Home / Technology / Nvidia's 50,000-Core Rubin GPU Redefines AI Supercomputing
Nvidia's 50,000-Core Rubin GPU Redefines AI Supercomputing
22 Nov
Summary
- Nvidia's new Rubin HPC boasts over 50,000 CUDA cores for scientific tasks.
- It surpasses the RTX 5090's core count by nearly three times.
- The Rubin HPC is slated for deployment in systems starting in 2028.

Nvidia has announced its new Rubin HPC accelerator, featuring over 50,000 CUDA cores. This specialized processor is engineered for demanding scientific simulations and extensive data analysis, moving beyond traditional gaming applications.
The Rubin HPC significantly exceeds the core count of consumer GPUs like the RTX 5090, promising substantial performance gains. Nvidia anticipates that this new chip will be integrated into deployed systems beginning in 2028.
This accelerator is designed to bolster research in fields such as materials science, molecular modeling, and climate system analysis. It incorporates advancements in data movement and sparse data acceleration to ensure efficient processing.
NVLink technology is also updated to its seventh generation, enhancing inter-GPU communication for distributed computing. Nvidia highlights the role of Rubin HPC in accelerating AI-driven scientific discovery, enabling researchers to explore more possibilities faster.