A project creating a domestic supercomputer complex for one of Russia's largest banks demonstrates significant progress in import substitution for high-performance computing. Collaboration between AMD Technologies and E-Flops resulted in a unique "dual-circuit" architecture that effectively separates data streams for scientific calculations and artificial intelligence tasks.
A key feature is the use of modern NVIDIA H100 accelerators with 80GB memory, providing unprecedented performance of 700 Petaflops for AI tasks. This power allows the banking sector to transition to more complex machine learning models, including natural language processing and real-time predictive analytics.
Importantly, the project was implemented entirely by Russian specialists, holding strategic significance for the country's technological sovereignty. High-speed InfiniBand NDR interfaces with 200 Gbit/s throughput ensure minimal latency during data transmission between computing nodes.
Deploying such infrastructure in the financial sector opens new possibilities for process automation, enhanced transaction security, and developing personalized financial products. This precedent could become a starting point for scaling similar solutions across other sectors of the Russian economy.
System parameters, including over 4,000 processor cores and nearly 35 terabytes of RAM, indicate the seriousness of Russian companies' intentions to independently solve tasks previously requiring foreign solutions.