A analysis mission performed by Columbia College professors on GPU servers situated greater than 5,000 miles away in Paraguay has produced a outcome that HIVE Digital Applied sciences (Nasdaq: HIVE) calls a proof of idea for the way forward for distributed AI infrastructure.
HIVE Digital, a publicly traded information middle and digital asset mining firm listed on the TSX and Nasdaq, introduced on June 22 that it has accomplished its first AI analysis collaboration with the Division of Industrial Engineering and Operations Analysis at Columbia College.
The analysis was offered on the Convention on Neural Info Processing Methods (NeurIPS), one of many three most influential machine studying conferences on the earth, together with ICLR and ICML.
What the analysis actually discovered
The Columbia staff performed iterative AI coaching experiments remotely on HIVE’s A40 GPU cluster situated in Asunción, Paraguay, whereas working from New York Metropolis.
The research targeted on pre-training neural networks, the elemental course of by which massive AI fashions be taught from information earlier than fine-tuning them for particular duties.
The headline’s discovering is critical. After two months of optimizing code designed particularly for HIVE’s A40 {hardware}, Columbia researchers discovered that efficiency matched that of Nvidia’s H100 GPUs, a more moderen and significantly costlier era of chip extensively thought of the gold normal for AI coaching workloads.
The comparability was carried out after normalizing the uncooked efficiency of every {hardware} platform.
The staff additionally examined the service’s efficiency and latency on a 1.4 billion-parameter language mannequin, working normal benchmarks on LLaMA fashions as a part of the research.
“In our pre-training LLM use case of as much as 1.4 billion parameters, our outcomes match these of H100 after normalizing the uncooked efficiency of every {hardware},” mentioned a Columbia researcher from the IEOR division.
“This work improves our understanding of recent neural community optimization, together with matrix-aware optimizers like Muon and associated scale-invariance strategies.”
Developments on TheStreet Roundtable:
- Wall Avenue Analyst Predicts 400% Rise for MicroStrategy Inventory
- Billionaire reveals astonishing proposal for Satoshi’s Bitcoin
- Mark Zuckerberg eyes one other billion-dollar market
Why Paraguay and what comes subsequent
The outcomes from Asunción now function a efficiency foundation for a bigger infrastructure mission that HIVE is constructing in Yguazú, Paraguay. The corporate has a 100 megawatt substation underneath development there and the civil works are already accomplished.
The substation is predicted to be energized in September 2026, development of a brand new Tier III information middle will start within the fall and the supply date for service is deliberate for the second half of 2027.
HIVE’s dedication to Paraguay is rooted within the nation’s vitality profile: considerable and low-cost hydropower that makes it a beautiful location for energy-intensive computing workloads.
HIVE govt hairdresser Frank Holmes mentioned: “Paraguay has the facility, the strategic location and now the testing level.”
“Excessive efficiency computing doesn’t should be restricted by geography. Our imaginative and prescient is to offer the precise energy, information middle design, software program stack and execution, in order that by means of HIVE, Paraguay can immediately take part within the world AI economic system.”
HIVE’s wider historical past
HIVE was one of many first publicly traded corporations to mine Bitcoin utilizing inexperienced vitality when it was based in 2017. As we speak it operates Tier I and III information facilities in Canada, Sweden and Paraguay, serving each Bitcoin mining shoppers and high-performance computing shoppers.
The corporate has beforehand collaborated with Intel Company on customized mining {hardware} and have become certainly one of Sweden’s largest demand response contributors, serving to to steadiness the nationwide energy grid.
The Columbia collaboration represents HIVE’s clearest step but towards AI infrastructure as a business line of enterprise, shifting past cryptocurrency mining into the broader GPU computing market.
“Innovation is in our DNA“mentioned HIVE President and CEO Aydin Kilic. ““We see a vibrant future for AI in Paraguay.”
Kilic added that nice engineering can generate vital worth. HIVE will proceed to put money into communities whereas bringing developments in our expertise to our information facilities globally, he added.
