Evidence40%Reported
FactConfirmedProduct·March 19, 2026
Nvidia integrates Groq 3 LPU into Vera Rubin platform, removes Rubin CPX from roadmap
Nvidia's GTC 2026 keynote omitted the previously announced Rubin CPX context phase accelerator, replacing it with the Groq 3 LPU (512MB SRAM, 150 TB/s bandwidth per die) as the dedicated inference co-processor, with the LP30 chip shipping Q3 2026 on Samsung 4nm.
Evidence Strength
Evidence40%Reported
Based on trade press
Single publisher source
Key Development
Transformative development (significance 9-10)
Covered by 4 sources
Confirmed — verified event
Insights
First tracked
March 17, 2026
Last updated
March 19, 2026
Sources
4 sources
Related Developments
NVIDIA Launches Nemotron 3 Super 120B Open Model for Agentic AINvidia launches BlueField-4 STX storage architecture for agentic AINVIDIA Restarting Manufacturing of China-Specific AI Chip VariantNvidia updates long-term data center roadmap: Feynman GPU with die stacking, Rosa CPU, optical NVLink for 2028Nvidia demonstrates Rubin Ultra tray with 1TB HBM4E, Kyber NVL144 rack design for 2027
Sources (4)
Source Timeline
Evidence Strength
Evidence40%Reported
Based on trade press
Single publisher source
Key Development
Transformative development (significance 9-10)
Covered by 4 sources
Confirmed — verified event
Insights
First tracked
March 17, 2026
Last updated
March 19, 2026
Sources
4 sources
Related Developments
NVIDIA Launches Nemotron 3 Super 120B Open Model for Agentic AINvidia launches BlueField-4 STX storage architecture for agentic AINVIDIA Restarting Manufacturing of China-Specific AI Chip VariantNvidia updates long-term data center roadmap: Feynman GPU with die stacking, Rosa CPU, optical NVLink for 2028Nvidia demonstrates Rubin Ultra tray with 1TB HBM4E, Kyber NVL144 rack design for 2027