In 3.0 stage, the rapid upgrade and expansion of business volume put forward higher requirements for underlying computing power. The massive user level, coupled with different scenarios, is an urgent need for new technologies and new platforms to promote rapid innovation and development to meet Kakao's multi-angle, all-round business layout and daily maintenance, which all put forward higher requirements for the computing performance, interconnection and scalability of the AI servers supporting the Kakao i platform.
As the second largest Internet company, Kakao Corp. has become a global company with vertical and horizontal business layout since its establishment. At present, Kakao has multiple business platforms covering travel, mobile payment, AI/blockchain, games, music, video and other businesses, including Kakao Talk, Kakao Bank, Kakao Game, Kakao Taxi, Kakao Map, etc. Among them, Kakao Talk is widely downloaded in national level with over 43 million monthly active users. After observing the development of Kakao Group, its growth can be roughly divided into three stages: from Kakao 1.0 to Kakao 3.0. In 3.0 stage, artificial intelligence is one of the most important developing directions.The AI laboratory was formally established in December 2019, and the two main businesses currently developed are Kakao i (AI platform) and Kakao Mini C (smart speakers); Kakao i is an integrated AI platform that combines AI core technology including voice engine, visual engine, dialogue engine, recommendation engine, translation engine, etc.; Kakao Mini C is a smart speaker that combines Kakao AI platform Kakao i, providing various information, including news, exchange rates, stock prices, sports , Movies/TV, real-time popular searches, language dictionaries, etc.
At present, AI has become a hot spot, and core competitive factors such as interaction capabilities, scenario learning, and ecological construction are indispensable, which together form the basis for an AI product to continue to attract users and maintain a good user experience. In the early days of the establishment of the Kakao AI Lab, Inspur provided Kakao with a product portfolio solution with the Inspur high-density inference server NF5468M5 as the core, supplemented by NF5180M5 and NF5280M5 general models, and passed the first test verification, which promoted Kakao’s AI Business layout.
1) Launch product portfolio to adapt to various scenarios and build a complete ecosystem.
In order to satisfy the requirements of clients, Inspur has played a "combined punch" to meet customers' data layering and data storage demands in 3.0 stage by combining different products. Inspur chosed general model NF5180M5 and matched different types of hard disks, which greatly met the requirements for data layering, thereby further realizing the calculation, operation and management functions of Hadoop cluster data; at the same time, NF5280M5, the star model, also has advantages in flexible configuration and high-density deployment. In addition to support multiple types of hard drives for data storage, it also provides audio services for Kakao by matching 4 TA GPU cards; when it comes to more core AI services, in order to meet clients' requirements for power redundancy, Inspur choses NF5468M5-S l—— 40 units adopting NVLink technology, the single machine can support up to 4 2200W power supplies with the stronger power support performance, better heat dissipation design and scalability greatly improving the response level. At the same time, the product has 4 PCIe-switches, which can guarantee the realization of load balancing. In this way, through the group of NF5180M5, NF5280M5 and NF5468M5-S models, customer requirements for adapting to multiple scenarios such as Hadoop, audio services, and AI computing have been achieved.
2) Complete channel network & self-developed components guarantee agile delivery
Facing the explosive growth of customer business, Inspur has built a more complete channel network and used Inspur's self-developed components to greatly shorten the procurement and stocking cycle to achieve extremely agile delivery. On the one hand, maintain regular communication and training with channel partners in the early testing phase to enhance their understanding of the different features of the GPU models that are being promoted this time, and enable distributors to have a better understanding of customer requirements during the stocking and distribution process; on the other hand, by using SAS cards, PHY cards and other components independently developed by Inspur, it not only shortens the supply cycle, but also reduces the overall cost. By controlling the entire process from production to launch, it took only 25 days to realize the entire process from the customer's order to the final launch.
As the technology continues to iterate and update, Kakao accelerates the incubation of more AI products and injects more AI genes into the two major business platforms of the Kakao Group, Kakao i and Kakao Mini C. By using the NF5468M5 of NVLink technology to build an AI training platform, the cluster performance and linear scalability of the Kakao i artificial intelligence platform are increased by more than 30% compared with ordinary servers. For the smart speaker Kakao Mini C, its human-computer interaction experience and the depiction of user portraits have also been greatly improved and perfected.
Smart speakers can respond faster after collecting voices and performing intelligent analysis. At the same time, they can provide corresponding answers and operations corresponding to different contents, making the interaction with people more real and natural. -Kakao Mini C's deep learning capabilities have been further optimized. Through frequent interactions implemented, the AI platform can automatically record user habits and upload data to the background in real time.
With the support of the powerful computing power provided by Inspur's product combinations, the original data can be further analyzed to build user behavior model, depict user portraits, and push content more suitable for users. With the help of Inspur, Kakao's AI products have made outstanding improvements in Hadoop's data operations, audio analysis, and AI calculations, bringing a wider range of AI application scenarios to Kakao.