A analysis mission carried out by Columbia College professors on GPU servers situated greater than 5,000 miles away in Paraguay has produced outcomes that HIVE Digital Applied sciences (Nasdaq: HIVE) is in search of as a proof of idea for future distributed AI infrastructure.
HIVE Digital, a publicly traded information middle and digital asset mining firm based in 2017 and traded on each the TSX and Nasdaq, introduced on June 22 that it has accomplished its first AI analysis collaboration with Columbia College’s Division of Industrial Engineering and Operations Analysis.
The analysis was introduced on the Neural Data Processing Methods Convention (NeurIPS), one of many world’s most influential machine studying conferences together with ICLR and ICML.
What the analysis really discovered
Working from New York Metropolis, the Colombian crew remotely ran iterative AI coaching experiments on HIVE’s A40 GPU cluster in Asunción, Paraguay.
The analysis targeted on pre-training neural networks, the basic course of by which large-scale AI fashions study from information earlier than being fine-tuned for a particular process.
Headline discovery is essential. After two months of particular code optimizations for HIVE’s A40 {hardware}, Columbia researchers discovered that it matched the efficiency of Nvidia’s H100 GPU, a more moderen and significantly dearer era of chips that’s extensively thought-about the gold commonplace for AI coaching workloads.
Comparisons had been made after normalizing the uncooked efficiency of every {hardware} platform.
The crew additionally ran commonplace benchmarks on the LLaMA mannequin as a part of their analysis, testing service throughput and latency on the 1.4 billion parameter language mannequin.
“For the use case of pre-training an LLM with as much as 1.4B parameters, the outcomes match the H100 outcomes after normalizing the uncooked efficiency of every {hardware},” mentioned Columbia researchers within the IEOR division.
“This analysis advances our understanding of recent neural community optimization, together with matrix-aware optimizers reminiscent of Muon and associated scale-invariant strategies.”
Trending on TheStreet Roundtable:
- Wall Road Analysts Predict 400% Rise in MicroStrategy Inventory
- Billionaire reveals Satoshi’s shocking proposal on Bitcoin
- Mark Zuckerberg eyes new $1 billion market
Why Paraguay and what occurs subsequent?
The Asunción outcomes are at present serving as a efficiency baseline for a big infrastructure mission that HIVE is constructing in Iguazu, Paraguay. The corporate is constructing a 100-megawatt substation there, and civil works have already been accomplished.
The substation is scheduled to be energized in September 2026, and development of the brand new Tier-III information middle will start within the fall, with a goal service date of late 2027.
HIVE’s wager on Paraguay is rooted within the nation’s power profile. Low cost and considerable hydropower makes it a gorgeous location for power-intensive computing workloads.
Frank Holmes, HIVE Government Director, mentioned: “Paraguay has energy, it has a strategic location, and now it proves it.”
“Excessive-performance computing would not must be restricted by geography. Our imaginative and prescient is to offer the suitable energy, information middle design, software program stack, and execution to allow Paraguay to take part immediately within the international AI financial system via HIVE.”
HIVE’s broader story
When based in 2017, HIVE was one of many first publicly traded firms to make use of inexperienced power to mine Bitcoin. We at present function Tier-I and Tier-III information facilities in Canada, Sweden, and Paraguay, serving each Bitcoin mining purchasers and excessive efficiency computing prospects.
The corporate has beforehand collaborated with Intel Company on customized mining {hardware}, making it certainly one of Sweden’s largest demand response members and serving to to steadiness the nationwide energy grid.
The collaboration with Columbia represents HIVE’s clearest step but into AI infrastructure as a industrial enterprise line, transferring past crypto mining and into the broader marketplace for GPU computing.
”Innovation is in our DNA” mentioned Aydin Kilic, President and CEO of HIVE. ”We consider that the way forward for AI in Paraguay is shiny. ”
Kilic added that good engineering supplies nice worth. He added that HIVE will proceed to speculate locally whereas bringing developments in its know-how stack to information facilities all over the world.
