HIVE Digital Technologies partnered with researchers from Columbia University to successfully demonstrate that its older-generation Nvidia A40 GPUs located in Asunción, Paraguay, can match the performance of Nvidia’s advanced H100 chips for specific AI training tasks. This groundbreaking research, which involved the remote training of large language models over a distance of 5,000 miles, highlights the effective software optimizations made by the team, reinforcing the potential for intercontinental AI training frameworks. The results from this collaboration have been submitted to the Neural Information Processing Systems (NeurIPS) conference, underlining HIVE’s strategy to leverage Paraguay’s abundant renewable energy resources as a foundation for expanding its AI infrastructure, as the demand for powerful computing capabilities continues to grow worldwide.
NeurIPS: NeurIPS, or the Neural Information Processing Systems conference, is one of the world’s premier academic events for machine learning and artificial intelligence research. HIVE and Columbia University submitted their paper on optimizing older GPUs for LLM pretraining to the conference. Acceptance would provide peer-reviewed validation of the intercontinental training results.
Frank Holmes: Frank Holmes serves as Executive Chairman and co-founder of HIVE Digital Technologies, where he oversees strategic initiatives including the company’s pivot toward AI infrastructure. He highlighted the Paraguay research as a key proof point for expanding high-performance computing in non-traditional locations like South America. Holmes has emphasized the role of renewable electricity in enabling global AI development.
Columbia University: Columbia University is a leading research institution with a Department of Industrial Engineering and Operations Research focused on optimization techniques and AI applications. Researchers from the university carried out the remote LLM pretraining experiments using HIVE’s GPUs in Paraguay while based in New York. The team submitted their findings on intercontinental AI training to the NeurIPS conference.
HIVE Digital Technologies: HIVE Digital Technologies is a cryptocurrency mining and digital infrastructure company that operates facilities powered by renewable energy across multiple countries. It has been expanding its operations into high-performance computing and AI services to diversify beyond Bitcoin mining. The company conducted the Columbia University research demonstrating the viability of its Paraguay GPU infrastructure for advanced AI workloads.
`json
{
“Performance Optimization”: “Software and algorithmic improvements are enabling older GPU hardware to deliver competitive performance in specific AI training scenarios when compared to newer flagship processors.”,
“Renewable Power Advantage”: “Access to abundant, low-cost renewable electricity is becoming a vital enabler for the development of AI infrastructure in various global locations.”,
“AI Infrastructure Expansion”: “Facilities initially used for Bitcoin mining, powered by renewable energy, are increasingly being repurposed for AI workloads due to growing global demand for computing resources.”
}
`