AI chip developed utilizing ancient Samsung tech is declared to be as quickly as Nvidia A100 GPU- model is smaller sized and far more power effective however is it simply too excellent to be real?

March 28, 2024 by admin

(Image credit: Getty Images)

Researchers from the Korea Advanced Institute of Science and Technology (KAIST) have actually revealed an AI chip that they declare can match the speed of Nvidia’s A100 GPU however with a smaller sized size and considerably lower power usage. The chip was established utilizing Samsung’s 28-nanometer production procedure, an innovation thought about fairly old in the fast-moving world of semiconductors.

The group, led by Professor Yoo Hoi-jun at KAIST’s processing-in-memory proving ground, has actually established what it states is the world’s very first ‘Complementary-Transformer’ (C-Transformer) AI chip. This neuromorphic computing system imitates the structure and operations of the human brain, utilizing a deep knowing design frequently used in visual information processing.

“Neuromorphic computing is an innovation that even business like IBM and Intel have actually not had the ability to execute, and we are happy to be the very first worldwide to run the LLM with a low-power neuromorphic accelerator,” Yoo stated.

Concerns stay

This innovation discovers context and significance by tracking relationships within information, such as words in a sentence, which is an essential innovation for generative AI services like ChatGPT.

Throughout a presentation at the ICT ministry’s head office, employee Kim Sang-yeob showcased the chip’s abilities. On a laptop computer geared up with the chip he carried out jobs such as Q&A sessions, sentence summation, and translations utilizing OpenAI’s LLM, GPT-2. The jobs were finished a minimum of 3 times much faster and, sometimes, approximately 9 times faster than when running GPT-2 on an internet-connected laptop computer.

Executing LLMs in generative AI jobs usually needs various GPUs and 250 watts of power, however the group declares their semiconductor utilizes just 1/625 of the power of Nvidia’s GPU for the exact same jobs. In addition, since it is likewise 1/41 of the size, determining simply 4.5 mm by 4.5 mm, it might eventually be utilized in gadgets like smart phones.

Whether the chip can provide on its pledges in real-world applications stays to be seen. As Tom’s Hardware reports, “Though we are informed that the KAIST C-Transformer chip can do the exact same LLM processing jobs as one of Nvidia’s sturdy A100 GPUs, none of journalism nor conference products we have actually supplied any direct relative efficiency metrics. That’s a substantial figure, noticeable by its lack, and the negative would most likely speculate that an efficiency contrast does not do the C-Transformer any favors.”

Register to the TechRadar Pro newsletter to get all the leading news, viewpoint, functions and assistance your company requires to be successful!