- Kyber increases InstaDeep’s compute power 10x, facilitating AI innovation.
- Powered by NVIDIA H100 GPUs, with ~0.5 exaFLOPs in FP16, ranking among the top 100 global AI clusters.
- Investment strengthens InstaDeep’s focus on innovation and growth in France.
24 October 2024: InstaDeep, a leader in artificial intelligence and deep learning, has announced the launch of Kyber, its near-exascale supercomputer. This substantial multi-million-pound investment boosts InstaDeep’s computational capabilities tenfold. Kyber ranks among the world’s top 100 most powerful clusters and is among the top 20 H100 GPU clusters.
Nearing exascale levels, it boasts ~0.5 exaFLOPs of compute power – a major leap in scalable AI infrastructure, repeatable and scalable for future demands. Kyber will play a critical role in advancing InstaDeep’s research in fields such as industrial optimisation and life sciences.
Kyber is hosted at the Equinix data centre in Saint-Denis – just outside Paris – demonstrating InstaDeep’s ongoing commitment to France as a key market. The Paris office is a vital part of the company’s efforts to contribute to the French AI ecosystem and collaborate with local talent, research, and educational institutions.
Karim Beguir, CEO and co-founder of InstaDeep, said: “Kyber is a transformative leap forward for InstaDeep. By integrating cutting-edge NVIDIA H100 GPUs, with our HPC expertise and our dedicated Aichor ML ops software, we now have the capacity to accelerate AI breakthroughs and deliver on our goals for innovative products to our clients.
“This investment highlights France’s strategic importance for InstaDeep. Our Paris office, home to over 130 AI engineers and researchers, plays a key role in driving our capabilities in the region. Hosting Kyber in a French data centre also reflects our commitment to fostering AI innovation and growth in Europe.”
Kyber is fully managed in-house, reinforcing InstaDeep’s commitment to independence in accessing critical computing power for AI workflows. Kyber’s cutting-edge infrastructure, powered by NVIDIA H100 GPUs, offers ~0.5 exaFLOPs in FP16 performance, and enables InstaDeep to tackle highly complex AI challenges, driving new advances in areas like life sciences, industrial optimisation, and advanced AI-driven research.
Nacef Labidi, Head of Infrastructure at InstaDeep, said: “From the outset, InstaDeep made the strategic decision to build and manage its own supercomputing cluster alongside cloud partnerships. This approach, enabled by our High-Performance Computing (HPC) expertise, ensures our independence in accessing critical computing power for AI workflows. Owning and optimising our hardware allows us to tailor software specifically for our needs, driving more efficiency and savings on computing costs.”
InstaDeep’s investment in Kyber underscores its dedication to pushing the boundaries of AI research and development, particularly in France and Europe. With this state-of-the-art supercomputer, InstaDeep is well-positioned to continue delivering impactful products across a range of industries, paving the way for the next generation of AI-driven breakthroughs.
