Nvidia is reportedly exploring reduced memory configurations for its upcoming Rubin Ultra accelerator due to anticipated shortages of High Bandwidth Memory (HBM). Designs are being tested with as little as 192 GB of memory and may revert to using HBM4 instead of the previously announced HBM4E. These potential adjustments stem from manufacturing complexities and supply constraints impacting memory manufacturers like Micron, TSMC, and SK Hynix, with some reports indicating memory shortages could persist through 2030. AI
IMPACT Potential memory constraints could impact the rollout and performance of next-generation AI accelerators.
RANK_REASON Report on potential product configuration changes for a major GPU due to supply chain constraints. [lever_c_demoted from significant: ic=1 ai=0.7]
- HBM4
- HBM4E
- Kyber NVL144
- Micron
- NVIDIA
- Nvidia GTC
- Rubin ultra
- SemiAnalysis
- SK Hynix
- The Information
- TSMC
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →