Samsung Unveils Large-Scale Computing System With Nearly 100 AMD Instinct MI100 GPUs and PIM Technology
PIM chips combine a processor and RAM on a single chip. This technological advancement is expected to aid in the performance of AI workloads on a large scale. In October, Samsung revealed that it had demonstrated how the performance of a PIM chip could be combined with an AMD graphics processing unit. When the company used this next-generation breakthrough to train the language model algorithm T5, the performance increased more than twice as much while consuming less power in a system without PIMs.
The world’s first large-scale computing system now houses 96 AMD Instinct MI100 GPUs with PIM ‘Processing-In-Memory’ technology, allowing for a boost in AI algorithms.
We have built a large computing system by combining 96 GPUs (MI100) from AMD of the United States, each loaded with a PIM chip. It is the only one of its kind in the world.
— Choi Chang-kyu, Vice President and Head of the AI Research Center, Samsung Electronics Advanced Institute of Technology
Artificial intelligence for science and technology has gained acceptance in today’s world, as well as widespread acceptance in semiconductor manufacturing processes. According to Choi, the company is also investigating methods to integrate the company’s engineers’ knowledge with inspection data and find a use for it in production.
As the head of the AI research center, I want to make Samsung a semiconductor company that uses AI better than any other company.
The current method of checking yields is limited, as testing can only be done every three to six months when wafers are put into and taken out of the fab. We need to move on to a stage where yields can be predicted using AI sensors and inspection data.
The AI Semiconductor Future Technology Conference brought organisations such as the AI Semiconductor Forum, the Next Generation Intelligence Semiconductor Foundation, the PIM Artificial Intelligence Semiconductor Project Group, and industry demonstrations such as DEEPX, Eyenix, ETRI, Furiosa AI, Nextchip, Nepes, Openedges Technology, SAPEON Korea, and Telechips to the Korea conference site.
News Source: Business Korea