NVIDIA Blackwell Ultra GPU to Be Renamed B300 Series

According to TrendForce, NVIDIA has announced a rebranding of its Blackwell Ultra product line to the B300 series, aiming to clearly differentiate it from the upcoming B100 and B200 products. However, the transition to enhanced memory configurations and performance may come at a higher cost.

The B200 Ultra will now be known as the B300, while the GB200 Ultra will be renamed GB300. Other updates include renaming the B200A Ultra to B300A and the GB200A Ultra to GB300A. While NVIDIA has not officially confirmed this rebranding, the previous roadmap indicated "Blackwell Ultra" as a product series, albeit without specific product names.

The B100, B200, and GB200 series will feature 192GB of 8-layer HBM3E memory utilizing 8-layer HBM3E memory stacks. In contrast, the B300 and GB300 will be equipped with 288GB of HBM3E, utilizing 8 stacks of 12-layer memory. This increase in memory capacity, along with potential enhancements such as higher frequency GPUs and additional processing cores, suggests that the pricing for the B300 series will exceed that of the B100 and B200 solutions. TrendForce analysts note that the more cost-effective B300A and GB300A will include 144GB of memory using 4 stacks of 12-layer HBM3E, highlighting a shift for NVIDIA's GPUs towards 12-layer HBM3E, which will necessitate adjustments in the supply chain for DRAM manufacturers.

The B100, B200, and GB200 series are expected to begin shipping between the fourth quarter of 2024 and the first quarter of 2025. Although NVIDIA encountered and resolved some hardware issues, the launch of the B300 series will experience delays. TrendForce anticipates that the B300 series will be introduced in the second or third quarter of 2025, with the B300A—designed specifically for OEMs—entering mass production by mid-2025. Interestingly, TrendForce reports that NVIDIA originally planned to launch the B200A for OEMs but later opted to go directly with the B300A due to lower-than-expected demand for lower-tier GPUs.

While the OEM market is significant for NVIDIA, the company has indicated a priority on its NVL rack solutions, which include configurations with either 36 or 72 Blackwell GPUs, tailored for cloud service providers such as Amazon AWS, Meta, and Microsoft. NVIDIA's NVL36 and NVL72 machines will feature more expensive GPUs, leading to increased costs compared to traditional GPU servers. This is partly due to the high power consumption of the new Blackwell GPUs, which require liquid cooling solutions. Transitioning to such advanced equipment will necessitate businesses to upgrade to more sophisticated data center infrastructures, thereby raising overall operational costs.