Chelsea_SunChelsea_Sun ・ Jul. 5, 2024
Moore Threads Upgrades KUAE Intelligent Computing Center to Support AI Power
With the intense competition in China's AI large model industry, new models are being released almost every month, and companies must upgrade their models within two weeks to stay competitive.

TMTPOST--Moore Threads, a Chinese AI computing solutions company, has showcased its upgraded AI flagship product, the KUAE Intelligent Computing Cluster Solution, expanding from the current thousand GPU cards level to a massive ten thousand GPU cards scale.

The product debuted at the 2024 World Artificial Intelligence Conference (WAIC 2024) on Thursday.

Moore Threads aims to build a leading platform that can support a scale of ten thousand GPU cards and achieve petaFLOPS-level floating-point computing power. This platform is designed specifically for training complex large models with trillions of parameters.

Additionally, Moore Threads announced strategic partnerships with China Mobile Qinghai Company, China Unicom Qinghai Company, Beijing Dedaoxin Technology Group, China Energy Engineering Group, and Guilin Huajue Big Data Technology Company to jointly build three ten-thousand-card cluster projects, focusing on building a domestic GPU cluster.

Zhang Jianzhong, the founder and CEO of Moore Threads, emphasized the importance of scaling in the AI large model industry, saying that the demand for computing power grows as models scale up.

He pointed out that while international companies like OpenAI, Google, and Meta are procuring tens of thousands of NVIDIA GPUs to build data center clusters, domestic companies still face challenges in reaching such scales, with most managing to procure only a few thousand GPU cards. Only a few top companies, like ByteDance, have computing clusters on a similar scale.

Zhang highlighted that the number of computing cards directly determines the time required to train a large model. With the intense competition in China's AI large model industry, new models are being released almost every month, and companies must upgrade their models within two weeks to stay competitive.

Moore Threads, established in October 2020, focuses on providing accelerated computing infrastructure and one-stop solutions, supporting the digital and intelligent transformation across industries. Zhang , who previously served as NVIDIA's Global Vice President and General Manager of China, founded the company.

In December 2023, Moore Threads launched the KUAE Intelligent Computing Center product suite, featuring the KUAE Intelligent Computing Cluster, based on the MTTS4000 large model computing accelerator card. This suite aims to build the computational infrastructure for large models and artificial general intelligence (AGI).

In the current AI landscape, the availability of GPUs and the ability to build intelligent computing centers are crucial for training more powerful and larger-scale models.

“Scaling from thousand GPU cards to ten thousand GPU cards centers involves more than just increasing the number of cards; it requires addressing exponential challenges in expansion and quality maintenance,” Zhang pointed out.

He acknowledged the conflict between short-term capital and long-term infrastructure investments, noting that the return on investment for intelligent computing centers is around five years, which is generally acceptable to investors.

He also considered the cost-effectiveness of developing GPU cards in China compared to the U.S., emphasizing the significant growth potential in the domestic market.

Currently, Moore Threads' thousand GPU cards computing center has achieved a 50% utilization rate, with a goal of reaching 60% utilization for the ten-thousand-card cluster while maintaining over 99% stability. This places them close to international standards, though challenges remain to achieve a leading position.

Zhang expressed confidence in the future of AGI, anticipating that advancements in algorithms and computing platforms will validate more large model application scenarios.

He noted that the journey to massive-scale clusters is incremental, moving from ten thousand GPU cards clusters to potentially million GPU cards clusters in the future.

Moore Threads holds this milestone sets a new benchmark for domestic GPU technology, enabling a significant leap in computing capabilities for China's AI sector.

“Moore Threads’ commitment to leveraging the KUAE computing cluster to accelerate the development of a better world through enhanced computing capabilities, bridging the gap between industries and benefiting humanity,”Zhang added.

LIKE 0
Related Posts
Traditional Chinese Medicine Embraces the Age of Big Models
Traditional Chinese Medicine Embraces the Age of Big Models
Chinese AI Startup Baichuan AI Raises $5 Billion in Funding
Chinese AI Startup Baichuan AI Raises $5 Billion in Funding
Disappearing Small Banks in China
Disappearing Small Banks in China
Intel Says 'Microcode Algorithm' Voltage Errors the Primary Cause of 13th and 14th Gen CPU Crashes
Intel Says 'Microcode Algorithm' Voltage Errors the Primary Cause of 13th and 14th Gen CPU Crashes
Eli Lilly's Weight Loss Drug Tirzepatide Approved in China Following Novo Nordisk's Wegovy
Eli Lilly's Weight Loss Drug Tirzepatide Approved in China Following Novo Nordisk's Wegovy
Tesla Extends Auto Loan Incentive in China as Q2 Earnings Miss Sends Stocks 12% Lower
Tesla Extends Auto Loan Incentive in China as Q2 Earnings Miss Sends Stocks 12% Lower

  • Subscribe To Our News