Nvidia trims HBM4 bandwidth for Rubin platform, sources say

DIGITIMESen

Nvidia trims HBM4 bandwidth for Rubin platform, sources say

Nvidia is reportedly adjusting the fourth-generation high-bandwidth memory (HBM4) specification for its Vera Rubin NVL72 platform, cutting bandwidth per GPU from the previously announced 22TB/s to 19.2TB/s, a reduction of about 13%. South Korean industry observers said the move could increase supply flexibility and help balance performance with mass-production delivery, although the reason for the change and its actual impact on supplier orders remain unconfirmed.

This is a short summary published by AI Global Wire. The full article is owned and hosted by DIGITIMES — open it there to read it in full.

Read the full story at DIGITIMES

    Related AI news