律动BlockBeats|Aug 03, 2026 06:39
[SemiAnalysis: NVIDIA Rubin Ultra Mainline Version HBM May Be Reduced to 192GB, Data Center Power Supply and HBM Supply as Key Constraints]
BlockBeats News, August 3, research institution SemiAnalysis released a report stating that NVIDIA's Rubin Ultra specifications previewed to key customers are lower than previous market expectations. The report indicates that the Rubin Ultra mainline SKU will still use HBM4, with theoretical peak FLOPs remaining unchanged, but the HBM configuration will be reduced to 8-Hi, 192GB, compared to the previous Rubin's 12-Hi, 288GB. Chip-level power consumption is approximately 1800W, similar to the initial Rubin shipment version.
The report highlights that the primary upgrade point in the current Rubin Ultra roadmap focuses on NVL576-level large-scale interconnects. One system configuration involves interconnecting 8 racks of 72 GPUs, using NPO to achieve cross-rack connections between switches. SemiAnalysis believes this configuration is one of the more likely candidate solutions.
SemiAnalysis notes that NVIDIA may still offer a Max-P high-power version ranging from 2600W to 2800W, as well as a 1200W version targeting lower computational loads such as token decoding. The 1800W Max-Q version, however, aligns better with the current data center power supply-constrained environment, offering greater deployment efficiency in terms of FLOPs/Watt.
SemiAnalysis believes the core reason for NVIDIA's adjustment of Rubin Ultra specifications lies in tight HBM supply and data center power constraints. Reducing HBM capacity and power consumption specifications helps improve deployment feasibility for customers, lower material costs, and mitigate the pressure of HBM price increases next year.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink