2 Sources
[1]
NVIDIA renames Blackwell Ultra to B300 series: HBM3E 12-Hi memory, TSMC CoWoS-L packaging
NVIDIA has reportedly rebranded all of its upcoming Blackwell Ultra products to the B300 series, with the beefed-up B300 and GB300 chips to also reportedly use TSMC's new CoWoS-L advanced packaging. In a new report from TrendForce, we're learning that the B200 Ultra has been renamed to the B300, while the GB200 Ultra has been renamed to the GB300. On top of that, the B200A Ultra and GB200A Ultra will be called the B300A and GB300A, respectively. NVIDIA is expected to launch its now rebranded B300 and GB300 chips in Q2 2025 to Q3 2025, whilew the B200 and GB200 is shipping in small quantities now, more throughout Q4 2024 and things really kick off in Q1 2025. TrendForce notes that NVIDIA is "refining its Blackwell chip segmentation to better meet the performance demands of CSPs and the cost-performance needs of server OEMs, allowing for flexibility based on supply chain capabilities. The B300A is specifically aimed at OEM clients, with mass production projected to commence in the second quarter of 2025 following peak shipments of the H200". NVIDIA's new B300 and GB300 AI chips will be using new HBM3E 12-Hi stack memory, which SK hynix recently showed off. The faster HBM3E 12-Hi memory is expected to enter production between Q4 2024 and Q1 2025, with suppliers requiring around 6 months to refine their processes and stabilize production yields reports TrendForce.
[2]
NVIDIA Blackwell Ultra AI GPUs To Be Called B300, Feature 12-Hi HBM3E & TSMC CoWoS-L Tech
NVIDIA's next-gen Blackwell Ultra GPUs for AI computing are going to use the B300 branding, & feature upgraded HBM3e specs. NVIDIA To Debut The B300 "Blackwell Ultra" AI Series, An Upgraded Package With Heightened HBM3e Capabilities TrendForce reports that the original Blackwell Ultra lineup, which was said to be the immediate successor to the firm's Blackwell products, has reportedly been shifted to a new naming convention, called the B300 series (via TrendForce), which is a relatively more compact tag. Under this branding, the B200 Ultra AI accelerator and the GB200 Ultra AI server, which have now been renamed as the B300 and the GB300, respectively, and a similar case would apply to the "A" series AI products, such as the B200A Ultra. While the motive behind this move hasn't been defined yet, NVIDIA is likely switching up the naming scheme to depict the architectural changes the firm is going to make with the B300 series. Interestingly, the original B200A series planned by NVIDIA isn't anticipated to be released in the markets. The B200A series would now be adjusted with the launch of the B300 lineup under the B300A "Blackwell Ultra" tag, which is hoped to attract enterprise clients who aren't looking to go all out when it comes to building their own AI computing power. NVIDIA's existing Blackwell portfolio has been booked out for the next 12 months and seeing an incredible adoption rate backed by insane demand. The company is expected to earn Billions in revenues in the coming quarter and even more so at the start of 2025. The NVIDIA B300 Blackwell Ultra lineup will feature a mix of HGX, MGX and NVL configurations with NVL-36 and NVL72 being the powerhouse for servers and datacenters. These chips will incorporate up to 288 GB of HBM3e memory and also come in 144 GB packages For comparison, the recently launched MI325X GPU from AMD offers 256 GB capacities and the upcoming MI350X which is going to target a 2H 2025 launch will offer up to 288 GB capacities too. In terms of release dates, NVIDIA's B300 "Blackwell Ultra" AI products are expected to be available in the markets by Q2-Q4 of 2025. Team Green is looking to adopt 12-hi HBM3e memory along with CoWoS-L in the lineup, with the intent of bringing massive performance uplifts.
Share
Copy Link
NVIDIA has renamed its upcoming Blackwell Ultra products to the B300 series, incorporating HBM3E 12-Hi memory and TSMC's CoWoS-L packaging. The new lineup aims to meet diverse performance demands in the AI chip market.
NVIDIA, a leader in AI and graphics processing technology, has announced a significant rebranding of its upcoming Blackwell Ultra product line to the B300 series. This move, reported by TrendForce, marks a strategic shift in NVIDIA's product segmentation and naming convention 1.
The rebranding affects several products in NVIDIA's lineup:
This new naming scheme is believed to reflect architectural changes in the B300 series, although NVIDIA has not officially stated the reason for the shift 2.
The B300 series will incorporate cutting-edge technologies:
HBM3E 12-Hi Memory: The new chips will utilize advanced High Bandwidth Memory (HBM) technology, with SK hynix's HBM3E 12-Hi stack memory expected to enter production between Q4 2024 and Q1 2025 1.
TSMC CoWoS-L Packaging: NVIDIA will employ TSMC's advanced Chip-on-Wafer-on-Substrate with Local Silicon Interconnect (CoWoS-L) packaging technology for the B300 and GB300 chips 1.
Memory Capacity: The B300 series is expected to offer up to 288 GB of HBM3e memory, with 144 GB packages also available. This puts NVIDIA in direct competition with AMD's upcoming MI350X GPU, which is also targeting 288 GB capacities 2.
NVIDIA's rebranding strategy appears to be aimed at better meeting the diverse needs of its client base:
This staggered release strategy allows NVIDIA to refine its chip segmentation, catering to both the high-performance demands of Cloud Service Providers (CSPs) and the cost-performance needs of server OEMs 1.
Elon Musk's companies X and xAI have filed a lawsuit against Apple and OpenAI, alleging anticompetitive practices in the integration of ChatGPT into iOS, claiming it stifles competition in the AI chatbot market.
50 Sources
Technology
17 hrs ago
50 Sources
Technology
17 hrs ago
YouTube has been secretly testing AI-powered video enhancement on select Shorts, leading to backlash from creators who noticed unexpected changes in their content. The platform claims it's using traditional machine learning, not generative AI, to improve video quality.
7 Sources
Technology
17 hrs ago
7 Sources
Technology
17 hrs ago
IBM and AMD announce a partnership to develop next-generation computing architectures that combine quantum computers with high-performance computing, aiming to solve complex problems beyond the reach of traditional computing methods.
4 Sources
Technology
9 hrs ago
4 Sources
Technology
9 hrs ago
An investigation into how AI chatbot design choices, particularly sycophancy and anthropomorphization, are leading to concerning cases of AI-related psychosis and addiction among vulnerable users.
5 Sources
Technology
17 hrs ago
5 Sources
Technology
17 hrs ago
Leading tech firms and investors create a network of political action committees to advocate for AI-friendly policies and oppose strict regulations ahead of the 2026 midterms.
5 Sources
Policy
17 hrs ago
5 Sources
Policy
17 hrs ago