NVIDIA Rubin Ultra HBM reduced to 192GB due to supply and power constraints

Huoxing Finance reports that on August 3, research firm SemiAnalysis released a report stating that NVIDIA’s preview specifications for Rubin Ultra fell below prior market expectations. The report notes that the mainline SKU of Rubin Ultra will still utilize HBM4, with theoretical peak FLOPs remaining unchanged, but HBM configuration has been reduced to 8-Hi, 192GB, down from Rubin’s previous 12-Hi, 288GB. The chip-level power consumption is approximately 1800W, similar to the initial Rubin shipping version. The report highlights that the primary upgrades in Rubin Ultra’s current roadmap focus on scalable interconnects at the NVL576 level. One system configuration involves interconnecting eight 72-GPU racks using NPO to enable cross-rack switch connectivity. SemiAnalysis considers this configuration one of the most plausible current options. SemiAnalysis indicates that NVIDIA may still offer a high-power Max-P variant ranging from 2600W to 2800W, as well as a lower-power 1200W version tailored for lighter workloads such as token decoding. The 1800W Max-Q variant better aligns with current data center power constraints and offers superior deployment efficiency in terms of FLOPs per watt. SemiAnalysis believes NVIDIA’s adjustment of Rubin Ultra’s specifications is primarily driven by HBM supply constraints and data center power limitations. Reducing HBM capacity and power requirements enhances customer deployability, lowers material costs, and mitigates pressure from anticipated HBM price increases next year.