SK Telecom (SKT) unveiled its internally developed artificial intelligence (AI) processor today and shed some light on its AI ambitions.
The company's Sapeon X220 is designed to process large amounts of data in parallel, making it ideal for AI workloads. The company claims the chip can achieve 6,700 frames per second in deep learning workloads, roughly 150% faster than conventional GPUs used for inference.
The operator also claims the chip can do so for half the cost and while consuming 20% less energy than competing GPUs. The Sapeon X220 has a peak power draw of 60 watts. Despite these claims, SKT did not draw any direct comparisons to which GPUs it was comparing against, nor did it provide any concrete benchmarks.
SKT Aims to Expand AI PresenceSKT plans to expand the use of the Sapeon X220 across its product line up. This includes using the chip within its NUGU AI service — think Google Assistant, Siri, or Alexa — to improve voice recognition capabilities. Additionally, the chip will be made available to affiliate companies.
One such company is home security vendor ADT Caps, which plans to use the chip to enhance the performance of its video monitoring service called T View. And in a joint venture with Sinclair Broadcast Group, the Sapeon X220 will be used to power the Cast.era streaming platform.
AI EverywhereOn the communications front, SKT sees advantages to placing AI compute at the edge where, in conjunction with 5G connectivity, users will be able to enjoy lower latency AI-enhanced services.
The operator said it will offer the chip both as a cloud-style service,and as a prepackaged offering that combines the company's hardware and software expertise, algorithms, and training libraries to enable customers to hit the ground running.
Customers will be able to quickly deploy content recommendation, voice and video recognition, and media upscaling workloads using the kit, according to SKT.
"For instance, an over the top service provider wanting to adopt an AI-based content curation service will be able to easily implement the service by simply using SKT's solution that comes with all that is needed," the company claimed in a statement.
Comments