NVIDIA’s next-generation Blackwell Ultra GPUs for AI computing use the B300 branding and feature an upgraded HBM3e specification.
NVIDIA Announces B300 “Blackwell Ultra” AI Series, an Upgrade Package with Enhanced HBM3e Features
TrendForce reports that the original Blackwell Ultra lineup, which was said to be the successor to the company’s Blackwell products, has reportedly been changed to a new naming convention called the B300 series (via TrendForce), a relatively compact tag. Masu. Under this brand, the B200 Ultra AI accelerator and GB200 Ultra AI server (now renamed B300 and GB300, respectively) and similar cases also apply to “A” series AI products such as the B200A Ultra. Masu. .
The motivation behind this move has not been defined yet, but NVIDIA will likely change the naming scheme to represent the architectural changes the company plans to make with the B300 series.
Interestingly, NVIDIA’s planned original B200A series will not be released to the market. The B200A series will be aligned with the launch of the B300 lineup under the B300A “Blackwell Ultra” tag and is expected to attract enterprise customers who don’t want to go all out when it comes to building their own AI computing power. .
NVIDIA’s existing Blackwell portfolio is fully booked for the next 12 months and is seeing incredible adoption rates backed by insane demand. The company is expected to generate billions of dollars in revenue in the next quarter and even more in early 2025.
The NVIDIA B300 Blackwell Ultra lineup features a combination of HGX, MGX, and NVL configurations, with NVL-36 and NVL72 serving as server and data center powerhouses. These chips include up to 288 GB of HBM3e memory and are also available in 144 GB packages. For comparison, AMD’s recently launched MI325X GPU offers 256 GB of capacity, and the upcoming MI350X, scheduled to launch in the second half of 2025, will offer up to 288 GB of capacity. The capacity is also GB.
In terms of release dates, NVIDIA’s B300 “Blackwell Ultra” AI products are expected to hit the market between Q2 and Q4 of 2025. Team Green is considering 12 Hi HBM3e memory along with CoWoS-L in its lineup. It is intended to significantly improve performance.
