SK Telecom deploys Xilinx FPGAs for AI
5X performance and 16X performance per watt over GPU
SK Telecom is definitely a forward looking companies and will be one of the first companies to trial and start 5G. It is not wasting time when it comes to speech recognition and AI as SK Telecom has announced the first commercial adoption of FPGA accelerators from Xilinx in the AI domain for large scale data centers in South Korea.
These two companies made some big claims over the efficiency of the FPGA based solution. SK Telecom has chosen the Xilinx Kintex UltraScale FPGAs as its artificial intelligence (AI) accelerators in its data center.
5 X the performance 16X performance per watt over GPU
Xilinx FPGA will take on the responsible task of running SKT’s automatic speech-recognition (ASR) application to accelerate Nugu, SKT’s voice-activated assistant. Initial testing shows that SKT achieved up to five times higher performance in automatic speech recognition (ASR) applications compared to GPU based solutions.
More importantly they achieved 16 times better performance per watt, and of course this means that this results in a much lower energy bill at the end of the month. Nugu is being made avaiable to SK customers as an Alexa/Google Home based smart AI voice assistant speaker and it is getting a lot of traction in the native Korea.
Explosion of AI Data acceleration is happening now as many companies want to creat a wealth from its data lakes that usually are sitting on some dark corners of a data center / cloud. Adaptable acceleration data center market can grow to $26B until 2022 and SK telecom realised this and had an early FPGA infused AI start.
Explosion of AI Data acceleration need is happening now, and will grow to $26B in less then four years.
FPGA AI makes the Total Cost of Ownership lower
Anyone who ever took an interesting in servers and data centers knows how important the TCO parameters are. The total cost of ownership is lower using FPGAs compared to CPUs and GPU bases solutions. FPGA based Xilinx Kintex UltraScale card simply fits the empty slots and it is an easly and simple solution for SKT to accelerate multiple voice service channels. One FPGA card provides more than five time the performance of a single server and results in substantial cost savings.
Kang-Won Lee, senior vice president, software research and development center at SKT said: “Over many years we have seen the shape of the industry evolve, and we are proud to be at the forefront of developing AI accelerators. By designing our solution based on the Xilinx KCU1500 board and our own bitstream image, we have developed a cost-effective, high-performance application.”
SK Telecom joins a fast growing list of high-profile commercial data centers that have deployed FPGAs for compute acceleration. Xilinx has a unique position with very little competition in FPGA AI applications and it seems that a lot of companies want to go after FPGA approach rather than CPU and GPU accelerated AI.
Manish Muthal, vice president, marketing for data center at Xilinx, said: “We are delighted to have the opportunity to deploy Xilinx FPGAs to SKT’s AI data center, a first in South Korea. “Xilinx Kintex UltraScale KCU1500 FPGAs show that Xilinx has a major competitive edge in applications. Xilinx will continue to focus its technological capabilities and innovations on data center acceleration.”
SK Telecom plans to demonstrate its FPGA AI based solution at the Xilinx developer forum in Silicon Valley on October 1st and 2nd and Beijing on October 16.
