Chip manufacturer Ampere Computing has announced a partnership with Qualcomm to develop a new generation AI server, using Ampere's CPUs and Qualcomm's AI Cloud AI 100 Ultra inference chip, promising high performance and energy savings. .
In the development roadmap for next year, Ampere Computing - a company specializing in manufacturing ARM chips for data center servers - surprised everyone by announcing a partnership with Qualcomm, a name famous for its mobile chips.
The combination aims to take advantage of the strengths of both sides to create a new line of AI inference servers, using Ampere's CPU and Qualcomm's Cloud AI 100 Ultra chip. This solution is designed to run large-scale AI models, especially large language models (LLMs) and generative AI applications, in an efficient and energy-efficient manner.
According to Jeff Wittich, Chief Technology Officer at Arm, “We worked with Qualcomm to create a server-grade solution that effectively combines Ampere CPUs with their high-performance cards. This is not just a common development, but a revolution in the way we approach AI.”
Test results show that Meta's Llama 3 language model running on Ampere CPU at Oracle Cloud achieves performance equivalent to Nvidia A100 GPU but consumes less power. Ampere and Qualcomm believe the new server will be an attractive choice for businesses looking for powerful and energy-efficient AI solutions.
In addition, Ampere also introduced the new AmpereOne chip with 256 cores, manufactured on an advanced 3nm process, supporting 12-channel DDR5 RAM, allowing data center customers to flexibly adjust memory access according to needs. .
Not stopping there, Ampere also cooperated with NETINT to build a solution that combines its CPU with NETINT's video processing chip. The new server is capable of encoding 360 channels of live video in parallel and uses OpenAI's Whisper text-to-speech model to create captions for 40 video streams.
These collaborations demonstrate Ampere's efforts to expand the market beyond the traditional data center sector, towards AI applications in media, entertainment, retail and manufacturing.
