NVIDIA has revealed the sneak preview teaser of its next-gen Blackwell B100 GPUs which will more than double the efficiency of Hopper H200 in 2024.
NVIDIA Blackwell B100 AI GPUs To Offer More Than 2x Performance Versus Hopper H200 GPUs In 2024
Throughout its SC23 unique address, NVIDIA teased the efficiency of its next-gen GPUs codenamed Blackwell which will use more than 2x the AI efficiency of Hopper GPUs when they make their launching in 2024. The GPU utilized was the next-generation B100 which will prosper the Hopper H200 and can be seen simply squashing the GPT-3 175B reasoning efficiency criteria, showcasing its huge AI efficiency capacity.
For the previous 2 years, NVIDIA has actually counted on its Hopper and Ampere GPUs to serve the requirements of AI & HPC consumers worldwide, working together with numerous partners, however all of that will alter in 2024 with the arrival of Blackwell. NVIDIA saw a huge increase to its information center and general business income thanks to the AI fad and it appears like that train is going complete steam ahead as the green group is intending to release 2 brand name brand-new GPU households by 2025.
The very first of these brand-new AI/HPC GPU households from NVIDIA is going to be Blackwell, called after David Harold Blackwell(1919-2010). The GPU will be the follower to the GH200 Hopper series & will utilize the B100 chip. The business intends on using numerous items consisting of GB200NVL( NVLINK ), the basic GB200, and the B40 for visual calculate velocity. The next-gen lineup is anticipated to be revealed at the next GTC (2024) followed by a launch at some point later on in 2024.
Current reports approximate that NVIDIA will be using the TSMC 3nm procedure node for producing its Blackwell GPUs and the very first consumers will be provided the chips by the end of 2024 (Q4) though the current reports have actually highlighted that NVIDIA is fast-tracking the production to Q2 2024 which is the exact same time its just recently revealed Hopper H200 GPUs will be offered. Samsung is stated to be a significant memory service provider for NVIDIA's next-gen Blackwell GPUs too.
The GPU is likewise anticipated to be the very first HPC/AI accelerator from NVIDIA to make use of a chiplet style and will be taking on AMD's Instinct MI300 accelerator which is likewise going to be a huge offer within the AI area as the red group has actually promoted it to be.
The other chip that has actually been divulged is the GX200 and this one is the follow-up to Blackwell with a launch arranged for 2025. Now NVIDIA has actually been following a two-year cadence in between its AI & HPC items so it is most likely that we may just see a statement of the chip by 2025 with real systems to start deliveries by 2026.
The lineup will be based upon the X100 GPU and will consist of a GX200 lineup of items and a different X40 lineup for Enterprise consumers. NVIDIA is understood to call its GPUs after widely known researchers and it currently utilizes the Xavier codename for its Jetson series so we can anticipate a various researcher name for the X100 series. That, there's little that we understand about the X100 GPUs however it is much better than the Hopper-Next codenames that NVIDIA is utilizing in previous roadmaps.
NVIDIA likewise prepares to provide significant "doubling" upgrades on its Quantum and Spectrum-X with brand-new Bluefield and Spectrum items, providing to 800 Gb/s transfer speeds by 2024 and approximately 1600 Gb/s transfer speeds by 2025. These brand-new networking and adjoin user interfaces will likewise assist the HPC/ AI sector a lot in attaining the needed efficiency.