NVIDIA has revised the specifications for its upcoming Rubin GPU, reducing the HBM4 memory bandwidth and NVLink interconnect bandwidth compared to earlier projections. The Rubin GPU will now offer 19.2 TB/second of memory bandwidth per GPU, down from a previous target of 22 TB/second, and NVLink bandwidth will be reduced from 3.6 TB/second to 3.0 TB/second. However, scale-out networking bandwidth has been increased to 450 GB/second, up from 400 GB/second. These changes are occurring as NVIDIA begins the production ramp-up for Vera Rubin systems, with the first NVL72 systems becoming available to customers. AI
IMPACT Potential performance implications for AI workloads due to revised memory and interconnect bandwidth specifications.
RANK_REASON Hardware specification update for an upcoming major GPU release from a leading AI chip manufacturer. [lever_c_demoted from significant: ic=1 ai=0.7]
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →