained that the model will be more than twice the size of the current model in proportion to the expansion of computing performance.

SKT is also expanding the universality of its ultra-large AI. Currently, Adot is providing services in conjunction with domestic media platforms such as FLO, TMAP, and Wave.

SKT A. Promotion Team Manager Kim Young-jun said, “We expect that this expansion of the supercomputer will play a major role in improving Adot’s ability to learn more sophisticatedly than before, as well as its ability to improve the flow of conversations with people and the completeness of its answers to be closer to the human level.” He added, “We will continue to strive to lead AI technology leadership through aggressive R&D investment, infrastructure expansion, and talent recruitment.”