Archive — US — turtlesai
Nvidia Vera Rubin, Rubin Ultra confirmed for 2026 and 2027. Up to 3.3x and 14xBlackwell Ultra GB300
Nvidia is set to launch its AI systems based on the Rubin and Rubin Ultra platforms in 2026 and 2027, featuring Vera GPU and CPU chips, enhanced memory capacity, and innovative NVLink interconnects. The NVL144 system aims for 50 PFLOPS performance, while the NVL576 will double the GPU capacity, delivering up to 100 PFLOPS and 365 TB memory, marking a transformative moment in AI computing infrastructure.
Key points
- Nvidia will introduce the Rubin and Rubin Ultra platforms in 2026 and 2027.
- The systems will leverage new Vera GPU and CPU chips, significantly improving compute and memory performance.
- The NVL144 system will achieve up to 50 PFLOPS of FP4 performance and 288 GB of HBM4e memory.
- The NVLink architecture will enhance bandwidth and efficiency, offering up to 1.8 TB/s communication.
- Rubin Ultra NVL576 doubles the GPU configurations, achieving 100 PFLOPS and up to 1 TB of HBM4e memory.
- Projected performance improvements of 3.3x for NVL144 and 14x for NVL576 over previous models.
Why it matters
- The developments align with increasing demands for AI capabilities amidst the rise of deep learning applications.
- Relates to ongoing IT trends focusing on high-performance computing and AI-driven solutions.
- NVIDIA's technological strides reflect broader shifts in the tech industry towards enhancing infrastructure for data-intensive operations.
Nvidia's upcoming Rubin platforms signify a substantial leap in AI computing capabilities, setting new performance benchmarks and transforming data center architectures, reinforcing its role as a leader in AI innovations.