NEC Corporation
NEC Corporation
11/27/2024 | Press release | Distributed by Public on 11/28/2024 22:51
NEC strengthens its generative AI, NEC cotomi, for specialized business use, maintaining high speed and achieving world-class accuracy
Tokyo, November 27, 2024 - NEC Corporation (NEC; TSE: 6701) has enhanced its generative AI, "NEC cotomi," achieving world-class accuracy while maintaining high speed in Japanese language benchmarks. At the same time, NEC has also developed technologies that double the computational efficiency of GPUs while maintaining the performance of generative AI.
The use of NEC cotomi, a key technology in the value creation model for digital transformation (DX) "NEC BluStellar" (*1), enables NEC to strongly promote the use of generative AI in actual business. Moreover, NEC technologies can help to alleviate issues such as GPU shortages and power problems associated with the growing demand for AI, while contributing to the creation of an environment in which AI can be used more easily and comfortably.
Due to the rapid development of generative AI, various companies and public institutions are capitalizing on it to promote business innovation. In line with this, there is a demand for improvements in the functions and performance of generative AI itself, such as support for use cases tailored to more specialized tasks and for incorporation into business systems.
Meanwhile, studies have shown that global demand for the generative AI market is expected to grow by about 20 times from 2023 to 2030. In line with this, the demand for GPUs is also rapidly increasing, and the increase in power consumption in data centers is becoming an issue.
In response to these challenges, NEC has developed technology to enhance the performance of NEC cotomi and increase the computational efficiency of GPUs, providing conditions for the use of generative AI that are more environmentally friendly.
Now, NEC cotomi has enhanced its performance and achieved accuracy in the Japanese LLM (Large Language Model) benchmark "Japanese MT-Bench" that is comparable to global top-level LLMs such as Claude, GPT-4, and Qwen (*2). Furthermore, it has achieved a speed that is six times faster than Qwen and twice as fast as GPT-4o, successfully achieving both overwhelming inference speed and high accuracy. With this performance enhancement, NEC is making the automation of advanced specialized tasks using generative AI a reality. NEC has also developed technology to maximize the performance of generative AI and improve the computational efficiency of GPUs. When generative AI is used in business, it tends to run more frequently, which increases the load on the GPU and the total power consumption. Furthermore, there is a concern that these issues will increase in the future as the application of generative AI to business systems and other uses becomes more widespread.
To address this, NEC developed a technology to reduce the power consumption per inference process. This technology improves the efficiency of parallel processing by increasing the utilization efficiency of the arithmetic elements inside the GPU, doubling the arithmetic efficiency of the GPU (compared to NEC's previous technology), and reducing the power consumption per inference process. This makes it possible to perform inference processes with fewer GPUs than before, and to reduce power utilization. NEC offers end-to-end DX services, from strategy and concept consulting to implementation-focused offerings, based on the three pillars of business models, technology, and organization/talent. Additionally, in its shift from a traditional systems integrator to a "Value Driver," NEC restructured its value creation model under the name NEC BluStellar, which leverages NEC's cutting-edge technologies, developed and refined through years of experience and proven cross-industry expertise, aiming to transform business models, address social challenges, resolve management issues faced by customers, and lead them into a brighter future.
The use of NEC cotomi, a key technology in the value creation model for digital transformation (DX) "NEC BluStellar" (*1), enables NEC to strongly promote the use of generative AI in actual business. Moreover, NEC technologies can help to alleviate issues such as GPU shortages and power problems associated with the growing demand for AI, while contributing to the creation of an environment in which AI can be used more easily and comfortably.
Due to the rapid development of generative AI, various companies and public institutions are capitalizing on it to promote business innovation. In line with this, there is a demand for improvements in the functions and performance of generative AI itself, such as support for use cases tailored to more specialized tasks and for incorporation into business systems.
Meanwhile, studies have shown that global demand for the generative AI market is expected to grow by about 20 times from 2023 to 2030. In line with this, the demand for GPUs is also rapidly increasing, and the increase in power consumption in data centers is becoming an issue.
In response to these challenges, NEC has developed technology to enhance the performance of NEC cotomi and increase the computational efficiency of GPUs, providing conditions for the use of generative AI that are more environmentally friendly.
Now, NEC cotomi has enhanced its performance and achieved accuracy in the Japanese LLM (Large Language Model) benchmark "Japanese MT-Bench" that is comparable to global top-level LLMs such as Claude, GPT-4, and Qwen (*2). Furthermore, it has achieved a speed that is six times faster than Qwen and twice as fast as GPT-4o, successfully achieving both overwhelming inference speed and high accuracy. With this performance enhancement, NEC is making the automation of advanced specialized tasks using generative AI a reality. NEC has also developed technology to maximize the performance of generative AI and improve the computational efficiency of GPUs. When generative AI is used in business, it tends to run more frequently, which increases the load on the GPU and the total power consumption. Furthermore, there is a concern that these issues will increase in the future as the application of generative AI to business systems and other uses becomes more widespread.
To address this, NEC developed a technology to reduce the power consumption per inference process. This technology improves the efficiency of parallel processing by increasing the utilization efficiency of the arithmetic elements inside the GPU, doubling the arithmetic efficiency of the GPU (compared to NEC's previous technology), and reducing the power consumption per inference process. This makes it possible to perform inference processes with fewer GPUs than before, and to reduce power utilization. NEC offers end-to-end DX services, from strategy and concept consulting to implementation-focused offerings, based on the three pillars of business models, technology, and organization/talent. Additionally, in its shift from a traditional systems integrator to a "Value Driver," NEC restructured its value creation model under the name NEC BluStellar, which leverages NEC's cutting-edge technologies, developed and refined through years of experience and proven cross-industry expertise, aiming to transform business models, address social challenges, resolve management issues faced by customers, and lead them into a brighter future.