IGenius, a company specializing in AI models for highly regulated industries, announced the Colosseum computing platform. It is claimed to be one of the world’s most powerful AI supercomputers powered by the NVIDIA DGX SuperPOD platform with thousands of GB200 (Blackwell) accelerators.
It is known that the Colosseum complex is located in Europe. The full characteristics of the supercomputer are not disclosed. It is noted that it provides performance of up to 115 Eflops on AI operations (FP4 with sparsity). It talks about the use of an advanced liquid cooling system. Power is supplied from renewable energy sources in Italy.
According to Reuters, Colosseum will include about 80 GB200 NVL72 super accelerators. This brings the total number of Blackwell accelerators to 5,760. The total system power consumption should be almost 10 MW. The cost of the project has not been announced. But the head of iGenius, Uljan Sharka, notes that the company has raised approximately €650 million for development during 2024 and intends to obtain additional funding for the Colosseum project. It is emphasized that iGenius is one of the few AI startups in Europe whose capitalization exceeds $1 billion.
Image source: iGenius
IGenius plans to use Colosseum for computationally intensive AI applications, including training large language models (LLMs) with trillions of parameters, as well as working with open source generative AI models. It is emphasized that the creation of Colosseum will become the basis for the next stage of cooperation between iGenius and NVIDIA in the field of AI to support tasks that require maximum data security, reliability and accuracy: this could be financial consulting, patient care in the healthcare system, government planning, etc.
IGenius AI models built using the NVIDIA AI Enterprise platform, NVIDIA Nemotron and the NVIDIA NeMo framework will be offered as NVIDIA NIM microservices. According to iGenius, Colosseum will help meet the growing needs for AI computing. Colosseum will also serve as a hub of sorts, bringing together businesses, academic institutions and government agencies.
It should be noted that about a month ago, DeepL, a company specializing in the development of automatic translation tools based on AI, announced its intention to deploy a platform based on NVIDIA DGX GB200 SuperPod in Sweden. DeepL will use this complex for research tasks, in particular, to develop advanced AI models.