SKT Doubles the Scale of A.Dot Supercomputer Deployment

Capable of Processing 1.71 Quintillion Operations per Second
Enables More Precise Learning Than Before

[Asia Economy Reporter Lim Hye-seon] SK Telecom announced on the 12th that it has doubled the scale of its supercomputer Titan, which serves as the brain of its massive artificial intelligence (AI) model A.Dot.

To operate massive AI models handling tens of billions of parameters, a supercomputer infrastructure capable of processing vast amounts of data quickly and accurately is essential. SKT has been building and operating its own supercomputer since 2021, and this time expanded it with 1,040 NVIDIA A100 graphics processing units (GPUs), more than doubling its previous scale.

Image of SKT's supercomputer 'Titan'

SKT explained that its supercomputer supports performance exceeding 17.1 petaflops (one quadrillion mathematical operations per second), making it one of the top-performing systems globally. One petaflop is the speed capable of performing one thousand trillion floating-point operations per second, and 17.1 petaflops means it can process 17,100 trillion operations per second.

Last year, SKT’s supercomputer ranked 85th in the global Top 500 supercomputer list, becoming the first domestic telecommunications company to enter the ranking. The Top 500 Foundation conducts biannual performance assessments and publishes the global supercomputer performance rankings based on these results.

SKT expects that the expansion of its supercomputer will enable its massive AI model A.Dot to perform more sophisticated learning than before.

A.Dot has already applied models with hundreds of billions of parameters in commercial services, and with the expanded computing power, it plans to advance to models more than twice the current scale.

Based on years of internalized massive AI research and development (R&D) capabilities and supercomputing advancements, SKT plans to infinitely expand the versatility of massive AI.