NVIDIA Unveils Blackwell B100 AI GPUs Doubling Hopper H200 Performance
During the SC23 special address, NVIDIA presented a sneak peek into the performance of their upcoming next-gen GPUs, codenamed Blackwell, which are set to exceed Hopper GPUs’ AI performance by more than double when they debut in 2024.
Drawing from the success of its Hopper and Ampere GPUs over the past two years, NVIDIA’s new Blackwell GPUs will be a game-changer for the AI and HPC industry. The company witnessed a significant revenue boost in its data center and overall business due to the AI craze, setting the stage for the upcoming launch of two brand new GPU families by 2025.
The first of these new AI/HPC GPU families is the Blackwell series, named after renowned mathematician David Harold Blackwell (1919-2010). It is set to succeed the Hopper H200 series and will utilize the B100 chip. The product lineup is expected to include various offerings such as GB200NVL (NVLINK), the standard GB200, and the B40 for visual compute acceleration, with an anticipated unveiling at the GTC in 2024 and a subsequent launch in the same year.
Rumors suggest that NVIDIA will manufacture the Blackwell GPUs using the TSMC 3nm process node, with initial deliveries to customers projected by the end of 2024 (Q4). Moreover, NVIDIA is reportedly fast-tracking production to Q2 2024, aligning with the availability of their Hopper H200 GPUs. Samsung is expected to play a significant role as a memory provider for the Blackwell GPUs.
The Blackwell GPUs are also anticipated to be the first HPC/AI accelerators from NVIDIA to feature a chiplet design, posing a direct competition to AMD’s Instinct MI300 accelerator in the AI space. Additionally, NVIDIA is set to unveil the GX200 chip as a follow-up to Blackwell in 2025, following its established two-year cadence between AI and HPC products.
These new developments are part of NVIDIA’s broader strategy to deliver major performance upgrades with new networking and interconnect interfaces, offering up to 800 Gb/s transfer speeds by 2024 and up to 1600 Gb/s transfer speeds by 2025, as part of its Quantum and Spectrum-X products.
With the latest announcement, NVIDIA is cementing its position as a key player in the AI and HPC industry, poised to drive major advancements in GPU technology and AI performance.