The innovation in artificial intelligence (AI) has reached a new level with the announcement of ExaAILabs, which has implemented its Exacluster, one of the first industry clusters based on Nvidia H200 Hopper GPUs. This cluster is not only a technological milestone but also aims to revolutionize the way we interact with search engines, with the ambitious goal of surpassing Google in understanding and processing complex queries.
A Look at the Exacluster
The Exacluster stands out for its impressive power and specialized design:
- Hardware: It consists of 18 servers, each equipped with 8 Nvidia H200 GPUs, totaling 144 GPUs with 20TB of HBM3E memory (141GB per GPU). This design offers a combined performance of 570 PetaTOPS, a crucial capacity for training advanced neural networks.
- Processors: 192 CPUs with 96 cores, reaching a total of 3,456 processing cores.
- Memory and Storage:
- 36TB of DDR5 memory.
- 270TB of NVMe storage.
- Cooling and Energy Efficiency: Despite its power, the Exacluster uses standard air cooling and consumes 100kW of energy. Only two machines are installed per rack to ensure optimal cooling.
A Strategic and Economic Investment
The total cost of the Exacluster amounts to $5 million, which equals approximately $277,777 per machine. This surprisingly low price for such an advanced system has been made possible thanks to ExaAILabs’ close relationship with Nvidia, one of its main investors alongside Lightspeed and YCombinator.
Early access to the H200 GPUs also highlights ExaAI’s privileged position in the industry, allowing it to leverage the latest technology for unique AI projects.
The Goal: Transforming Search Engines
The Exacluster will be dedicated to training advanced neural networks with a clear purpose: to develop a search engine capable of understanding user queries more accurately and delivering more relevant results than the current leading engines, including Google. This engine aims to better interpret natural language and respond to complex queries, redefining what we expect from a search engine.
Impact and Challenges
The deployment of the Exacluster underscores how high-performance computing (HPC) and AI are converging to tackle increasingly complex problems. If ExaAILabs achieves its goal, its search engine could significantly disrupt the dynamics of the tech industry and open new possibilities in using AI to understand and process data.
However, competition in this space is fierce, and expectations are high. ExaAI’s ability to innovate and scale its technology will determine whether this cluster becomes the catalyst for a new era in search engines.
via: tom’s hardware